1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/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/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
125 if (C->log()) {
126 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
127 vmIntrinsics::name_at(intrinsic_id()),
128 (is_virtual() ? " virtual='1'" : ""),
129 C->unique() - nodes);
130 }
131 // Push the result from the inlined method onto the stack.
132 kit.push_result();
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 assert(jvms->map() == kit.map(), "Out of sync JVM state");
139 if (jvms->has_method()) {
140 // Not a root compile.
141 const char* msg;
142 if (callee->intrinsic_candidate()) {
143 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
144 } else {
145 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
146 : "failed to inline (intrinsic), method not annotated";
147 }
148 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
149 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
150 } else {
151 // Root compile
152 ResourceMark rm;
153 stringStream msg_stream;
154 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
155 vmIntrinsics::name_at(intrinsic_id()),
156 is_virtual() ? " (virtual)" : "", bci);
157 const char *msg = msg_stream.freeze();
158 log_debug(jit, inlining)("%s", msg);
159 if (C->print_intrinsics() || C->print_inlining()) {
160 tty->print("%s", msg);
161 }
162 }
163 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
164
165 return nullptr;
166 }
167
168 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
169 LibraryCallKit kit(jvms, this);
170 Compile* C = kit.C;
171 int nodes = C->unique();
172 _last_predicate = predicate;
173 #ifndef PRODUCT
174 assert(is_predicated() && predicate < predicates_count(), "sanity");
175 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
176 char buf[1000];
177 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
178 tty->print_cr("Predicate for intrinsic %s", str);
179 }
180 #endif
181 ciMethod* callee = kit.callee();
182 const int bci = kit.bci();
183
184 Node* slow_ctl = kit.try_to_predicate(predicate);
185 if (!kit.failing()) {
186 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
187 : "(intrinsic, predicate)";
188 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
189 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
190
191 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
192 if (C->log()) {
193 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
194 vmIntrinsics::name_at(intrinsic_id()),
195 (is_virtual() ? " virtual='1'" : ""),
196 C->unique() - nodes);
197 }
198 return slow_ctl; // Could be null if the check folds.
199 }
200
201 // The intrinsic bailed out
202 if (jvms->has_method()) {
203 // Not a root compile.
204 const char* msg = "failed to generate predicate for intrinsic";
205 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
206 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
207 } else {
208 // Root compile
209 ResourceMark rm;
210 stringStream msg_stream;
211 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
212 vmIntrinsics::name_at(intrinsic_id()),
213 is_virtual() ? " (virtual)" : "", bci);
214 const char *msg = msg_stream.freeze();
215 log_debug(jit, inlining)("%s", msg);
216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
217 }
218 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
219 return nullptr;
220 }
221
222 bool LibraryCallKit::try_to_inline(int predicate) {
223 // Handle symbolic names for otherwise undistinguished boolean switches:
224 const bool is_store = true;
225 const bool is_compress = true;
226 const bool is_static = true;
227 const bool is_volatile = true;
228
229 if (!jvms()->has_method()) {
230 // Root JVMState has a null method.
231 assert(map()->memory()->Opcode() == Op_Parm, "");
232 // Insert the memory aliasing node
233 set_all_memory(reset_memory());
234 }
235 assert(merged_memory(), "");
236
237 switch (intrinsic_id()) {
238 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
239 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
240 case vmIntrinsics::_getClass: return inline_native_getClass();
241
242 case vmIntrinsics::_ceil:
243 case vmIntrinsics::_floor:
244 case vmIntrinsics::_rint:
245 case vmIntrinsics::_dsin:
246 case vmIntrinsics::_dcos:
247 case vmIntrinsics::_dtan:
248 case vmIntrinsics::_dtanh:
249 case vmIntrinsics::_dcbrt:
250 case vmIntrinsics::_dabs:
251 case vmIntrinsics::_fabs:
252 case vmIntrinsics::_iabs:
253 case vmIntrinsics::_labs:
254 case vmIntrinsics::_datan2:
255 case vmIntrinsics::_dsqrt:
256 case vmIntrinsics::_dsqrt_strict:
257 case vmIntrinsics::_dexp:
258 case vmIntrinsics::_dlog:
259 case vmIntrinsics::_dlog10:
260 case vmIntrinsics::_dpow:
261 case vmIntrinsics::_dcopySign:
262 case vmIntrinsics::_fcopySign:
263 case vmIntrinsics::_dsignum:
264 case vmIntrinsics::_roundF:
265 case vmIntrinsics::_roundD:
266 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
267
268 case vmIntrinsics::_notify:
269 case vmIntrinsics::_notifyAll:
270 return inline_notify(intrinsic_id());
271
272 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
273 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
274 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
275 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
276 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
277 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
278 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
279 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
280 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
281 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
282 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
283 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
284 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
285 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
286
287 case vmIntrinsics::_arraycopy: return inline_arraycopy();
288
289 case vmIntrinsics::_arraySort: return inline_array_sort();
290 case vmIntrinsics::_arrayPartition: return inline_array_partition();
291
292 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
293 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
294 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
295 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
296
297 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
298 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
299 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
300 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
301 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
302 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
303 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
304 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
305
306 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
307
308 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
309
310 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
311 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
312 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
313 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
314
315 case vmIntrinsics::_compressStringC:
316 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
317 case vmIntrinsics::_inflateStringC:
318 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
319
320 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
321 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
322 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
323 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
324 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
325 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
326 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
327 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
328 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
329
330 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
331 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
332 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
333 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
334 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
335 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
336 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
337 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
338 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
339
340 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
341 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
342 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
343 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
344 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
345 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
346 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
347 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
348 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
349
350 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
351 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
352 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
353 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
354 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
355 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
356 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
357 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
358 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
359
360 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
361 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
362 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
363 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
364
365 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
366 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
367 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
368 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
369
370 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
371 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
372 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
373 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
374 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
375 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
376 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
377 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
378 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
379
380 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
381 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
382 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
383 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
384 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
385 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
386 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
387 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
388 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
389
390 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
391 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
392 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
393 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
394 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
395 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
396 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
397 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
398 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
399
400 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
401 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
402 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
403 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
404 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
405 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
406 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
407 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
408 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
409
410 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
415
416 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
417 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
418 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
419 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
420 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
421 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
422 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
423 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
424 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
436
437 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
438 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
439 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
440 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
441 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
442 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
443 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
444 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
445 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
446 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
452
453 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
457
458 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
463
464 case vmIntrinsics::_loadFence:
465 case vmIntrinsics::_storeFence:
466 case vmIntrinsics::_storeStoreFence:
467 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
468
469 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
470
471 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
472 case vmIntrinsics::_currentThread: return inline_native_currentThread();
473 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
474
475 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
476 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
477
478 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
479 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
480
481 #if INCLUDE_JVMTI
482 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
483 "notifyJvmtiStart", true, false);
484 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
485 "notifyJvmtiEnd", false, true);
486 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
487 "notifyJvmtiMount", false, false);
488 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
489 "notifyJvmtiUnmount", false, false);
490 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
491 #endif
492
493 #ifdef JFR_HAVE_INTRINSICS
494 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
495 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
496 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
497 #endif
498 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
499 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
500 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
501 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
502 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
503 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
504 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
505 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
506 case vmIntrinsics::_getLength: return inline_native_getLength();
507 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
508 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
509 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
510 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
511 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
512 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
513 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
514
515 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
516 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
517
518 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
519
520 case vmIntrinsics::_isInstance:
521 case vmIntrinsics::_isHidden:
522 case vmIntrinsics::_getSuperclass:
523 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
524
525 case vmIntrinsics::_floatToRawIntBits:
526 case vmIntrinsics::_floatToIntBits:
527 case vmIntrinsics::_intBitsToFloat:
528 case vmIntrinsics::_doubleToRawLongBits:
529 case vmIntrinsics::_doubleToLongBits:
530 case vmIntrinsics::_longBitsToDouble:
531 case vmIntrinsics::_floatToFloat16:
532 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
533 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
534 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
535 case vmIntrinsics::_floatIsFinite:
536 case vmIntrinsics::_floatIsInfinite:
537 case vmIntrinsics::_doubleIsFinite:
538 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
539
540 case vmIntrinsics::_numberOfLeadingZeros_i:
541 case vmIntrinsics::_numberOfLeadingZeros_l:
542 case vmIntrinsics::_numberOfTrailingZeros_i:
543 case vmIntrinsics::_numberOfTrailingZeros_l:
544 case vmIntrinsics::_bitCount_i:
545 case vmIntrinsics::_bitCount_l:
546 case vmIntrinsics::_reverse_i:
547 case vmIntrinsics::_reverse_l:
548 case vmIntrinsics::_reverseBytes_i:
549 case vmIntrinsics::_reverseBytes_l:
550 case vmIntrinsics::_reverseBytes_s:
551 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
552
553 case vmIntrinsics::_compress_i:
554 case vmIntrinsics::_compress_l:
555 case vmIntrinsics::_expand_i:
556 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
557
558 case vmIntrinsics::_compareUnsigned_i:
559 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
560
561 case vmIntrinsics::_divideUnsigned_i:
562 case vmIntrinsics::_divideUnsigned_l:
563 case vmIntrinsics::_remainderUnsigned_i:
564 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
565
566 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
567
568 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
569 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
570 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
571 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
572 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
573
574 case vmIntrinsics::_Class_cast: return inline_Class_cast();
575
576 case vmIntrinsics::_aescrypt_encryptBlock:
577 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
578
579 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
580 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
581 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
582
583 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
584 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
585 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
586
587 case vmIntrinsics::_counterMode_AESCrypt:
588 return inline_counterMode_AESCrypt(intrinsic_id());
589
590 case vmIntrinsics::_galoisCounterMode_AESCrypt:
591 return inline_galoisCounterMode_AESCrypt();
592
593 case vmIntrinsics::_md5_implCompress:
594 case vmIntrinsics::_sha_implCompress:
595 case vmIntrinsics::_sha2_implCompress:
596 case vmIntrinsics::_sha5_implCompress:
597 case vmIntrinsics::_sha3_implCompress:
598 return inline_digestBase_implCompress(intrinsic_id());
599 case vmIntrinsics::_double_keccak:
600 return inline_double_keccak();
601
602 case vmIntrinsics::_digestBase_implCompressMB:
603 return inline_digestBase_implCompressMB(predicate);
604
605 case vmIntrinsics::_multiplyToLen:
606 return inline_multiplyToLen();
607
608 case vmIntrinsics::_squareToLen:
609 return inline_squareToLen();
610
611 case vmIntrinsics::_mulAdd:
612 return inline_mulAdd();
613
614 case vmIntrinsics::_montgomeryMultiply:
615 return inline_montgomeryMultiply();
616 case vmIntrinsics::_montgomerySquare:
617 return inline_montgomerySquare();
618
619 case vmIntrinsics::_bigIntegerRightShiftWorker:
620 return inline_bigIntegerShift(true);
621 case vmIntrinsics::_bigIntegerLeftShiftWorker:
622 return inline_bigIntegerShift(false);
623
624 case vmIntrinsics::_vectorizedMismatch:
625 return inline_vectorizedMismatch();
626
627 case vmIntrinsics::_ghash_processBlocks:
628 return inline_ghash_processBlocks();
629 case vmIntrinsics::_chacha20Block:
630 return inline_chacha20Block();
631 case vmIntrinsics::_kyberNtt:
632 return inline_kyberNtt();
633 case vmIntrinsics::_kyberInverseNtt:
634 return inline_kyberInverseNtt();
635 case vmIntrinsics::_kyberNttMult:
636 return inline_kyberNttMult();
637 case vmIntrinsics::_kyberAddPoly_2:
638 return inline_kyberAddPoly_2();
639 case vmIntrinsics::_kyberAddPoly_3:
640 return inline_kyberAddPoly_3();
641 case vmIntrinsics::_kyber12To16:
642 return inline_kyber12To16();
643 case vmIntrinsics::_kyberBarrettReduce:
644 return inline_kyberBarrettReduce();
645 case vmIntrinsics::_dilithiumAlmostNtt:
646 return inline_dilithiumAlmostNtt();
647 case vmIntrinsics::_dilithiumAlmostInverseNtt:
648 return inline_dilithiumAlmostInverseNtt();
649 case vmIntrinsics::_dilithiumNttMult:
650 return inline_dilithiumNttMult();
651 case vmIntrinsics::_dilithiumMontMulByConstant:
652 return inline_dilithiumMontMulByConstant();
653 case vmIntrinsics::_dilithiumDecomposePoly:
654 return inline_dilithiumDecomposePoly();
655 case vmIntrinsics::_base64_encodeBlock:
656 return inline_base64_encodeBlock();
657 case vmIntrinsics::_base64_decodeBlock:
658 return inline_base64_decodeBlock();
659 case vmIntrinsics::_poly1305_processBlocks:
660 return inline_poly1305_processBlocks();
661 case vmIntrinsics::_intpoly_montgomeryMult_P256:
662 return inline_intpoly_montgomeryMult_P256();
663 case vmIntrinsics::_intpoly_assign:
664 return inline_intpoly_assign();
665 case vmIntrinsics::_encodeISOArray:
666 case vmIntrinsics::_encodeByteISOArray:
667 return inline_encodeISOArray(false);
668 case vmIntrinsics::_encodeAsciiArray:
669 return inline_encodeISOArray(true);
670
671 case vmIntrinsics::_updateCRC32:
672 return inline_updateCRC32();
673 case vmIntrinsics::_updateBytesCRC32:
674 return inline_updateBytesCRC32();
675 case vmIntrinsics::_updateByteBufferCRC32:
676 return inline_updateByteBufferCRC32();
677
678 case vmIntrinsics::_updateBytesCRC32C:
679 return inline_updateBytesCRC32C();
680 case vmIntrinsics::_updateDirectByteBufferCRC32C:
681 return inline_updateDirectByteBufferCRC32C();
682
683 case vmIntrinsics::_updateBytesAdler32:
684 return inline_updateBytesAdler32();
685 case vmIntrinsics::_updateByteBufferAdler32:
686 return inline_updateByteBufferAdler32();
687
688 case vmIntrinsics::_profileBoolean:
689 return inline_profileBoolean();
690 case vmIntrinsics::_isCompileConstant:
691 return inline_isCompileConstant();
692
693 case vmIntrinsics::_countPositives:
694 return inline_countPositives();
695
696 case vmIntrinsics::_fmaD:
697 case vmIntrinsics::_fmaF:
698 return inline_fma(intrinsic_id());
699
700 case vmIntrinsics::_isDigit:
701 case vmIntrinsics::_isLowerCase:
702 case vmIntrinsics::_isUpperCase:
703 case vmIntrinsics::_isWhitespace:
704 return inline_character_compare(intrinsic_id());
705
706 case vmIntrinsics::_min:
707 case vmIntrinsics::_max:
708 case vmIntrinsics::_min_strict:
709 case vmIntrinsics::_max_strict:
710 case vmIntrinsics::_minL:
711 case vmIntrinsics::_maxL:
712 case vmIntrinsics::_minF:
713 case vmIntrinsics::_maxF:
714 case vmIntrinsics::_minD:
715 case vmIntrinsics::_maxD:
716 case vmIntrinsics::_minF_strict:
717 case vmIntrinsics::_maxF_strict:
718 case vmIntrinsics::_minD_strict:
719 case vmIntrinsics::_maxD_strict:
720 return inline_min_max(intrinsic_id());
721
722 case vmIntrinsics::_VectorUnaryOp:
723 return inline_vector_nary_operation(1);
724 case vmIntrinsics::_VectorBinaryOp:
725 return inline_vector_nary_operation(2);
726 case vmIntrinsics::_VectorUnaryLibOp:
727 return inline_vector_call(1);
728 case vmIntrinsics::_VectorBinaryLibOp:
729 return inline_vector_call(2);
730 case vmIntrinsics::_VectorTernaryOp:
731 return inline_vector_nary_operation(3);
732 case vmIntrinsics::_VectorFromBitsCoerced:
733 return inline_vector_frombits_coerced();
734 case vmIntrinsics::_VectorMaskOp:
735 return inline_vector_mask_operation();
736 case vmIntrinsics::_VectorLoadOp:
737 return inline_vector_mem_operation(/*is_store=*/false);
738 case vmIntrinsics::_VectorLoadMaskedOp:
739 return inline_vector_mem_masked_operation(/*is_store*/false);
740 case vmIntrinsics::_VectorStoreOp:
741 return inline_vector_mem_operation(/*is_store=*/true);
742 case vmIntrinsics::_VectorStoreMaskedOp:
743 return inline_vector_mem_masked_operation(/*is_store=*/true);
744 case vmIntrinsics::_VectorGatherOp:
745 return inline_vector_gather_scatter(/*is_scatter*/ false);
746 case vmIntrinsics::_VectorScatterOp:
747 return inline_vector_gather_scatter(/*is_scatter*/ true);
748 case vmIntrinsics::_VectorReductionCoerced:
749 return inline_vector_reduction();
750 case vmIntrinsics::_VectorTest:
751 return inline_vector_test();
752 case vmIntrinsics::_VectorBlend:
753 return inline_vector_blend();
754 case vmIntrinsics::_VectorRearrange:
755 return inline_vector_rearrange();
756 case vmIntrinsics::_VectorSelectFrom:
757 return inline_vector_select_from();
758 case vmIntrinsics::_VectorCompare:
759 return inline_vector_compare();
760 case vmIntrinsics::_VectorBroadcastInt:
761 return inline_vector_broadcast_int();
762 case vmIntrinsics::_VectorConvert:
763 return inline_vector_convert();
764 case vmIntrinsics::_VectorInsert:
765 return inline_vector_insert();
766 case vmIntrinsics::_VectorExtract:
767 return inline_vector_extract();
768 case vmIntrinsics::_VectorCompressExpand:
769 return inline_vector_compress_expand();
770 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
771 return inline_vector_select_from_two_vectors();
772 case vmIntrinsics::_IndexVector:
773 return inline_index_vector();
774 case vmIntrinsics::_IndexPartiallyInUpperRange:
775 return inline_index_partially_in_upper_range();
776
777 case vmIntrinsics::_getObjectSize:
778 return inline_getObjectSize();
779
780 case vmIntrinsics::_blackhole:
781 return inline_blackhole();
782
783 default:
784 // If you get here, it may be that someone has added a new intrinsic
785 // to the list in vmIntrinsics.hpp without implementing it here.
786 #ifndef PRODUCT
787 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
788 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
789 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
790 }
791 #endif
792 return false;
793 }
794 }
795
796 Node* LibraryCallKit::try_to_predicate(int predicate) {
797 if (!jvms()->has_method()) {
798 // Root JVMState has a null method.
799 assert(map()->memory()->Opcode() == Op_Parm, "");
800 // Insert the memory aliasing node
801 set_all_memory(reset_memory());
802 }
803 assert(merged_memory(), "");
804
805 switch (intrinsic_id()) {
806 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
807 return inline_cipherBlockChaining_AESCrypt_predicate(false);
808 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
809 return inline_cipherBlockChaining_AESCrypt_predicate(true);
810 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
811 return inline_electronicCodeBook_AESCrypt_predicate(false);
812 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
813 return inline_electronicCodeBook_AESCrypt_predicate(true);
814 case vmIntrinsics::_counterMode_AESCrypt:
815 return inline_counterMode_AESCrypt_predicate();
816 case vmIntrinsics::_digestBase_implCompressMB:
817 return inline_digestBase_implCompressMB_predicate(predicate);
818 case vmIntrinsics::_galoisCounterMode_AESCrypt:
819 return inline_galoisCounterMode_AESCrypt_predicate();
820
821 default:
822 // If you get here, it may be that someone has added a new intrinsic
823 // to the list in vmIntrinsics.hpp without implementing it here.
824 #ifndef PRODUCT
825 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
826 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
827 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
828 }
829 #endif
830 Node* slow_ctl = control();
831 set_control(top()); // No fast path intrinsic
832 return slow_ctl;
833 }
834 }
835
836 //------------------------------set_result-------------------------------
837 // Helper function for finishing intrinsics.
838 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
839 record_for_igvn(region);
840 set_control(_gvn.transform(region));
841 set_result( _gvn.transform(value));
842 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
843 }
844
845 //------------------------------generate_guard---------------------------
846 // Helper function for generating guarded fast-slow graph structures.
847 // The given 'test', if true, guards a slow path. If the test fails
848 // then a fast path can be taken. (We generally hope it fails.)
849 // In all cases, GraphKit::control() is updated to the fast path.
850 // The returned value represents the control for the slow path.
851 // The return value is never 'top'; it is either a valid control
852 // or null if it is obvious that the slow path can never be taken.
853 // Also, if region and the slow control are not null, the slow edge
854 // is appended to the region.
855 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
856 if (stopped()) {
857 // Already short circuited.
858 return nullptr;
859 }
860
861 // Build an if node and its projections.
862 // If test is true we take the slow path, which we assume is uncommon.
863 if (_gvn.type(test) == TypeInt::ZERO) {
864 // The slow branch is never taken. No need to build this guard.
865 return nullptr;
866 }
867
868 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
869
870 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
871 if (if_slow == top()) {
872 // The slow branch is never taken. No need to build this guard.
873 return nullptr;
874 }
875
876 if (region != nullptr)
877 region->add_req(if_slow);
878
879 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
880 set_control(if_fast);
881
882 return if_slow;
883 }
884
885 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
886 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
887 }
888 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
889 return generate_guard(test, region, PROB_FAIR);
890 }
891
892 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
893 Node* *pos_index) {
894 if (stopped())
895 return nullptr; // already stopped
896 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
897 return nullptr; // index is already adequately typed
898 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
899 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
900 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
901 if (is_neg != nullptr && pos_index != nullptr) {
902 // Emulate effect of Parse::adjust_map_after_if.
903 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
904 (*pos_index) = _gvn.transform(ccast);
905 }
906 return is_neg;
907 }
908
909 // Make sure that 'position' is a valid limit index, in [0..length].
910 // There are two equivalent plans for checking this:
911 // A. (offset + copyLength) unsigned<= arrayLength
912 // B. offset <= (arrayLength - copyLength)
913 // We require that all of the values above, except for the sum and
914 // difference, are already known to be non-negative.
915 // Plan A is robust in the face of overflow, if offset and copyLength
916 // are both hugely positive.
917 //
918 // Plan B is less direct and intuitive, but it does not overflow at
919 // all, since the difference of two non-negatives is always
920 // representable. Whenever Java methods must perform the equivalent
921 // check they generally use Plan B instead of Plan A.
922 // For the moment we use Plan A.
923 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
924 Node* subseq_length,
925 Node* array_length,
926 RegionNode* region) {
927 if (stopped())
928 return nullptr; // already stopped
929 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
930 if (zero_offset && subseq_length->eqv_uncast(array_length))
931 return nullptr; // common case of whole-array copy
932 Node* last = subseq_length;
933 if (!zero_offset) // last += offset
934 last = _gvn.transform(new AddINode(last, offset));
935 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
936 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
937 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
938 return is_over;
939 }
940
941 // Emit range checks for the given String.value byte array
942 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
943 if (stopped()) {
944 return; // already stopped
945 }
946 RegionNode* bailout = new RegionNode(1);
947 record_for_igvn(bailout);
948 if (char_count) {
949 // Convert char count to byte count
950 count = _gvn.transform(new LShiftINode(count, intcon(1)));
951 }
952
953 // Offset and count must not be negative
954 generate_negative_guard(offset, bailout);
955 generate_negative_guard(count, bailout);
956 // Offset + count must not exceed length of array
957 generate_limit_guard(offset, count, load_array_length(array), bailout);
958
959 if (bailout->req() > 1) {
960 PreserveJVMState pjvms(this);
961 set_control(_gvn.transform(bailout));
962 uncommon_trap(Deoptimization::Reason_intrinsic,
963 Deoptimization::Action_maybe_recompile);
964 }
965 }
966
967 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
968 bool is_immutable) {
969 ciKlass* thread_klass = env()->Thread_klass();
970 const Type* thread_type
971 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
972
973 Node* thread = _gvn.transform(new ThreadLocalNode());
974 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
975 tls_output = thread;
976
977 Node* thread_obj_handle
978 = (is_immutable
979 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
980 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
981 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
982 thread_obj_handle = _gvn.transform(thread_obj_handle);
983
984 DecoratorSet decorators = IN_NATIVE;
985 if (is_immutable) {
986 decorators |= C2_IMMUTABLE_MEMORY;
987 }
988 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
989 }
990
991 //--------------------------generate_current_thread--------------------
992 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
993 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
994 /*is_immutable*/false);
995 }
996
997 //--------------------------generate_virtual_thread--------------------
998 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
999 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1000 !C->method()->changes_current_thread());
1001 }
1002
1003 //------------------------------make_string_method_node------------------------
1004 // Helper method for String intrinsic functions. This version is called with
1005 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1006 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1007 // containing the lengths of str1 and str2.
1008 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1009 Node* result = nullptr;
1010 switch (opcode) {
1011 case Op_StrIndexOf:
1012 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1013 str1_start, cnt1, str2_start, cnt2, ae);
1014 break;
1015 case Op_StrComp:
1016 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1017 str1_start, cnt1, str2_start, cnt2, ae);
1018 break;
1019 case Op_StrEquals:
1020 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1021 // Use the constant length if there is one because optimized match rule may exist.
1022 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1023 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1024 break;
1025 default:
1026 ShouldNotReachHere();
1027 return nullptr;
1028 }
1029
1030 // All these intrinsics have checks.
1031 C->set_has_split_ifs(true); // Has chance for split-if optimization
1032 clear_upper_avx();
1033
1034 return _gvn.transform(result);
1035 }
1036
1037 //------------------------------inline_string_compareTo------------------------
1038 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1039 Node* arg1 = argument(0);
1040 Node* arg2 = argument(1);
1041
1042 arg1 = must_be_not_null(arg1, true);
1043 arg2 = must_be_not_null(arg2, true);
1044
1045 // Get start addr and length of first argument
1046 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1047 Node* arg1_cnt = load_array_length(arg1);
1048
1049 // Get start addr and length of second argument
1050 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1051 Node* arg2_cnt = load_array_length(arg2);
1052
1053 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1054 set_result(result);
1055 return true;
1056 }
1057
1058 //------------------------------inline_string_equals------------------------
1059 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1060 Node* arg1 = argument(0);
1061 Node* arg2 = argument(1);
1062
1063 // paths (plus control) merge
1064 RegionNode* region = new RegionNode(3);
1065 Node* phi = new PhiNode(region, TypeInt::BOOL);
1066
1067 if (!stopped()) {
1068
1069 arg1 = must_be_not_null(arg1, true);
1070 arg2 = must_be_not_null(arg2, true);
1071
1072 // Get start addr and length of first argument
1073 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1074 Node* arg1_cnt = load_array_length(arg1);
1075
1076 // Get start addr and length of second argument
1077 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1078 Node* arg2_cnt = load_array_length(arg2);
1079
1080 // Check for arg1_cnt != arg2_cnt
1081 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1082 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1083 Node* if_ne = generate_slow_guard(bol, nullptr);
1084 if (if_ne != nullptr) {
1085 phi->init_req(2, intcon(0));
1086 region->init_req(2, if_ne);
1087 }
1088
1089 // Check for count == 0 is done by assembler code for StrEquals.
1090
1091 if (!stopped()) {
1092 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1093 phi->init_req(1, equals);
1094 region->init_req(1, control());
1095 }
1096 }
1097
1098 // post merge
1099 set_control(_gvn.transform(region));
1100 record_for_igvn(region);
1101
1102 set_result(_gvn.transform(phi));
1103 return true;
1104 }
1105
1106 //------------------------------inline_array_equals----------------------------
1107 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1108 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1109 Node* arg1 = argument(0);
1110 Node* arg2 = argument(1);
1111
1112 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1113 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1114 clear_upper_avx();
1115
1116 return true;
1117 }
1118
1119
1120 //------------------------------inline_countPositives------------------------------
1121 bool LibraryCallKit::inline_countPositives() {
1122 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1123 return false;
1124 }
1125
1126 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1127 // no receiver since it is static method
1128 Node* ba = argument(0);
1129 Node* offset = argument(1);
1130 Node* len = argument(2);
1131
1132 ba = must_be_not_null(ba, true);
1133
1134 // Range checks
1135 generate_string_range_check(ba, offset, len, false);
1136 if (stopped()) {
1137 return true;
1138 }
1139 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1140 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1141 set_result(_gvn.transform(result));
1142 clear_upper_avx();
1143 return true;
1144 }
1145
1146 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1147 Node* index = argument(0);
1148 Node* length = bt == T_INT ? argument(1) : argument(2);
1149 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1150 return false;
1151 }
1152
1153 // check that length is positive
1154 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1155 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1156
1157 {
1158 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1159 uncommon_trap(Deoptimization::Reason_intrinsic,
1160 Deoptimization::Action_make_not_entrant);
1161 }
1162
1163 if (stopped()) {
1164 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1165 return true;
1166 }
1167
1168 // length is now known positive, add a cast node to make this explicit
1169 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1170 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1171 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1172 ConstraintCastNode::RegularDependency, bt);
1173 casted_length = _gvn.transform(casted_length);
1174 replace_in_map(length, casted_length);
1175 length = casted_length;
1176
1177 // Use an unsigned comparison for the range check itself
1178 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1179 BoolTest::mask btest = BoolTest::lt;
1180 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1181 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1182 _gvn.set_type(rc, rc->Value(&_gvn));
1183 if (!rc_bool->is_Con()) {
1184 record_for_igvn(rc);
1185 }
1186 set_control(_gvn.transform(new IfTrueNode(rc)));
1187 {
1188 PreserveJVMState pjvms(this);
1189 set_control(_gvn.transform(new IfFalseNode(rc)));
1190 uncommon_trap(Deoptimization::Reason_range_check,
1191 Deoptimization::Action_make_not_entrant);
1192 }
1193
1194 if (stopped()) {
1195 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1196 return true;
1197 }
1198
1199 // index is now known to be >= 0 and < length, cast it
1200 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1201 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1202 ConstraintCastNode::RegularDependency, bt);
1203 result = _gvn.transform(result);
1204 set_result(result);
1205 replace_in_map(index, result);
1206 return true;
1207 }
1208
1209 //------------------------------inline_string_indexOf------------------------
1210 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1211 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1212 return false;
1213 }
1214 Node* src = argument(0);
1215 Node* tgt = argument(1);
1216
1217 // Make the merge point
1218 RegionNode* result_rgn = new RegionNode(4);
1219 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1220
1221 src = must_be_not_null(src, true);
1222 tgt = must_be_not_null(tgt, true);
1223
1224 // Get start addr and length of source string
1225 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1226 Node* src_count = load_array_length(src);
1227
1228 // Get start addr and length of substring
1229 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1230 Node* tgt_count = load_array_length(tgt);
1231
1232 Node* result = nullptr;
1233 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1234
1235 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1236 // Divide src size by 2 if String is UTF16 encoded
1237 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1238 }
1239 if (ae == StrIntrinsicNode::UU) {
1240 // Divide substring size by 2 if String is UTF16 encoded
1241 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1242 }
1243
1244 if (call_opt_stub) {
1245 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1246 StubRoutines::_string_indexof_array[ae],
1247 "stringIndexOf", TypePtr::BOTTOM, src_start,
1248 src_count, tgt_start, tgt_count);
1249 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1250 } else {
1251 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1252 result_rgn, result_phi, ae);
1253 }
1254 if (result != nullptr) {
1255 result_phi->init_req(3, result);
1256 result_rgn->init_req(3, control());
1257 }
1258 set_control(_gvn.transform(result_rgn));
1259 record_for_igvn(result_rgn);
1260 set_result(_gvn.transform(result_phi));
1261
1262 return true;
1263 }
1264
1265 //-----------------------------inline_string_indexOfI-----------------------
1266 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1267 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1268 return false;
1269 }
1270 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1271 return false;
1272 }
1273
1274 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1275 Node* src = argument(0); // byte[]
1276 Node* src_count = argument(1); // char count
1277 Node* tgt = argument(2); // byte[]
1278 Node* tgt_count = argument(3); // char count
1279 Node* from_index = argument(4); // char index
1280
1281 src = must_be_not_null(src, true);
1282 tgt = must_be_not_null(tgt, true);
1283
1284 // Multiply byte array index by 2 if String is UTF16 encoded
1285 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1286 src_count = _gvn.transform(new SubINode(src_count, from_index));
1287 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1288 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1289
1290 // Range checks
1291 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1292 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1293 if (stopped()) {
1294 return true;
1295 }
1296
1297 RegionNode* region = new RegionNode(5);
1298 Node* phi = new PhiNode(region, TypeInt::INT);
1299 Node* result = nullptr;
1300
1301 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1302
1303 if (call_opt_stub) {
1304 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1305 StubRoutines::_string_indexof_array[ae],
1306 "stringIndexOf", TypePtr::BOTTOM, src_start,
1307 src_count, tgt_start, tgt_count);
1308 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1309 } else {
1310 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1311 region, phi, ae);
1312 }
1313 if (result != nullptr) {
1314 // The result is index relative to from_index if substring was found, -1 otherwise.
1315 // Generate code which will fold into cmove.
1316 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1317 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1318
1319 Node* if_lt = generate_slow_guard(bol, nullptr);
1320 if (if_lt != nullptr) {
1321 // result == -1
1322 phi->init_req(3, result);
1323 region->init_req(3, if_lt);
1324 }
1325 if (!stopped()) {
1326 result = _gvn.transform(new AddINode(result, from_index));
1327 phi->init_req(4, result);
1328 region->init_req(4, control());
1329 }
1330 }
1331
1332 set_control(_gvn.transform(region));
1333 record_for_igvn(region);
1334 set_result(_gvn.transform(phi));
1335 clear_upper_avx();
1336
1337 return true;
1338 }
1339
1340 // Create StrIndexOfNode with fast path checks
1341 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1342 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1343 // Check for substr count > string count
1344 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1345 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1346 Node* if_gt = generate_slow_guard(bol, nullptr);
1347 if (if_gt != nullptr) {
1348 phi->init_req(1, intcon(-1));
1349 region->init_req(1, if_gt);
1350 }
1351 if (!stopped()) {
1352 // Check for substr count == 0
1353 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1354 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1355 Node* if_zero = generate_slow_guard(bol, nullptr);
1356 if (if_zero != nullptr) {
1357 phi->init_req(2, intcon(0));
1358 region->init_req(2, if_zero);
1359 }
1360 }
1361 if (!stopped()) {
1362 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1363 }
1364 return nullptr;
1365 }
1366
1367 //-----------------------------inline_string_indexOfChar-----------------------
1368 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1369 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1370 return false;
1371 }
1372 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1373 return false;
1374 }
1375 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1376 Node* src = argument(0); // byte[]
1377 Node* int_ch = argument(1);
1378 Node* from_index = argument(2);
1379 Node* max = argument(3);
1380
1381 src = must_be_not_null(src, true);
1382
1383 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1384 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1385 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1386
1387 // Range checks
1388 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1389
1390 // Check for int_ch >= 0
1391 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1392 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1393 {
1394 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1395 uncommon_trap(Deoptimization::Reason_intrinsic,
1396 Deoptimization::Action_maybe_recompile);
1397 }
1398 if (stopped()) {
1399 return true;
1400 }
1401
1402 RegionNode* region = new RegionNode(3);
1403 Node* phi = new PhiNode(region, TypeInt::INT);
1404
1405 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1406 C->set_has_split_ifs(true); // Has chance for split-if optimization
1407 _gvn.transform(result);
1408
1409 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1410 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1411
1412 Node* if_lt = generate_slow_guard(bol, nullptr);
1413 if (if_lt != nullptr) {
1414 // result == -1
1415 phi->init_req(2, result);
1416 region->init_req(2, if_lt);
1417 }
1418 if (!stopped()) {
1419 result = _gvn.transform(new AddINode(result, from_index));
1420 phi->init_req(1, result);
1421 region->init_req(1, control());
1422 }
1423 set_control(_gvn.transform(region));
1424 record_for_igvn(region);
1425 set_result(_gvn.transform(phi));
1426 clear_upper_avx();
1427
1428 return true;
1429 }
1430 //---------------------------inline_string_copy---------------------
1431 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1432 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1433 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1434 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1435 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1436 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1437 bool LibraryCallKit::inline_string_copy(bool compress) {
1438 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1439 return false;
1440 }
1441 int nargs = 5; // 2 oops, 3 ints
1442 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1443
1444 Node* src = argument(0);
1445 Node* src_offset = argument(1);
1446 Node* dst = argument(2);
1447 Node* dst_offset = argument(3);
1448 Node* length = argument(4);
1449
1450 // Check for allocation before we add nodes that would confuse
1451 // tightly_coupled_allocation()
1452 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1453
1454 // Figure out the size and type of the elements we will be copying.
1455 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1456 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1457 if (src_type == nullptr || dst_type == nullptr) {
1458 return false;
1459 }
1460 BasicType src_elem = src_type->elem()->array_element_basic_type();
1461 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1462 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1463 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1464 "Unsupported array types for inline_string_copy");
1465
1466 src = must_be_not_null(src, true);
1467 dst = must_be_not_null(dst, true);
1468
1469 // Convert char[] offsets to byte[] offsets
1470 bool convert_src = (compress && src_elem == T_BYTE);
1471 bool convert_dst = (!compress && dst_elem == T_BYTE);
1472 if (convert_src) {
1473 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1474 } else if (convert_dst) {
1475 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1476 }
1477
1478 // Range checks
1479 generate_string_range_check(src, src_offset, length, convert_src);
1480 generate_string_range_check(dst, dst_offset, length, convert_dst);
1481 if (stopped()) {
1482 return true;
1483 }
1484
1485 Node* src_start = array_element_address(src, src_offset, src_elem);
1486 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1487 // 'src_start' points to src array + scaled offset
1488 // 'dst_start' points to dst array + scaled offset
1489 Node* count = nullptr;
1490 if (compress) {
1491 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1492 } else {
1493 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1494 }
1495
1496 if (alloc != nullptr) {
1497 if (alloc->maybe_set_complete(&_gvn)) {
1498 // "You break it, you buy it."
1499 InitializeNode* init = alloc->initialization();
1500 assert(init->is_complete(), "we just did this");
1501 init->set_complete_with_arraycopy();
1502 assert(dst->is_CheckCastPP(), "sanity");
1503 assert(dst->in(0)->in(0) == init, "dest pinned");
1504 }
1505 // Do not let stores that initialize this object be reordered with
1506 // a subsequent store that would make this object accessible by
1507 // other threads.
1508 // Record what AllocateNode this StoreStore protects so that
1509 // escape analysis can go from the MemBarStoreStoreNode to the
1510 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1511 // based on the escape status of the AllocateNode.
1512 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1513 }
1514 if (compress) {
1515 set_result(_gvn.transform(count));
1516 }
1517 clear_upper_avx();
1518
1519 return true;
1520 }
1521
1522 #ifdef _LP64
1523 #define XTOP ,top() /*additional argument*/
1524 #else //_LP64
1525 #define XTOP /*no additional argument*/
1526 #endif //_LP64
1527
1528 //------------------------inline_string_toBytesU--------------------------
1529 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1530 bool LibraryCallKit::inline_string_toBytesU() {
1531 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1532 return false;
1533 }
1534 // Get the arguments.
1535 Node* value = argument(0);
1536 Node* offset = argument(1);
1537 Node* length = argument(2);
1538
1539 Node* newcopy = nullptr;
1540
1541 // Set the original stack and the reexecute bit for the interpreter to reexecute
1542 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1543 { PreserveReexecuteState preexecs(this);
1544 jvms()->set_should_reexecute(true);
1545
1546 // Check if a null path was taken unconditionally.
1547 value = null_check(value);
1548
1549 RegionNode* bailout = new RegionNode(1);
1550 record_for_igvn(bailout);
1551
1552 // Range checks
1553 generate_negative_guard(offset, bailout);
1554 generate_negative_guard(length, bailout);
1555 generate_limit_guard(offset, length, load_array_length(value), bailout);
1556 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1557 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1558
1559 if (bailout->req() > 1) {
1560 PreserveJVMState pjvms(this);
1561 set_control(_gvn.transform(bailout));
1562 uncommon_trap(Deoptimization::Reason_intrinsic,
1563 Deoptimization::Action_maybe_recompile);
1564 }
1565 if (stopped()) {
1566 return true;
1567 }
1568
1569 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1570 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1571 newcopy = new_array(klass_node, size, 0); // no arguments to push
1572 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1573 guarantee(alloc != nullptr, "created above");
1574
1575 // Calculate starting addresses.
1576 Node* src_start = array_element_address(value, offset, T_CHAR);
1577 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1578
1579 // Check if dst array address is aligned to HeapWordSize
1580 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1581 // If true, then check if src array address is aligned to HeapWordSize
1582 if (aligned) {
1583 const TypeInt* toffset = gvn().type(offset)->is_int();
1584 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1585 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1586 }
1587
1588 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1589 const char* copyfunc_name = "arraycopy";
1590 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1591 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1592 OptoRuntime::fast_arraycopy_Type(),
1593 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1594 src_start, dst_start, ConvI2X(length) XTOP);
1595 // Do not let reads from the cloned object float above the arraycopy.
1596 if (alloc->maybe_set_complete(&_gvn)) {
1597 // "You break it, you buy it."
1598 InitializeNode* init = alloc->initialization();
1599 assert(init->is_complete(), "we just did this");
1600 init->set_complete_with_arraycopy();
1601 assert(newcopy->is_CheckCastPP(), "sanity");
1602 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1603 }
1604 // Do not let stores that initialize this object be reordered with
1605 // a subsequent store that would make this object accessible by
1606 // other threads.
1607 // Record what AllocateNode this StoreStore protects so that
1608 // escape analysis can go from the MemBarStoreStoreNode to the
1609 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1610 // based on the escape status of the AllocateNode.
1611 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1612 } // original reexecute is set back here
1613
1614 C->set_has_split_ifs(true); // Has chance for split-if optimization
1615 if (!stopped()) {
1616 set_result(newcopy);
1617 }
1618 clear_upper_avx();
1619
1620 return true;
1621 }
1622
1623 //------------------------inline_string_getCharsU--------------------------
1624 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1625 bool LibraryCallKit::inline_string_getCharsU() {
1626 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1627 return false;
1628 }
1629
1630 // Get the arguments.
1631 Node* src = argument(0);
1632 Node* src_begin = argument(1);
1633 Node* src_end = argument(2); // exclusive offset (i < src_end)
1634 Node* dst = argument(3);
1635 Node* dst_begin = argument(4);
1636
1637 // Check for allocation before we add nodes that would confuse
1638 // tightly_coupled_allocation()
1639 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1640
1641 // Check if a null path was taken unconditionally.
1642 src = null_check(src);
1643 dst = null_check(dst);
1644 if (stopped()) {
1645 return true;
1646 }
1647
1648 // Get length and convert char[] offset to byte[] offset
1649 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1650 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1651
1652 // Range checks
1653 generate_string_range_check(src, src_begin, length, true);
1654 generate_string_range_check(dst, dst_begin, length, false);
1655 if (stopped()) {
1656 return true;
1657 }
1658
1659 if (!stopped()) {
1660 // Calculate starting addresses.
1661 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1662 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1663
1664 // Check if array addresses are aligned to HeapWordSize
1665 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1666 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1667 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1668 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1669
1670 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1671 const char* copyfunc_name = "arraycopy";
1672 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1673 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1674 OptoRuntime::fast_arraycopy_Type(),
1675 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1676 src_start, dst_start, ConvI2X(length) XTOP);
1677 // Do not let reads from the cloned object float above the arraycopy.
1678 if (alloc != nullptr) {
1679 if (alloc->maybe_set_complete(&_gvn)) {
1680 // "You break it, you buy it."
1681 InitializeNode* init = alloc->initialization();
1682 assert(init->is_complete(), "we just did this");
1683 init->set_complete_with_arraycopy();
1684 assert(dst->is_CheckCastPP(), "sanity");
1685 assert(dst->in(0)->in(0) == init, "dest pinned");
1686 }
1687 // Do not let stores that initialize this object be reordered with
1688 // a subsequent store that would make this object accessible by
1689 // other threads.
1690 // Record what AllocateNode this StoreStore protects so that
1691 // escape analysis can go from the MemBarStoreStoreNode to the
1692 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1693 // based on the escape status of the AllocateNode.
1694 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1695 } else {
1696 insert_mem_bar(Op_MemBarCPUOrder);
1697 }
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* value = argument(0);
1710 Node* index = argument(1);
1711 Node* ch = is_store ? argument(2) : nullptr;
1712
1713 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1714 // correctly requires matched array shapes.
1715 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1716 "sanity: byte[] and char[] bases agree");
1717 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1718 "sanity: byte[] and char[] scales agree");
1719
1720 // Bail when getChar over constants is requested: constant folding would
1721 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1722 // Java method would constant fold nicely instead.
1723 if (!is_store && value->is_Con() && index->is_Con()) {
1724 return false;
1725 }
1726
1727 // Save state and restore on bailout
1728 SavedState old_state(this);
1729
1730 value = must_be_not_null(value, true);
1731
1732 Node* adr = array_element_address(value, index, T_CHAR);
1733 if (adr->is_top()) {
1734 return false;
1735 }
1736 old_state.discard();
1737 if (is_store) {
1738 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1739 } else {
1740 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);
1741 set_result(ch);
1742 }
1743 return true;
1744 }
1745
1746
1747 //------------------------------inline_math-----------------------------------
1748 // public static double Math.abs(double)
1749 // public static double Math.sqrt(double)
1750 // public static double Math.log(double)
1751 // public static double Math.log10(double)
1752 // public static double Math.round(double)
1753 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1754 Node* arg = argument(0);
1755 Node* n = nullptr;
1756 switch (id) {
1757 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1758 case vmIntrinsics::_dsqrt:
1759 case vmIntrinsics::_dsqrt_strict:
1760 n = new SqrtDNode(C, control(), arg); break;
1761 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1762 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1763 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1764 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1765 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1766 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1767 default: fatal_unexpected_iid(id); break;
1768 }
1769 set_result(_gvn.transform(n));
1770 return true;
1771 }
1772
1773 //------------------------------inline_math-----------------------------------
1774 // public static float Math.abs(float)
1775 // public static int Math.abs(int)
1776 // public static long Math.abs(long)
1777 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1778 Node* arg = argument(0);
1779 Node* n = nullptr;
1780 switch (id) {
1781 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1782 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1783 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1784 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1785 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1786 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1787 default: fatal_unexpected_iid(id); break;
1788 }
1789 set_result(_gvn.transform(n));
1790 return true;
1791 }
1792
1793 //------------------------------runtime_math-----------------------------
1794 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1795 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1796 "must be (DD)D or (D)D type");
1797
1798 // Inputs
1799 Node* a = argument(0);
1800 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1801
1802 const TypePtr* no_memory_effects = nullptr;
1803 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1804 no_memory_effects,
1805 a, top(), b, b ? top() : nullptr);
1806 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1807 #ifdef ASSERT
1808 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1809 assert(value_top == top(), "second value must be top");
1810 #endif
1811
1812 set_result(value);
1813 return true;
1814 }
1815
1816 //------------------------------inline_math_pow-----------------------------
1817 bool LibraryCallKit::inline_math_pow() {
1818 Node* exp = argument(2);
1819 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1820 if (d != nullptr) {
1821 if (d->getd() == 2.0) {
1822 // Special case: pow(x, 2.0) => x * x
1823 Node* base = argument(0);
1824 set_result(_gvn.transform(new MulDNode(base, base)));
1825 return true;
1826 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1827 // Special case: pow(x, 0.5) => sqrt(x)
1828 Node* base = argument(0);
1829 Node* zero = _gvn.zerocon(T_DOUBLE);
1830
1831 RegionNode* region = new RegionNode(3);
1832 Node* phi = new PhiNode(region, Type::DOUBLE);
1833
1834 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1835 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1836 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1837 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1838 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1839
1840 Node* if_pow = generate_slow_guard(test, nullptr);
1841 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1842 phi->init_req(1, value_sqrt);
1843 region->init_req(1, control());
1844
1845 if (if_pow != nullptr) {
1846 set_control(if_pow);
1847 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1848 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1849 const TypePtr* no_memory_effects = nullptr;
1850 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1851 no_memory_effects, base, top(), exp, top());
1852 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1853 #ifdef ASSERT
1854 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1855 assert(value_top == top(), "second value must be top");
1856 #endif
1857 phi->init_req(2, value_pow);
1858 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1859 }
1860
1861 C->set_has_split_ifs(true); // Has chance for split-if optimization
1862 set_control(_gvn.transform(region));
1863 record_for_igvn(region);
1864 set_result(_gvn.transform(phi));
1865
1866 return true;
1867 }
1868 }
1869
1870 return StubRoutines::dpow() != nullptr ?
1871 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1872 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1873 }
1874
1875 //------------------------------inline_math_native-----------------------------
1876 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1877 switch (id) {
1878 case vmIntrinsics::_dsin:
1879 return StubRoutines::dsin() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1882 case vmIntrinsics::_dcos:
1883 return StubRoutines::dcos() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1886 case vmIntrinsics::_dtan:
1887 return StubRoutines::dtan() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1890 case vmIntrinsics::_dtanh:
1891 return StubRoutines::dtanh() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1893 case vmIntrinsics::_dcbrt:
1894 return StubRoutines::dcbrt() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1896 case vmIntrinsics::_dexp:
1897 return StubRoutines::dexp() != nullptr ?
1898 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1899 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1900 case vmIntrinsics::_dlog:
1901 return StubRoutines::dlog() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1903 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1904 case vmIntrinsics::_dlog10:
1905 return StubRoutines::dlog10() != nullptr ?
1906 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1907 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1908
1909 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_ceil:
1911 case vmIntrinsics::_floor:
1912 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1913
1914 case vmIntrinsics::_dsqrt:
1915 case vmIntrinsics::_dsqrt_strict:
1916 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1917 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1918 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1919 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1920 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1921
1922 case vmIntrinsics::_dpow: return inline_math_pow();
1923 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1924 case vmIntrinsics::_fcopySign: return inline_math(id);
1925 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1926 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1927 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1928
1929 // These intrinsics are not yet correctly implemented
1930 case vmIntrinsics::_datan2:
1931 return false;
1932
1933 default:
1934 fatal_unexpected_iid(id);
1935 return false;
1936 }
1937 }
1938
1939 //----------------------------inline_notify-----------------------------------*
1940 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1941 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1942 address func;
1943 if (id == vmIntrinsics::_notify) {
1944 func = OptoRuntime::monitor_notify_Java();
1945 } else {
1946 func = OptoRuntime::monitor_notifyAll_Java();
1947 }
1948 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1949 make_slow_call_ex(call, env()->Throwable_klass(), false);
1950 return true;
1951 }
1952
1953
1954 //----------------------------inline_min_max-----------------------------------
1955 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1956 Node* a = nullptr;
1957 Node* b = nullptr;
1958 Node* n = nullptr;
1959 switch (id) {
1960 case vmIntrinsics::_min:
1961 case vmIntrinsics::_max:
1962 case vmIntrinsics::_minF:
1963 case vmIntrinsics::_maxF:
1964 case vmIntrinsics::_minF_strict:
1965 case vmIntrinsics::_maxF_strict:
1966 case vmIntrinsics::_min_strict:
1967 case vmIntrinsics::_max_strict:
1968 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1969 a = argument(0);
1970 b = argument(1);
1971 break;
1972 case vmIntrinsics::_minD:
1973 case vmIntrinsics::_maxD:
1974 case vmIntrinsics::_minD_strict:
1975 case vmIntrinsics::_maxD_strict:
1976 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1977 a = argument(0);
1978 b = argument(2);
1979 break;
1980 case vmIntrinsics::_minL:
1981 case vmIntrinsics::_maxL:
1982 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1983 a = argument(0);
1984 b = argument(2);
1985 break;
1986 default:
1987 fatal_unexpected_iid(id);
1988 break;
1989 }
1990
1991 switch (id) {
1992 case vmIntrinsics::_min:
1993 case vmIntrinsics::_min_strict:
1994 n = new MinINode(a, b);
1995 break;
1996 case vmIntrinsics::_max:
1997 case vmIntrinsics::_max_strict:
1998 n = new MaxINode(a, b);
1999 break;
2000 case vmIntrinsics::_minF:
2001 case vmIntrinsics::_minF_strict:
2002 n = new MinFNode(a, b);
2003 break;
2004 case vmIntrinsics::_maxF:
2005 case vmIntrinsics::_maxF_strict:
2006 n = new MaxFNode(a, b);
2007 break;
2008 case vmIntrinsics::_minD:
2009 case vmIntrinsics::_minD_strict:
2010 n = new MinDNode(a, b);
2011 break;
2012 case vmIntrinsics::_maxD:
2013 case vmIntrinsics::_maxD_strict:
2014 n = new MaxDNode(a, b);
2015 break;
2016 case vmIntrinsics::_minL:
2017 n = new MinLNode(_gvn.C, a, b);
2018 break;
2019 case vmIntrinsics::_maxL:
2020 n = new MaxLNode(_gvn.C, a, b);
2021 break;
2022 default:
2023 fatal_unexpected_iid(id);
2024 break;
2025 }
2026
2027 set_result(_gvn.transform(n));
2028 return true;
2029 }
2030
2031 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2032 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2033 env()->ArithmeticException_instance())) {
2034 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2035 // so let's bail out intrinsic rather than risking deopting again.
2036 return false;
2037 }
2038
2039 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2040 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2041 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2042 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2043
2044 {
2045 PreserveJVMState pjvms(this);
2046 PreserveReexecuteState preexecs(this);
2047 jvms()->set_should_reexecute(true);
2048
2049 set_control(slow_path);
2050 set_i_o(i_o());
2051
2052 builtin_throw(Deoptimization::Reason_intrinsic,
2053 env()->ArithmeticException_instance(),
2054 /*allow_too_many_traps*/ false);
2055 }
2056
2057 set_control(fast_path);
2058 set_result(math);
2059 return true;
2060 }
2061
2062 template <typename OverflowOp>
2063 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2064 typedef typename OverflowOp::MathOp MathOp;
2065
2066 MathOp* mathOp = new MathOp(arg1, arg2);
2067 Node* operation = _gvn.transform( mathOp );
2068 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2069 return inline_math_mathExact(operation, ofcheck);
2070 }
2071
2072 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2073 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2074 }
2075
2076 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2077 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2078 }
2079
2080 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2081 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2082 }
2083
2084 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2085 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2086 }
2087
2088 bool LibraryCallKit::inline_math_negateExactI() {
2089 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2090 }
2091
2092 bool LibraryCallKit::inline_math_negateExactL() {
2093 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2094 }
2095
2096 bool LibraryCallKit::inline_math_multiplyExactI() {
2097 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2098 }
2099
2100 bool LibraryCallKit::inline_math_multiplyExactL() {
2101 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2102 }
2103
2104 bool LibraryCallKit::inline_math_multiplyHigh() {
2105 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2106 return true;
2107 }
2108
2109 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2110 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2111 return true;
2112 }
2113
2114 inline int
2115 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2116 const TypePtr* base_type = TypePtr::NULL_PTR;
2117 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2118 if (base_type == nullptr) {
2119 // Unknown type.
2120 return Type::AnyPtr;
2121 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2122 // Since this is a null+long form, we have to switch to a rawptr.
2123 base = _gvn.transform(new CastX2PNode(offset));
2124 offset = MakeConX(0);
2125 return Type::RawPtr;
2126 } else if (base_type->base() == Type::RawPtr) {
2127 return Type::RawPtr;
2128 } else if (base_type->isa_oopptr()) {
2129 // Base is never null => always a heap address.
2130 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2131 return Type::OopPtr;
2132 }
2133 // Offset is small => always a heap address.
2134 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2135 if (offset_type != nullptr &&
2136 base_type->offset() == 0 && // (should always be?)
2137 offset_type->_lo >= 0 &&
2138 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2139 return Type::OopPtr;
2140 } else if (type == T_OBJECT) {
2141 // off heap access to an oop doesn't make any sense. Has to be on
2142 // heap.
2143 return Type::OopPtr;
2144 }
2145 // Otherwise, it might either be oop+off or null+addr.
2146 return Type::AnyPtr;
2147 } else {
2148 // No information:
2149 return Type::AnyPtr;
2150 }
2151 }
2152
2153 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2154 Node* uncasted_base = base;
2155 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2156 if (kind == Type::RawPtr) {
2157 return basic_plus_adr(top(), uncasted_base, offset);
2158 } else if (kind == Type::AnyPtr) {
2159 assert(base == uncasted_base, "unexpected base change");
2160 if (can_cast) {
2161 if (!_gvn.type(base)->speculative_maybe_null() &&
2162 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2163 // According to profiling, this access is always on
2164 // heap. Casting the base to not null and thus avoiding membars
2165 // around the access should allow better optimizations
2166 Node* null_ctl = top();
2167 base = null_check_oop(base, &null_ctl, true, true, true);
2168 assert(null_ctl->is_top(), "no null control here");
2169 return basic_plus_adr(base, offset);
2170 } else if (_gvn.type(base)->speculative_always_null() &&
2171 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2172 // According to profiling, this access is always off
2173 // heap.
2174 base = null_assert(base);
2175 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2176 offset = MakeConX(0);
2177 return basic_plus_adr(top(), raw_base, offset);
2178 }
2179 }
2180 // We don't know if it's an on heap or off heap access. Fall back
2181 // to raw memory access.
2182 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2183 return basic_plus_adr(top(), raw, offset);
2184 } else {
2185 assert(base == uncasted_base, "unexpected base change");
2186 // We know it's an on heap access so base can't be null
2187 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2188 base = must_be_not_null(base, true);
2189 }
2190 return basic_plus_adr(base, offset);
2191 }
2192 }
2193
2194 //--------------------------inline_number_methods-----------------------------
2195 // inline int Integer.numberOfLeadingZeros(int)
2196 // inline int Long.numberOfLeadingZeros(long)
2197 //
2198 // inline int Integer.numberOfTrailingZeros(int)
2199 // inline int Long.numberOfTrailingZeros(long)
2200 //
2201 // inline int Integer.bitCount(int)
2202 // inline int Long.bitCount(long)
2203 //
2204 // inline char Character.reverseBytes(char)
2205 // inline short Short.reverseBytes(short)
2206 // inline int Integer.reverseBytes(int)
2207 // inline long Long.reverseBytes(long)
2208 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2209 Node* arg = argument(0);
2210 Node* n = nullptr;
2211 switch (id) {
2212 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2213 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2214 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2215 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2216 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2217 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2218 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2219 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2220 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2221 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2222 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2223 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2224 default: fatal_unexpected_iid(id); break;
2225 }
2226 set_result(_gvn.transform(n));
2227 return true;
2228 }
2229
2230 //--------------------------inline_bitshuffle_methods-----------------------------
2231 // inline int Integer.compress(int, int)
2232 // inline int Integer.expand(int, int)
2233 // inline long Long.compress(long, long)
2234 // inline long Long.expand(long, long)
2235 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2236 Node* n = nullptr;
2237 switch (id) {
2238 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2239 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2240 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2241 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2242 default: fatal_unexpected_iid(id); break;
2243 }
2244 set_result(_gvn.transform(n));
2245 return true;
2246 }
2247
2248 //--------------------------inline_number_methods-----------------------------
2249 // inline int Integer.compareUnsigned(int, int)
2250 // inline int Long.compareUnsigned(long, long)
2251 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2252 Node* arg1 = argument(0);
2253 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2254 Node* n = nullptr;
2255 switch (id) {
2256 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2257 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2258 default: fatal_unexpected_iid(id); break;
2259 }
2260 set_result(_gvn.transform(n));
2261 return true;
2262 }
2263
2264 //--------------------------inline_unsigned_divmod_methods-----------------------------
2265 // inline int Integer.divideUnsigned(int, int)
2266 // inline int Integer.remainderUnsigned(int, int)
2267 // inline long Long.divideUnsigned(long, long)
2268 // inline long Long.remainderUnsigned(long, long)
2269 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2270 Node* n = nullptr;
2271 switch (id) {
2272 case vmIntrinsics::_divideUnsigned_i: {
2273 zero_check_int(argument(1));
2274 // Compile-time detect of null-exception
2275 if (stopped()) {
2276 return true; // keep the graph constructed so far
2277 }
2278 n = new UDivINode(control(), argument(0), argument(1));
2279 break;
2280 }
2281 case vmIntrinsics::_divideUnsigned_l: {
2282 zero_check_long(argument(2));
2283 // Compile-time detect of null-exception
2284 if (stopped()) {
2285 return true; // keep the graph constructed so far
2286 }
2287 n = new UDivLNode(control(), argument(0), argument(2));
2288 break;
2289 }
2290 case vmIntrinsics::_remainderUnsigned_i: {
2291 zero_check_int(argument(1));
2292 // Compile-time detect of null-exception
2293 if (stopped()) {
2294 return true; // keep the graph constructed so far
2295 }
2296 n = new UModINode(control(), argument(0), argument(1));
2297 break;
2298 }
2299 case vmIntrinsics::_remainderUnsigned_l: {
2300 zero_check_long(argument(2));
2301 // Compile-time detect of null-exception
2302 if (stopped()) {
2303 return true; // keep the graph constructed so far
2304 }
2305 n = new UModLNode(control(), argument(0), argument(2));
2306 break;
2307 }
2308 default: fatal_unexpected_iid(id); break;
2309 }
2310 set_result(_gvn.transform(n));
2311 return true;
2312 }
2313
2314 //----------------------------inline_unsafe_access----------------------------
2315
2316 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2317 // Attempt to infer a sharper value type from the offset and base type.
2318 ciKlass* sharpened_klass = nullptr;
2319
2320 // See if it is an instance field, with an object type.
2321 if (alias_type->field() != nullptr) {
2322 if (alias_type->field()->type()->is_klass()) {
2323 sharpened_klass = alias_type->field()->type()->as_klass();
2324 }
2325 }
2326
2327 const TypeOopPtr* result = nullptr;
2328 // See if it is a narrow oop array.
2329 if (adr_type->isa_aryptr()) {
2330 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2331 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2332 if (elem_type != nullptr && elem_type->is_loaded()) {
2333 // Sharpen the value type.
2334 result = elem_type;
2335 }
2336 }
2337 }
2338
2339 // The sharpened class might be unloaded if there is no class loader
2340 // contraint in place.
2341 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2342 // Sharpen the value type.
2343 result = TypeOopPtr::make_from_klass(sharpened_klass);
2344 }
2345 if (result != nullptr) {
2346 #ifndef PRODUCT
2347 if (C->print_intrinsics() || C->print_inlining()) {
2348 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2349 tty->print(" sharpened value: "); result->dump(); tty->cr();
2350 }
2351 #endif
2352 }
2353 return result;
2354 }
2355
2356 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2357 switch (kind) {
2358 case Relaxed:
2359 return MO_UNORDERED;
2360 case Opaque:
2361 return MO_RELAXED;
2362 case Acquire:
2363 return MO_ACQUIRE;
2364 case Release:
2365 return MO_RELEASE;
2366 case Volatile:
2367 return MO_SEQ_CST;
2368 default:
2369 ShouldNotReachHere();
2370 return 0;
2371 }
2372 }
2373
2374 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2375 _kit(kit),
2376 _sp(kit->sp()),
2377 _jvms(kit->jvms()),
2378 _map(kit->clone_map()),
2379 _discarded(false)
2380 {
2381 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2382 Node* out = kit->control()->fast_out(i);
2383 if (out->is_CFG()) {
2384 _ctrl_succ.push(out);
2385 }
2386 }
2387 }
2388
2389 LibraryCallKit::SavedState::~SavedState() {
2390 if (_discarded) {
2391 _kit->destruct_map_clone(_map);
2392 return;
2393 }
2394 _kit->jvms()->set_map(_map);
2395 _kit->jvms()->set_sp(_sp);
2396 _map->set_jvms(_kit->jvms());
2397 _kit->set_map(_map);
2398 _kit->set_sp(_sp);
2399 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2400 Node* out = _kit->control()->fast_out(i);
2401 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2402 _kit->_gvn.hash_delete(out);
2403 out->set_req(0, _kit->C->top());
2404 _kit->C->record_for_igvn(out);
2405 --i; --imax;
2406 _kit->_gvn.hash_find_insert(out);
2407 }
2408 }
2409 }
2410
2411 void LibraryCallKit::SavedState::discard() {
2412 _discarded = true;
2413 }
2414
2415 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2416 if (callee()->is_static()) return false; // caller must have the capability!
2417 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2418 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2419 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2420 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2421
2422 if (is_reference_type(type)) {
2423 decorators |= ON_UNKNOWN_OOP_REF;
2424 }
2425
2426 if (unaligned) {
2427 decorators |= C2_UNALIGNED;
2428 }
2429
2430 #ifndef PRODUCT
2431 {
2432 ResourceMark rm;
2433 // Check the signatures.
2434 ciSignature* sig = callee()->signature();
2435 #ifdef ASSERT
2436 if (!is_store) {
2437 // Object getReference(Object base, int/long offset), etc.
2438 BasicType rtype = sig->return_type()->basic_type();
2439 assert(rtype == type, "getter must return the expected value");
2440 assert(sig->count() == 2, "oop getter has 2 arguments");
2441 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2442 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2443 } else {
2444 // void putReference(Object base, int/long offset, Object x), etc.
2445 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2446 assert(sig->count() == 3, "oop putter has 3 arguments");
2447 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2448 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2449 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2450 assert(vtype == type, "putter must accept the expected value");
2451 }
2452 #endif // ASSERT
2453 }
2454 #endif //PRODUCT
2455
2456 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2457
2458 Node* receiver = argument(0); // type: oop
2459
2460 // Build address expression.
2461 Node* heap_base_oop = top();
2462
2463 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2464 Node* base = argument(1); // type: oop
2465 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2466 Node* offset = argument(2); // type: long
2467 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2468 // to be plain byte offsets, which are also the same as those accepted
2469 // by oopDesc::field_addr.
2470 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2471 "fieldOffset must be byte-scaled");
2472 // 32-bit machines ignore the high half!
2473 offset = ConvL2X(offset);
2474
2475 // Save state and restore on bailout
2476 SavedState old_state(this);
2477
2478 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2479 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2480
2481 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2482 if (type != T_OBJECT) {
2483 decorators |= IN_NATIVE; // off-heap primitive access
2484 } else {
2485 return false; // off-heap oop accesses are not supported
2486 }
2487 } else {
2488 heap_base_oop = base; // on-heap or mixed access
2489 }
2490
2491 // Can base be null? Otherwise, always on-heap access.
2492 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2493
2494 if (!can_access_non_heap) {
2495 decorators |= IN_HEAP;
2496 }
2497
2498 Node* val = is_store ? argument(4) : nullptr;
2499
2500 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2501 if (adr_type == TypePtr::NULL_PTR) {
2502 return false; // off-heap access with zero address
2503 }
2504
2505 // Try to categorize the address.
2506 Compile::AliasType* alias_type = C->alias_type(adr_type);
2507 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2508
2509 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2510 alias_type->adr_type() == TypeAryPtr::RANGE) {
2511 return false; // not supported
2512 }
2513
2514 bool mismatched = false;
2515 BasicType bt = alias_type->basic_type();
2516 if (bt != T_ILLEGAL) {
2517 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2518 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2519 // Alias type doesn't differentiate between byte[] and boolean[]).
2520 // Use address type to get the element type.
2521 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2522 }
2523 if (is_reference_type(bt, true)) {
2524 // accessing an array field with getReference is not a mismatch
2525 bt = T_OBJECT;
2526 }
2527 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2528 // Don't intrinsify mismatched object accesses
2529 return false;
2530 }
2531 mismatched = (bt != type);
2532 } else if (alias_type->adr_type()->isa_oopptr()) {
2533 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2534 }
2535
2536 old_state.discard();
2537 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2538
2539 if (mismatched) {
2540 decorators |= C2_MISMATCHED;
2541 }
2542
2543 // First guess at the value type.
2544 const Type *value_type = Type::get_const_basic_type(type);
2545
2546 // Figure out the memory ordering.
2547 decorators |= mo_decorator_for_access_kind(kind);
2548
2549 if (!is_store && type == T_OBJECT) {
2550 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2551 if (tjp != nullptr) {
2552 value_type = tjp;
2553 }
2554 }
2555
2556 receiver = null_check(receiver);
2557 if (stopped()) {
2558 return true;
2559 }
2560 // Heap pointers get a null-check from the interpreter,
2561 // as a courtesy. However, this is not guaranteed by Unsafe,
2562 // and it is not possible to fully distinguish unintended nulls
2563 // from intended ones in this API.
2564
2565 if (!is_store) {
2566 Node* p = nullptr;
2567 // Try to constant fold a load from a constant field
2568 ciField* field = alias_type->field();
2569 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2570 // final or stable field
2571 p = make_constant_from_field(field, heap_base_oop);
2572 }
2573
2574 if (p == nullptr) { // Could not constant fold the load
2575 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2576 // Normalize the value returned by getBoolean in the following cases
2577 if (type == T_BOOLEAN &&
2578 (mismatched ||
2579 heap_base_oop == top() || // - heap_base_oop is null or
2580 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2581 // and the unsafe access is made to large offset
2582 // (i.e., larger than the maximum offset necessary for any
2583 // field access)
2584 ) {
2585 IdealKit ideal = IdealKit(this);
2586 #define __ ideal.
2587 IdealVariable normalized_result(ideal);
2588 __ declarations_done();
2589 __ set(normalized_result, p);
2590 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2591 __ set(normalized_result, ideal.ConI(1));
2592 ideal.end_if();
2593 final_sync(ideal);
2594 p = __ value(normalized_result);
2595 #undef __
2596 }
2597 }
2598 if (type == T_ADDRESS) {
2599 p = gvn().transform(new CastP2XNode(nullptr, p));
2600 p = ConvX2UL(p);
2601 }
2602 // The load node has the control of the preceding MemBarCPUOrder. All
2603 // following nodes will have the control of the MemBarCPUOrder inserted at
2604 // the end of this method. So, pushing the load onto the stack at a later
2605 // point is fine.
2606 set_result(p);
2607 } else {
2608 if (bt == T_ADDRESS) {
2609 // Repackage the long as a pointer.
2610 val = ConvL2X(val);
2611 val = gvn().transform(new CastX2PNode(val));
2612 }
2613 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2614 }
2615
2616 return true;
2617 }
2618
2619 //----------------------------inline_unsafe_load_store----------------------------
2620 // This method serves a couple of different customers (depending on LoadStoreKind):
2621 //
2622 // LS_cmp_swap:
2623 //
2624 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2625 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2626 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2627 //
2628 // LS_cmp_swap_weak:
2629 //
2630 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2631 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2632 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2633 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2634 //
2635 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2636 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2637 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2638 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2639 //
2640 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2641 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2642 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2643 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2644 //
2645 // LS_cmp_exchange:
2646 //
2647 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2648 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2649 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2650 //
2651 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2652 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2653 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2654 //
2655 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2656 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2657 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2658 //
2659 // LS_get_add:
2660 //
2661 // int getAndAddInt( Object o, long offset, int delta)
2662 // long getAndAddLong(Object o, long offset, long delta)
2663 //
2664 // LS_get_set:
2665 //
2666 // int getAndSet(Object o, long offset, int newValue)
2667 // long getAndSet(Object o, long offset, long newValue)
2668 // Object getAndSet(Object o, long offset, Object newValue)
2669 //
2670 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2671 // This basic scheme here is the same as inline_unsafe_access, but
2672 // differs in enough details that combining them would make the code
2673 // overly confusing. (This is a true fact! I originally combined
2674 // them, but even I was confused by it!) As much code/comments as
2675 // possible are retained from inline_unsafe_access though to make
2676 // the correspondences clearer. - dl
2677
2678 if (callee()->is_static()) return false; // caller must have the capability!
2679
2680 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2681 decorators |= mo_decorator_for_access_kind(access_kind);
2682
2683 #ifndef PRODUCT
2684 BasicType rtype;
2685 {
2686 ResourceMark rm;
2687 // Check the signatures.
2688 ciSignature* sig = callee()->signature();
2689 rtype = sig->return_type()->basic_type();
2690 switch(kind) {
2691 case LS_get_add:
2692 case LS_get_set: {
2693 // Check the signatures.
2694 #ifdef ASSERT
2695 assert(rtype == type, "get and set must return the expected type");
2696 assert(sig->count() == 3, "get and set has 3 arguments");
2697 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2698 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2699 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2700 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2701 #endif // ASSERT
2702 break;
2703 }
2704 case LS_cmp_swap:
2705 case LS_cmp_swap_weak: {
2706 // Check the signatures.
2707 #ifdef ASSERT
2708 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2709 assert(sig->count() == 4, "CAS has 4 arguments");
2710 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2711 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2712 #endif // ASSERT
2713 break;
2714 }
2715 case LS_cmp_exchange: {
2716 // Check the signatures.
2717 #ifdef ASSERT
2718 assert(rtype == type, "CAS must return the expected type");
2719 assert(sig->count() == 4, "CAS has 4 arguments");
2720 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2721 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2722 #endif // ASSERT
2723 break;
2724 }
2725 default:
2726 ShouldNotReachHere();
2727 }
2728 }
2729 #endif //PRODUCT
2730
2731 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2732
2733 // Get arguments:
2734 Node* receiver = nullptr;
2735 Node* base = nullptr;
2736 Node* offset = nullptr;
2737 Node* oldval = nullptr;
2738 Node* newval = nullptr;
2739 switch(kind) {
2740 case LS_cmp_swap:
2741 case LS_cmp_swap_weak:
2742 case LS_cmp_exchange: {
2743 const bool two_slot_type = type2size[type] == 2;
2744 receiver = argument(0); // type: oop
2745 base = argument(1); // type: oop
2746 offset = argument(2); // type: long
2747 oldval = argument(4); // type: oop, int, or long
2748 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2749 break;
2750 }
2751 case LS_get_add:
2752 case LS_get_set: {
2753 receiver = argument(0); // type: oop
2754 base = argument(1); // type: oop
2755 offset = argument(2); // type: long
2756 oldval = nullptr;
2757 newval = argument(4); // type: oop, int, or long
2758 break;
2759 }
2760 default:
2761 ShouldNotReachHere();
2762 }
2763
2764 // Build field offset expression.
2765 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2766 // to be plain byte offsets, which are also the same as those accepted
2767 // by oopDesc::field_addr.
2768 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2769 // 32-bit machines ignore the high half of long offsets
2770 offset = ConvL2X(offset);
2771 // Save state and restore on bailout
2772 SavedState old_state(this);
2773 Node* adr = make_unsafe_address(base, offset,type, false);
2774 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2775
2776 Compile::AliasType* alias_type = C->alias_type(adr_type);
2777 BasicType bt = alias_type->basic_type();
2778 if (bt != T_ILLEGAL &&
2779 (is_reference_type(bt) != (type == T_OBJECT))) {
2780 // Don't intrinsify mismatched object accesses.
2781 return false;
2782 }
2783
2784 old_state.discard();
2785
2786 // For CAS, unlike inline_unsafe_access, there seems no point in
2787 // trying to refine types. Just use the coarse types here.
2788 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2789 const Type *value_type = Type::get_const_basic_type(type);
2790
2791 switch (kind) {
2792 case LS_get_set:
2793 case LS_cmp_exchange: {
2794 if (type == T_OBJECT) {
2795 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2796 if (tjp != nullptr) {
2797 value_type = tjp;
2798 }
2799 }
2800 break;
2801 }
2802 case LS_cmp_swap:
2803 case LS_cmp_swap_weak:
2804 case LS_get_add:
2805 break;
2806 default:
2807 ShouldNotReachHere();
2808 }
2809
2810 // Null check receiver.
2811 receiver = null_check(receiver);
2812 if (stopped()) {
2813 return true;
2814 }
2815
2816 int alias_idx = C->get_alias_index(adr_type);
2817
2818 if (is_reference_type(type)) {
2819 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2820
2821 // Transformation of a value which could be null pointer (CastPP #null)
2822 // could be delayed during Parse (for example, in adjust_map_after_if()).
2823 // Execute transformation here to avoid barrier generation in such case.
2824 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2825 newval = _gvn.makecon(TypePtr::NULL_PTR);
2826
2827 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2828 // Refine the value to a null constant, when it is known to be null
2829 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2830 }
2831 }
2832
2833 Node* result = nullptr;
2834 switch (kind) {
2835 case LS_cmp_exchange: {
2836 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2837 oldval, newval, value_type, type, decorators);
2838 break;
2839 }
2840 case LS_cmp_swap_weak:
2841 decorators |= C2_WEAK_CMPXCHG;
2842 case LS_cmp_swap: {
2843 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2844 oldval, newval, value_type, type, decorators);
2845 break;
2846 }
2847 case LS_get_set: {
2848 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2849 newval, value_type, type, decorators);
2850 break;
2851 }
2852 case LS_get_add: {
2853 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2854 newval, value_type, type, decorators);
2855 break;
2856 }
2857 default:
2858 ShouldNotReachHere();
2859 }
2860
2861 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2862 set_result(result);
2863 return true;
2864 }
2865
2866 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2867 // Regardless of form, don't allow previous ld/st to move down,
2868 // then issue acquire, release, or volatile mem_bar.
2869 insert_mem_bar(Op_MemBarCPUOrder);
2870 switch(id) {
2871 case vmIntrinsics::_loadFence:
2872 insert_mem_bar(Op_LoadFence);
2873 return true;
2874 case vmIntrinsics::_storeFence:
2875 insert_mem_bar(Op_StoreFence);
2876 return true;
2877 case vmIntrinsics::_storeStoreFence:
2878 insert_mem_bar(Op_StoreStoreFence);
2879 return true;
2880 case vmIntrinsics::_fullFence:
2881 insert_mem_bar(Op_MemBarVolatile);
2882 return true;
2883 default:
2884 fatal_unexpected_iid(id);
2885 return false;
2886 }
2887 }
2888
2889 bool LibraryCallKit::inline_onspinwait() {
2890 insert_mem_bar(Op_OnSpinWait);
2891 return true;
2892 }
2893
2894 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2895 if (!kls->is_Con()) {
2896 return true;
2897 }
2898 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2899 if (klsptr == nullptr) {
2900 return true;
2901 }
2902 ciInstanceKlass* ik = klsptr->instance_klass();
2903 // don't need a guard for a klass that is already initialized
2904 return !ik->is_initialized();
2905 }
2906
2907 //----------------------------inline_unsafe_writeback0-------------------------
2908 // public native void Unsafe.writeback0(long address)
2909 bool LibraryCallKit::inline_unsafe_writeback0() {
2910 if (!Matcher::has_match_rule(Op_CacheWB)) {
2911 return false;
2912 }
2913 #ifndef PRODUCT
2914 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2915 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2916 ciSignature* sig = callee()->signature();
2917 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2918 #endif
2919 null_check_receiver(); // null-check, then ignore
2920 Node *addr = argument(1);
2921 addr = new CastX2PNode(addr);
2922 addr = _gvn.transform(addr);
2923 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2924 flush = _gvn.transform(flush);
2925 set_memory(flush, TypeRawPtr::BOTTOM);
2926 return true;
2927 }
2928
2929 //----------------------------inline_unsafe_writeback0-------------------------
2930 // public native void Unsafe.writeback0(long address)
2931 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2932 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2933 return false;
2934 }
2935 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2936 return false;
2937 }
2938 #ifndef PRODUCT
2939 assert(Matcher::has_match_rule(Op_CacheWB),
2940 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2941 : "found match rule for CacheWBPostSync but not CacheWB"));
2942
2943 #endif
2944 null_check_receiver(); // null-check, then ignore
2945 Node *sync;
2946 if (is_pre) {
2947 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2948 } else {
2949 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2950 }
2951 sync = _gvn.transform(sync);
2952 set_memory(sync, TypeRawPtr::BOTTOM);
2953 return true;
2954 }
2955
2956 //----------------------------inline_unsafe_allocate---------------------------
2957 // public native Object Unsafe.allocateInstance(Class<?> cls);
2958 bool LibraryCallKit::inline_unsafe_allocate() {
2959
2960 #if INCLUDE_JVMTI
2961 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2962 return false;
2963 }
2964 #endif //INCLUDE_JVMTI
2965
2966 if (callee()->is_static()) return false; // caller must have the capability!
2967
2968 null_check_receiver(); // null-check, then ignore
2969 Node* cls = null_check(argument(1));
2970 if (stopped()) return true;
2971
2972 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2973 kls = null_check(kls);
2974 if (stopped()) return true; // argument was like int.class
2975
2976 #if INCLUDE_JVMTI
2977 // Don't try to access new allocated obj in the intrinsic.
2978 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2979 // Deoptimize and allocate in interpreter instead.
2980 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2981 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2982 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2983 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2984 {
2985 BuildCutout unless(this, tst, PROB_MAX);
2986 uncommon_trap(Deoptimization::Reason_intrinsic,
2987 Deoptimization::Action_make_not_entrant);
2988 }
2989 if (stopped()) {
2990 return true;
2991 }
2992 #endif //INCLUDE_JVMTI
2993
2994 Node* test = nullptr;
2995 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2996 // Note: The argument might still be an illegal value like
2997 // Serializable.class or Object[].class. The runtime will handle it.
2998 // But we must make an explicit check for initialization.
2999 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3000 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3001 // can generate code to load it as unsigned byte.
3002 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3003 Node* bits = intcon(InstanceKlass::fully_initialized);
3004 test = _gvn.transform(new SubINode(inst, bits));
3005 // The 'test' is non-zero if we need to take a slow path.
3006 }
3007
3008 Node* obj = new_instance(kls, test);
3009 set_result(obj);
3010 return true;
3011 }
3012
3013 //------------------------inline_native_time_funcs--------------
3014 // inline code for System.currentTimeMillis() and System.nanoTime()
3015 // these have the same type and signature
3016 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3017 const TypeFunc* tf = OptoRuntime::void_long_Type();
3018 const TypePtr* no_memory_effects = nullptr;
3019 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3020 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3021 #ifdef ASSERT
3022 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3023 assert(value_top == top(), "second value must be top");
3024 #endif
3025 set_result(value);
3026 return true;
3027 }
3028
3029
3030 #if INCLUDE_JVMTI
3031
3032 // When notifications are disabled then just update the VTMS transition bit and return.
3033 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3034 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3035 if (!DoJVMTIVirtualThreadTransitions) {
3036 return true;
3037 }
3038 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3039 IdealKit ideal(this);
3040
3041 Node* ONE = ideal.ConI(1);
3042 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3043 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3044 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3045
3046 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3047 sync_kit(ideal);
3048 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3049 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3050 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3051 ideal.sync_kit(this);
3052 } ideal.else_(); {
3053 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3054 Node* thread = ideal.thread();
3055 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3056 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3057
3058 sync_kit(ideal);
3059 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3060 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3061
3062 ideal.sync_kit(this);
3063 } ideal.end_if();
3064 final_sync(ideal);
3065
3066 return true;
3067 }
3068
3069 // Always update the is_disable_suspend bit.
3070 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3071 if (!DoJVMTIVirtualThreadTransitions) {
3072 return true;
3073 }
3074 IdealKit ideal(this);
3075
3076 {
3077 // unconditionally update the is_disable_suspend bit in current JavaThread
3078 Node* thread = ideal.thread();
3079 Node* arg = _gvn.transform(argument(0)); // argument for notification
3080 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3081 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3082
3083 sync_kit(ideal);
3084 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3085 ideal.sync_kit(this);
3086 }
3087 final_sync(ideal);
3088
3089 return true;
3090 }
3091
3092 #endif // INCLUDE_JVMTI
3093
3094 #ifdef JFR_HAVE_INTRINSICS
3095
3096 /**
3097 * if oop->klass != null
3098 * // normal class
3099 * epoch = _epoch_state ? 2 : 1
3100 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3101 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3102 * }
3103 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3104 * else
3105 * // primitive class
3106 * if oop->array_klass != null
3107 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3108 * else
3109 * id = LAST_TYPE_ID + 1 // void class path
3110 * if (!signaled)
3111 * signaled = true
3112 */
3113 bool LibraryCallKit::inline_native_classID() {
3114 Node* cls = argument(0);
3115
3116 IdealKit ideal(this);
3117 #define __ ideal.
3118 IdealVariable result(ideal); __ declarations_done();
3119 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3120 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3121 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3122
3123
3124 __ if_then(kls, BoolTest::ne, null()); {
3125 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3126 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3127
3128 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3129 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3130 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3131 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3132 mask = _gvn.transform(new OrLNode(mask, epoch));
3133 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3134
3135 float unlikely = PROB_UNLIKELY(0.999);
3136 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3137 sync_kit(ideal);
3138 make_runtime_call(RC_LEAF,
3139 OptoRuntime::class_id_load_barrier_Type(),
3140 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3141 "class id load barrier",
3142 TypePtr::BOTTOM,
3143 kls);
3144 ideal.sync_kit(this);
3145 } __ end_if();
3146
3147 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3148 } __ else_(); {
3149 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3150 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3151 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3152 __ if_then(array_kls, BoolTest::ne, null()); {
3153 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3154 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3155 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3156 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3157 } __ else_(); {
3158 // void class case
3159 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3160 } __ end_if();
3161
3162 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3163 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3164 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3165 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3166 } __ end_if();
3167 } __ end_if();
3168
3169 final_sync(ideal);
3170 set_result(ideal.value(result));
3171 #undef __
3172 return true;
3173 }
3174
3175 //------------------------inline_native_jvm_commit------------------
3176 bool LibraryCallKit::inline_native_jvm_commit() {
3177 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3178
3179 // Save input memory and i_o state.
3180 Node* input_memory_state = reset_memory();
3181 set_all_memory(input_memory_state);
3182 Node* input_io_state = i_o();
3183
3184 // TLS.
3185 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3186 // Jfr java buffer.
3187 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3188 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3189 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3190
3191 // Load the current value of the notified field in the JfrThreadLocal.
3192 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3193 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3194
3195 // Test for notification.
3196 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3197 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3198 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3199
3200 // True branch, is notified.
3201 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3202 set_control(is_notified);
3203
3204 // Reset notified state.
3205 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3206 Node* notified_reset_memory = reset_memory();
3207
3208 // 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.
3209 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3210 // Convert the machine-word to a long.
3211 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3212
3213 // False branch, not notified.
3214 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3215 set_control(not_notified);
3216 set_all_memory(input_memory_state);
3217
3218 // Arg is the next position as a long.
3219 Node* arg = argument(0);
3220 // Convert long to machine-word.
3221 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3222
3223 // Store the next_position to the underlying jfr java buffer.
3224 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3225
3226 Node* commit_memory = reset_memory();
3227 set_all_memory(commit_memory);
3228
3229 // 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.
3230 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3231 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3232 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3233
3234 // And flags with lease constant.
3235 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3236
3237 // Branch on lease to conditionalize returning the leased java buffer.
3238 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3239 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3240 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3241
3242 // False branch, not a lease.
3243 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3244
3245 // True branch, is lease.
3246 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3247 set_control(is_lease);
3248
3249 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3250 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3251 OptoRuntime::void_void_Type(),
3252 SharedRuntime::jfr_return_lease(),
3253 "return_lease", TypePtr::BOTTOM);
3254 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3255
3256 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3257 record_for_igvn(lease_compare_rgn);
3258 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3259 record_for_igvn(lease_compare_mem);
3260 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3261 record_for_igvn(lease_compare_io);
3262 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3263 record_for_igvn(lease_result_value);
3264
3265 // Update control and phi nodes.
3266 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3267 lease_compare_rgn->init_req(_false_path, not_lease);
3268
3269 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3270 lease_compare_mem->init_req(_false_path, commit_memory);
3271
3272 lease_compare_io->init_req(_true_path, i_o());
3273 lease_compare_io->init_req(_false_path, input_io_state);
3274
3275 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3276 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3277
3278 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3279 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3280 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3281 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3282
3283 // Update control and phi nodes.
3284 result_rgn->init_req(_true_path, is_notified);
3285 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3286
3287 result_mem->init_req(_true_path, notified_reset_memory);
3288 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3289
3290 result_io->init_req(_true_path, input_io_state);
3291 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3292
3293 result_value->init_req(_true_path, current_pos);
3294 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3295
3296 // Set output state.
3297 set_control(_gvn.transform(result_rgn));
3298 set_all_memory(_gvn.transform(result_mem));
3299 set_i_o(_gvn.transform(result_io));
3300 set_result(result_rgn, result_value);
3301 return true;
3302 }
3303
3304 /*
3305 * The intrinsic is a model of this pseudo-code:
3306 *
3307 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3308 * jobject h_event_writer = tl->java_event_writer();
3309 * if (h_event_writer == nullptr) {
3310 * return nullptr;
3311 * }
3312 * oop threadObj = Thread::threadObj();
3313 * oop vthread = java_lang_Thread::vthread(threadObj);
3314 * traceid tid;
3315 * bool pinVirtualThread;
3316 * bool excluded;
3317 * if (vthread != threadObj) { // i.e. current thread is virtual
3318 * tid = java_lang_Thread::tid(vthread);
3319 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3320 * pinVirtualThread = VMContinuations;
3321 * excluded = vthread_epoch_raw & excluded_mask;
3322 * if (!excluded) {
3323 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3324 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3325 * if (vthread_epoch != current_epoch) {
3326 * write_checkpoint();
3327 * }
3328 * }
3329 * } else {
3330 * tid = java_lang_Thread::tid(threadObj);
3331 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3332 * pinVirtualThread = false;
3333 * excluded = thread_epoch_raw & excluded_mask;
3334 * }
3335 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3336 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3337 * if (tid_in_event_writer != tid) {
3338 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3339 * setField(event_writer, "excluded", excluded);
3340 * setField(event_writer, "threadID", tid);
3341 * }
3342 * return event_writer
3343 */
3344 bool LibraryCallKit::inline_native_getEventWriter() {
3345 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3346
3347 // Save input memory and i_o state.
3348 Node* input_memory_state = reset_memory();
3349 set_all_memory(input_memory_state);
3350 Node* input_io_state = i_o();
3351
3352 // The most significant bit of the u2 is used to denote thread exclusion
3353 Node* excluded_shift = _gvn.intcon(15);
3354 Node* excluded_mask = _gvn.intcon(1 << 15);
3355 // The epoch generation is the range [1-32767]
3356 Node* epoch_mask = _gvn.intcon(32767);
3357
3358 // TLS
3359 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3360
3361 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3362 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3363
3364 // Load the eventwriter jobject handle.
3365 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3366
3367 // Null check the jobject handle.
3368 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3369 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3370 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3371
3372 // False path, jobj is null.
3373 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3374
3375 // True path, jobj is not null.
3376 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3377
3378 set_control(jobj_is_not_null);
3379
3380 // Load the threadObj for the CarrierThread.
3381 Node* threadObj = generate_current_thread(tls_ptr);
3382
3383 // Load the vthread.
3384 Node* vthread = generate_virtual_thread(tls_ptr);
3385
3386 // If vthread != threadObj, this is a virtual thread.
3387 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3388 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3389 IfNode* iff_vthread_not_equal_threadObj =
3390 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3391
3392 // False branch, fallback to threadObj.
3393 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3394 set_control(vthread_equal_threadObj);
3395
3396 // Load the tid field from the vthread object.
3397 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3398
3399 // Load the raw epoch value from the threadObj.
3400 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3401 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3402 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3403 TypeInt::CHAR, T_CHAR,
3404 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3405
3406 // Mask off the excluded information from the epoch.
3407 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3408
3409 // True branch, this is a virtual thread.
3410 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3411 set_control(vthread_not_equal_threadObj);
3412
3413 // Load the tid field from the vthread object.
3414 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3415
3416 // Continuation support determines if a virtual thread should be pinned.
3417 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3418 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3419
3420 // Load the raw epoch value from the vthread.
3421 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3422 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3423 TypeInt::CHAR, T_CHAR,
3424 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3425
3426 // Mask off the excluded information from the epoch.
3427 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3428
3429 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3430 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3431 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3432 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3433
3434 // False branch, vthread is excluded, no need to write epoch info.
3435 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3436
3437 // True branch, vthread is included, update epoch info.
3438 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3439 set_control(included);
3440
3441 // Get epoch value.
3442 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3443
3444 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3445 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3446 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3447
3448 // Compare the epoch in the vthread to the current epoch generation.
3449 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3450 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3451 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3452
3453 // False path, epoch is equal, checkpoint information is valid.
3454 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3455
3456 // True path, epoch is not equal, write a checkpoint for the vthread.
3457 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3458
3459 set_control(epoch_is_not_equal);
3460
3461 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3462 // The call also updates the native thread local thread id and the vthread with the current epoch.
3463 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3464 OptoRuntime::jfr_write_checkpoint_Type(),
3465 SharedRuntime::jfr_write_checkpoint(),
3466 "write_checkpoint", TypePtr::BOTTOM);
3467 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3468
3469 // vthread epoch != current epoch
3470 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3471 record_for_igvn(epoch_compare_rgn);
3472 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3473 record_for_igvn(epoch_compare_mem);
3474 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3475 record_for_igvn(epoch_compare_io);
3476
3477 // Update control and phi nodes.
3478 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3479 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3480 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3481 epoch_compare_mem->init_req(_false_path, input_memory_state);
3482 epoch_compare_io->init_req(_true_path, i_o());
3483 epoch_compare_io->init_req(_false_path, input_io_state);
3484
3485 // excluded != true
3486 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3487 record_for_igvn(exclude_compare_rgn);
3488 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3489 record_for_igvn(exclude_compare_mem);
3490 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3491 record_for_igvn(exclude_compare_io);
3492
3493 // Update control and phi nodes.
3494 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3495 exclude_compare_rgn->init_req(_false_path, excluded);
3496 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3497 exclude_compare_mem->init_req(_false_path, input_memory_state);
3498 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3499 exclude_compare_io->init_req(_false_path, input_io_state);
3500
3501 // vthread != threadObj
3502 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3503 record_for_igvn(vthread_compare_rgn);
3504 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3505 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3506 record_for_igvn(vthread_compare_io);
3507 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3508 record_for_igvn(tid);
3509 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3510 record_for_igvn(exclusion);
3511 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3512 record_for_igvn(pinVirtualThread);
3513
3514 // Update control and phi nodes.
3515 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3516 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3517 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3518 vthread_compare_mem->init_req(_false_path, input_memory_state);
3519 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3520 vthread_compare_io->init_req(_false_path, input_io_state);
3521 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3522 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3523 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3524 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3525 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3526 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3527
3528 // Update branch state.
3529 set_control(_gvn.transform(vthread_compare_rgn));
3530 set_all_memory(_gvn.transform(vthread_compare_mem));
3531 set_i_o(_gvn.transform(vthread_compare_io));
3532
3533 // Load the event writer oop by dereferencing the jobject handle.
3534 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3535 assert(klass_EventWriter->is_loaded(), "invariant");
3536 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3537 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3538 const TypeOopPtr* const xtype = aklass->as_instance_type();
3539 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3540 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3541
3542 // Load the current thread id from the event writer object.
3543 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3544 // Get the field offset to, conditionally, store an updated tid value later.
3545 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3546 // Get the field offset to, conditionally, store an updated exclusion value later.
3547 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3548 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3549 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3550
3551 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3552 record_for_igvn(event_writer_tid_compare_rgn);
3553 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3554 record_for_igvn(event_writer_tid_compare_mem);
3555 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3556 record_for_igvn(event_writer_tid_compare_io);
3557
3558 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3559 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3560 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3561 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3562
3563 // False path, tids are the same.
3564 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3565
3566 // True path, tid is not equal, need to update the tid in the event writer.
3567 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3568 record_for_igvn(tid_is_not_equal);
3569
3570 // Store the pin state to the event writer.
3571 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3572
3573 // Store the exclusion state to the event writer.
3574 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3575 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3576
3577 // Store the tid to the event writer.
3578 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3579
3580 // Update control and phi nodes.
3581 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3582 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3583 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3584 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3585 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3586 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3587
3588 // Result of top level CFG, Memory, IO and Value.
3589 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3590 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3591 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3592 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3593
3594 // Result control.
3595 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3596 result_rgn->init_req(_false_path, jobj_is_null);
3597
3598 // Result memory.
3599 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3600 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3601
3602 // Result IO.
3603 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3604 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3605
3606 // Result value.
3607 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3608 result_value->init_req(_false_path, null()); // return null
3609
3610 // Set output state.
3611 set_control(_gvn.transform(result_rgn));
3612 set_all_memory(_gvn.transform(result_mem));
3613 set_i_o(_gvn.transform(result_io));
3614 set_result(result_rgn, result_value);
3615 return true;
3616 }
3617
3618 /*
3619 * The intrinsic is a model of this pseudo-code:
3620 *
3621 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3622 * if (carrierThread != thread) { // is virtual thread
3623 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3624 * bool excluded = vthread_epoch_raw & excluded_mask;
3625 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3626 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3627 * if (!excluded) {
3628 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3629 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3630 * }
3631 * Atomic::release_store(&tl->_vthread, true);
3632 * return;
3633 * }
3634 * Atomic::release_store(&tl->_vthread, false);
3635 */
3636 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3637 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3638
3639 Node* input_memory_state = reset_memory();
3640 set_all_memory(input_memory_state);
3641
3642 // The most significant bit of the u2 is used to denote thread exclusion
3643 Node* excluded_mask = _gvn.intcon(1 << 15);
3644 // The epoch generation is the range [1-32767]
3645 Node* epoch_mask = _gvn.intcon(32767);
3646
3647 Node* const carrierThread = generate_current_thread(jt);
3648 // If thread != carrierThread, this is a virtual thread.
3649 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3650 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3651 IfNode* iff_thread_not_equal_carrierThread =
3652 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3653
3654 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3655
3656 // False branch, is carrierThread.
3657 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3658 // Store release
3659 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3660
3661 set_all_memory(input_memory_state);
3662
3663 // True branch, is virtual thread.
3664 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3665 set_control(thread_not_equal_carrierThread);
3666
3667 // Load the raw epoch value from the vthread.
3668 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3669 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3670 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3671
3672 // Mask off the excluded information from the epoch.
3673 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3674
3675 // Load the tid field from the thread.
3676 Node* tid = load_field_from_object(thread, "tid", "J");
3677
3678 // Store the vthread tid to the jfr thread local.
3679 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3680 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3681
3682 // Branch is_excluded to conditionalize updating the epoch .
3683 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3684 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3685 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3686
3687 // True branch, vthread is excluded, no need to write epoch info.
3688 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3689 set_control(excluded);
3690 Node* vthread_is_excluded = _gvn.intcon(1);
3691
3692 // False branch, vthread is included, update epoch info.
3693 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3694 set_control(included);
3695 Node* vthread_is_included = _gvn.intcon(0);
3696
3697 // Get epoch value.
3698 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3699
3700 // Store the vthread epoch to the jfr thread local.
3701 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3702 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3703
3704 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3705 record_for_igvn(excluded_rgn);
3706 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3707 record_for_igvn(excluded_mem);
3708 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3709 record_for_igvn(exclusion);
3710
3711 // Merge the excluded control and memory.
3712 excluded_rgn->init_req(_true_path, excluded);
3713 excluded_rgn->init_req(_false_path, included);
3714 excluded_mem->init_req(_true_path, tid_memory);
3715 excluded_mem->init_req(_false_path, included_memory);
3716 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3717 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3718
3719 // Set intermediate state.
3720 set_control(_gvn.transform(excluded_rgn));
3721 set_all_memory(excluded_mem);
3722
3723 // Store the vthread exclusion state to the jfr thread local.
3724 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3725 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3726
3727 // Store release
3728 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3729
3730 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3731 record_for_igvn(thread_compare_rgn);
3732 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3733 record_for_igvn(thread_compare_mem);
3734 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3735 record_for_igvn(vthread);
3736
3737 // Merge the thread_compare control and memory.
3738 thread_compare_rgn->init_req(_true_path, control());
3739 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3740 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3741 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3742
3743 // Set output state.
3744 set_control(_gvn.transform(thread_compare_rgn));
3745 set_all_memory(_gvn.transform(thread_compare_mem));
3746 }
3747
3748 #endif // JFR_HAVE_INTRINSICS
3749
3750 //------------------------inline_native_currentCarrierThread------------------
3751 bool LibraryCallKit::inline_native_currentCarrierThread() {
3752 Node* junk = nullptr;
3753 set_result(generate_current_thread(junk));
3754 return true;
3755 }
3756
3757 //------------------------inline_native_currentThread------------------
3758 bool LibraryCallKit::inline_native_currentThread() {
3759 Node* junk = nullptr;
3760 set_result(generate_virtual_thread(junk));
3761 return true;
3762 }
3763
3764 //------------------------inline_native_setVthread------------------
3765 bool LibraryCallKit::inline_native_setCurrentThread() {
3766 assert(C->method()->changes_current_thread(),
3767 "method changes current Thread but is not annotated ChangesCurrentThread");
3768 Node* arr = argument(1);
3769 Node* thread = _gvn.transform(new ThreadLocalNode());
3770 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3771 Node* thread_obj_handle
3772 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3773 thread_obj_handle = _gvn.transform(thread_obj_handle);
3774 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3775 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3776
3777 // Change the _monitor_owner_id of the JavaThread
3778 Node* tid = load_field_from_object(arr, "tid", "J");
3779 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3780 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3781
3782 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3783 return true;
3784 }
3785
3786 const Type* LibraryCallKit::scopedValueCache_type() {
3787 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3788 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3789 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3790
3791 // Because we create the scopedValue cache lazily we have to make the
3792 // type of the result BotPTR.
3793 bool xk = etype->klass_is_exact();
3794 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3795 return objects_type;
3796 }
3797
3798 Node* LibraryCallKit::scopedValueCache_helper() {
3799 Node* thread = _gvn.transform(new ThreadLocalNode());
3800 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3801 // We cannot use immutable_memory() because we might flip onto a
3802 // different carrier thread, at which point we'll need to use that
3803 // carrier thread's cache.
3804 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3805 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3806 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3807 }
3808
3809 //------------------------inline_native_scopedValueCache------------------
3810 bool LibraryCallKit::inline_native_scopedValueCache() {
3811 Node* cache_obj_handle = scopedValueCache_helper();
3812 const Type* objects_type = scopedValueCache_type();
3813 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3814
3815 return true;
3816 }
3817
3818 //------------------------inline_native_setScopedValueCache------------------
3819 bool LibraryCallKit::inline_native_setScopedValueCache() {
3820 Node* arr = argument(0);
3821 Node* cache_obj_handle = scopedValueCache_helper();
3822 const Type* objects_type = scopedValueCache_type();
3823
3824 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3825 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3826
3827 return true;
3828 }
3829
3830 //------------------------inline_native_Continuation_pin and unpin-----------
3831
3832 // Shared implementation routine for both pin and unpin.
3833 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3834 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3835
3836 // Save input memory.
3837 Node* input_memory_state = reset_memory();
3838 set_all_memory(input_memory_state);
3839
3840 // TLS
3841 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3842 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3843 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3844
3845 // Null check the last continuation object.
3846 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3847 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3848 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3849
3850 // False path, last continuation is null.
3851 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3852
3853 // True path, last continuation is not null.
3854 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3855
3856 set_control(continuation_is_not_null);
3857
3858 // Load the pin count from the last continuation.
3859 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3860 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3861
3862 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3863 Node* pin_count_rhs;
3864 if (unpin) {
3865 pin_count_rhs = _gvn.intcon(0);
3866 } else {
3867 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3868 }
3869 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3870 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3871 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3872
3873 // True branch, pin count over/underflow.
3874 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3875 {
3876 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3877 // which will throw IllegalStateException for pin count over/underflow.
3878 // No memory changed so far - we can use memory create by reset_memory()
3879 // at the beginning of this intrinsic. No need to call reset_memory() again.
3880 PreserveJVMState pjvms(this);
3881 set_control(pin_count_over_underflow);
3882 uncommon_trap(Deoptimization::Reason_intrinsic,
3883 Deoptimization::Action_none);
3884 assert(stopped(), "invariant");
3885 }
3886
3887 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3888 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3889 set_control(valid_pin_count);
3890
3891 Node* next_pin_count;
3892 if (unpin) {
3893 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3894 } else {
3895 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3896 }
3897
3898 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3899
3900 // Result of top level CFG and Memory.
3901 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3902 record_for_igvn(result_rgn);
3903 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3904 record_for_igvn(result_mem);
3905
3906 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3907 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3908 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3909 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3910
3911 // Set output state.
3912 set_control(_gvn.transform(result_rgn));
3913 set_all_memory(_gvn.transform(result_mem));
3914
3915 return true;
3916 }
3917
3918 //---------------------------load_mirror_from_klass----------------------------
3919 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3920 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3921 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3922 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3923 // mirror = ((OopHandle)mirror)->resolve();
3924 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3925 }
3926
3927 //-----------------------load_klass_from_mirror_common-------------------------
3928 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3929 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3930 // and branch to the given path on the region.
3931 // If never_see_null, take an uncommon trap on null, so we can optimistically
3932 // compile for the non-null case.
3933 // If the region is null, force never_see_null = true.
3934 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3935 bool never_see_null,
3936 RegionNode* region,
3937 int null_path,
3938 int offset) {
3939 if (region == nullptr) never_see_null = true;
3940 Node* p = basic_plus_adr(mirror, offset);
3941 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3942 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3943 Node* null_ctl = top();
3944 kls = null_check_oop(kls, &null_ctl, never_see_null);
3945 if (region != nullptr) {
3946 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3947 region->init_req(null_path, null_ctl);
3948 } else {
3949 assert(null_ctl == top(), "no loose ends");
3950 }
3951 return kls;
3952 }
3953
3954 //--------------------(inline_native_Class_query helpers)---------------------
3955 // Use this for JVM_ACC_INTERFACE.
3956 // Fall through if (mods & mask) == bits, take the guard otherwise.
3957 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3958 ByteSize offset, const Type* type, BasicType bt) {
3959 // Branch around if the given klass has the given modifier bit set.
3960 // Like generate_guard, adds a new path onto the region.
3961 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3962 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3963 Node* mask = intcon(modifier_mask);
3964 Node* bits = intcon(modifier_bits);
3965 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3966 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3967 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3968 return generate_fair_guard(bol, region);
3969 }
3970 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3971 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3972 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3973 }
3974
3975 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3976 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3977 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3978 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3979 }
3980
3981 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3982 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3983 }
3984
3985 //-------------------------inline_native_Class_query-------------------
3986 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3987 const Type* return_type = TypeInt::BOOL;
3988 Node* prim_return_value = top(); // what happens if it's a primitive class?
3989 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3990 bool expect_prim = false; // most of these guys expect to work on refs
3991
3992 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3993
3994 Node* mirror = argument(0);
3995 Node* obj = top();
3996
3997 switch (id) {
3998 case vmIntrinsics::_isInstance:
3999 // nothing is an instance of a primitive type
4000 prim_return_value = intcon(0);
4001 obj = argument(1);
4002 break;
4003 case vmIntrinsics::_isHidden:
4004 prim_return_value = intcon(0);
4005 break;
4006 case vmIntrinsics::_getSuperclass:
4007 prim_return_value = null();
4008 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4009 break;
4010 case vmIntrinsics::_getClassAccessFlags:
4011 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4012 return_type = TypeInt::CHAR;
4013 break;
4014 default:
4015 fatal_unexpected_iid(id);
4016 break;
4017 }
4018
4019 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4020 if (mirror_con == nullptr) return false; // cannot happen?
4021
4022 #ifndef PRODUCT
4023 if (C->print_intrinsics() || C->print_inlining()) {
4024 ciType* k = mirror_con->java_mirror_type();
4025 if (k) {
4026 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4027 k->print_name();
4028 tty->cr();
4029 }
4030 }
4031 #endif
4032
4033 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4034 RegionNode* region = new RegionNode(PATH_LIMIT);
4035 record_for_igvn(region);
4036 PhiNode* phi = new PhiNode(region, return_type);
4037
4038 // The mirror will never be null of Reflection.getClassAccessFlags, however
4039 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4040 // if it is. See bug 4774291.
4041
4042 // For Reflection.getClassAccessFlags(), the null check occurs in
4043 // the wrong place; see inline_unsafe_access(), above, for a similar
4044 // situation.
4045 mirror = null_check(mirror);
4046 // If mirror or obj is dead, only null-path is taken.
4047 if (stopped()) return true;
4048
4049 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4050
4051 // Now load the mirror's klass metaobject, and null-check it.
4052 // Side-effects region with the control path if the klass is null.
4053 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4054 // If kls is null, we have a primitive mirror.
4055 phi->init_req(_prim_path, prim_return_value);
4056 if (stopped()) { set_result(region, phi); return true; }
4057 bool safe_for_replace = (region->in(_prim_path) == top());
4058
4059 Node* p; // handy temp
4060 Node* null_ctl;
4061
4062 // Now that we have the non-null klass, we can perform the real query.
4063 // For constant classes, the query will constant-fold in LoadNode::Value.
4064 Node* query_value = top();
4065 switch (id) {
4066 case vmIntrinsics::_isInstance:
4067 // nothing is an instance of a primitive type
4068 query_value = gen_instanceof(obj, kls, safe_for_replace);
4069 break;
4070
4071 case vmIntrinsics::_isHidden:
4072 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4073 if (generate_hidden_class_guard(kls, region) != nullptr)
4074 // A guard was added. If the guard is taken, it was an hidden class.
4075 phi->add_req(intcon(1));
4076 // If we fall through, it's a plain class.
4077 query_value = intcon(0);
4078 break;
4079
4080
4081 case vmIntrinsics::_getSuperclass:
4082 // The rules here are somewhat unfortunate, but we can still do better
4083 // with random logic than with a JNI call.
4084 // Interfaces store null or Object as _super, but must report null.
4085 // Arrays store an intermediate super as _super, but must report Object.
4086 // Other types can report the actual _super.
4087 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4088 if (generate_interface_guard(kls, region) != nullptr)
4089 // A guard was added. If the guard is taken, it was an interface.
4090 phi->add_req(null());
4091 if (generate_array_guard(kls, region) != nullptr)
4092 // A guard was added. If the guard is taken, it was an array.
4093 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4094 // If we fall through, it's a plain class. Get its _super.
4095 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4096 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4097 null_ctl = top();
4098 kls = null_check_oop(kls, &null_ctl);
4099 if (null_ctl != top()) {
4100 // If the guard is taken, Object.superClass is null (both klass and mirror).
4101 region->add_req(null_ctl);
4102 phi ->add_req(null());
4103 }
4104 if (!stopped()) {
4105 query_value = load_mirror_from_klass(kls);
4106 }
4107 break;
4108
4109 case vmIntrinsics::_getClassAccessFlags:
4110 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4111 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4112 break;
4113
4114 default:
4115 fatal_unexpected_iid(id);
4116 break;
4117 }
4118
4119 // Fall-through is the normal case of a query to a real class.
4120 phi->init_req(1, query_value);
4121 region->init_req(1, control());
4122
4123 C->set_has_split_ifs(true); // Has chance for split-if optimization
4124 set_result(region, phi);
4125 return true;
4126 }
4127
4128 //-------------------------inline_Class_cast-------------------
4129 bool LibraryCallKit::inline_Class_cast() {
4130 Node* mirror = argument(0); // Class
4131 Node* obj = argument(1);
4132 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4133 if (mirror_con == nullptr) {
4134 return false; // dead path (mirror->is_top()).
4135 }
4136 if (obj == nullptr || obj->is_top()) {
4137 return false; // dead path
4138 }
4139 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4140
4141 // First, see if Class.cast() can be folded statically.
4142 // java_mirror_type() returns non-null for compile-time Class constants.
4143 ciType* tm = mirror_con->java_mirror_type();
4144 if (tm != nullptr && tm->is_klass() &&
4145 tp != nullptr) {
4146 if (!tp->is_loaded()) {
4147 // Don't use intrinsic when class is not loaded.
4148 return false;
4149 } else {
4150 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4151 if (static_res == Compile::SSC_always_true) {
4152 // isInstance() is true - fold the code.
4153 set_result(obj);
4154 return true;
4155 } else if (static_res == Compile::SSC_always_false) {
4156 // Don't use intrinsic, have to throw ClassCastException.
4157 // If the reference is null, the non-intrinsic bytecode will
4158 // be optimized appropriately.
4159 return false;
4160 }
4161 }
4162 }
4163
4164 // Bailout intrinsic and do normal inlining if exception path is frequent.
4165 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4166 return false;
4167 }
4168
4169 // Generate dynamic checks.
4170 // Class.cast() is java implementation of _checkcast bytecode.
4171 // Do checkcast (Parse::do_checkcast()) optimizations here.
4172
4173 mirror = null_check(mirror);
4174 // If mirror is dead, only null-path is taken.
4175 if (stopped()) {
4176 return true;
4177 }
4178
4179 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4180 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4181 RegionNode* region = new RegionNode(PATH_LIMIT);
4182 record_for_igvn(region);
4183
4184 // Now load the mirror's klass metaobject, and null-check it.
4185 // If kls is null, we have a primitive mirror and
4186 // nothing is an instance of a primitive type.
4187 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4188
4189 Node* res = top();
4190 if (!stopped()) {
4191 Node* bad_type_ctrl = top();
4192 // Do checkcast optimizations.
4193 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4194 region->init_req(_bad_type_path, bad_type_ctrl);
4195 }
4196 if (region->in(_prim_path) != top() ||
4197 region->in(_bad_type_path) != top()) {
4198 // Let Interpreter throw ClassCastException.
4199 PreserveJVMState pjvms(this);
4200 set_control(_gvn.transform(region));
4201 uncommon_trap(Deoptimization::Reason_intrinsic,
4202 Deoptimization::Action_maybe_recompile);
4203 }
4204 if (!stopped()) {
4205 set_result(res);
4206 }
4207 return true;
4208 }
4209
4210
4211 //--------------------------inline_native_subtype_check------------------------
4212 // This intrinsic takes the JNI calls out of the heart of
4213 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4214 bool LibraryCallKit::inline_native_subtype_check() {
4215 // Pull both arguments off the stack.
4216 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4217 args[0] = argument(0);
4218 args[1] = argument(1);
4219 Node* klasses[2]; // corresponding Klasses: superk, subk
4220 klasses[0] = klasses[1] = top();
4221
4222 enum {
4223 // A full decision tree on {superc is prim, subc is prim}:
4224 _prim_0_path = 1, // {P,N} => false
4225 // {P,P} & superc!=subc => false
4226 _prim_same_path, // {P,P} & superc==subc => true
4227 _prim_1_path, // {N,P} => false
4228 _ref_subtype_path, // {N,N} & subtype check wins => true
4229 _both_ref_path, // {N,N} & subtype check loses => false
4230 PATH_LIMIT
4231 };
4232
4233 RegionNode* region = new RegionNode(PATH_LIMIT);
4234 Node* phi = new PhiNode(region, TypeInt::BOOL);
4235 record_for_igvn(region);
4236
4237 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4238 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4239 int class_klass_offset = java_lang_Class::klass_offset();
4240
4241 // First null-check both mirrors and load each mirror's klass metaobject.
4242 int which_arg;
4243 for (which_arg = 0; which_arg <= 1; which_arg++) {
4244 Node* arg = args[which_arg];
4245 arg = null_check(arg);
4246 if (stopped()) break;
4247 args[which_arg] = arg;
4248
4249 Node* p = basic_plus_adr(arg, class_klass_offset);
4250 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4251 klasses[which_arg] = _gvn.transform(kls);
4252 }
4253
4254 // Having loaded both klasses, test each for null.
4255 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4256 for (which_arg = 0; which_arg <= 1; which_arg++) {
4257 Node* kls = klasses[which_arg];
4258 Node* null_ctl = top();
4259 kls = null_check_oop(kls, &null_ctl, never_see_null);
4260 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4261 region->init_req(prim_path, null_ctl);
4262 if (stopped()) break;
4263 klasses[which_arg] = kls;
4264 }
4265
4266 if (!stopped()) {
4267 // now we have two reference types, in klasses[0..1]
4268 Node* subk = klasses[1]; // the argument to isAssignableFrom
4269 Node* superk = klasses[0]; // the receiver
4270 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4271 // now we have a successful reference subtype check
4272 region->set_req(_ref_subtype_path, control());
4273 }
4274
4275 // If both operands are primitive (both klasses null), then
4276 // we must return true when they are identical primitives.
4277 // It is convenient to test this after the first null klass check.
4278 set_control(region->in(_prim_0_path)); // go back to first null check
4279 if (!stopped()) {
4280 // Since superc is primitive, make a guard for the superc==subc case.
4281 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4282 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4283 generate_guard(bol_eq, region, PROB_FAIR);
4284 if (region->req() == PATH_LIMIT+1) {
4285 // A guard was added. If the added guard is taken, superc==subc.
4286 region->swap_edges(PATH_LIMIT, _prim_same_path);
4287 region->del_req(PATH_LIMIT);
4288 }
4289 region->set_req(_prim_0_path, control()); // Not equal after all.
4290 }
4291
4292 // these are the only paths that produce 'true':
4293 phi->set_req(_prim_same_path, intcon(1));
4294 phi->set_req(_ref_subtype_path, intcon(1));
4295
4296 // pull together the cases:
4297 assert(region->req() == PATH_LIMIT, "sane region");
4298 for (uint i = 1; i < region->req(); i++) {
4299 Node* ctl = region->in(i);
4300 if (ctl == nullptr || ctl == top()) {
4301 region->set_req(i, top());
4302 phi ->set_req(i, top());
4303 } else if (phi->in(i) == nullptr) {
4304 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4305 }
4306 }
4307
4308 set_control(_gvn.transform(region));
4309 set_result(_gvn.transform(phi));
4310 return true;
4311 }
4312
4313 //---------------------generate_array_guard_common------------------------
4314 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4315 bool obj_array, bool not_array, Node** obj) {
4316
4317 if (stopped()) {
4318 return nullptr;
4319 }
4320
4321 // If obj_array/non_array==false/false:
4322 // Branch around if the given klass is in fact an array (either obj or prim).
4323 // If obj_array/non_array==false/true:
4324 // Branch around if the given klass is not an array klass of any kind.
4325 // If obj_array/non_array==true/true:
4326 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4327 // If obj_array/non_array==true/false:
4328 // Branch around if the kls is an oop array (Object[] or subtype)
4329 //
4330 // Like generate_guard, adds a new path onto the region.
4331 jint layout_con = 0;
4332 Node* layout_val = get_layout_helper(kls, layout_con);
4333 if (layout_val == nullptr) {
4334 bool query = (obj_array
4335 ? Klass::layout_helper_is_objArray(layout_con)
4336 : Klass::layout_helper_is_array(layout_con));
4337 if (query == not_array) {
4338 return nullptr; // never a branch
4339 } else { // always a branch
4340 Node* always_branch = control();
4341 if (region != nullptr)
4342 region->add_req(always_branch);
4343 set_control(top());
4344 return always_branch;
4345 }
4346 }
4347 // Now test the correct condition.
4348 jint nval = (obj_array
4349 ? (jint)(Klass::_lh_array_tag_type_value
4350 << Klass::_lh_array_tag_shift)
4351 : Klass::_lh_neutral_value);
4352 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4353 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4354 // invert the test if we are looking for a non-array
4355 if (not_array) btest = BoolTest(btest).negate();
4356 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4357 Node* ctrl = generate_fair_guard(bol, region);
4358 Node* is_array_ctrl = not_array ? control() : ctrl;
4359 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4360 // Keep track of the fact that 'obj' is an array to prevent
4361 // array specific accesses from floating above the guard.
4362 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4363 }
4364 return ctrl;
4365 }
4366
4367
4368 //-----------------------inline_native_newArray--------------------------
4369 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4370 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4371 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4372 Node* mirror;
4373 Node* count_val;
4374 if (uninitialized) {
4375 null_check_receiver();
4376 mirror = argument(1);
4377 count_val = argument(2);
4378 } else {
4379 mirror = argument(0);
4380 count_val = argument(1);
4381 }
4382
4383 mirror = null_check(mirror);
4384 // If mirror or obj is dead, only null-path is taken.
4385 if (stopped()) return true;
4386
4387 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4388 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4389 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4390 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4391 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4392
4393 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4394 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4395 result_reg, _slow_path);
4396 Node* normal_ctl = control();
4397 Node* no_array_ctl = result_reg->in(_slow_path);
4398
4399 // Generate code for the slow case. We make a call to newArray().
4400 set_control(no_array_ctl);
4401 if (!stopped()) {
4402 // Either the input type is void.class, or else the
4403 // array klass has not yet been cached. Either the
4404 // ensuing call will throw an exception, or else it
4405 // will cache the array klass for next time.
4406 PreserveJVMState pjvms(this);
4407 CallJavaNode* slow_call = nullptr;
4408 if (uninitialized) {
4409 // Generate optimized virtual call (holder class 'Unsafe' is final)
4410 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4411 } else {
4412 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4413 }
4414 Node* slow_result = set_results_for_java_call(slow_call);
4415 // this->control() comes from set_results_for_java_call
4416 result_reg->set_req(_slow_path, control());
4417 result_val->set_req(_slow_path, slow_result);
4418 result_io ->set_req(_slow_path, i_o());
4419 result_mem->set_req(_slow_path, reset_memory());
4420 }
4421
4422 set_control(normal_ctl);
4423 if (!stopped()) {
4424 // Normal case: The array type has been cached in the java.lang.Class.
4425 // The following call works fine even if the array type is polymorphic.
4426 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4427 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4428 result_reg->init_req(_normal_path, control());
4429 result_val->init_req(_normal_path, obj);
4430 result_io ->init_req(_normal_path, i_o());
4431 result_mem->init_req(_normal_path, reset_memory());
4432
4433 if (uninitialized) {
4434 // Mark the allocation so that zeroing is skipped
4435 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4436 alloc->maybe_set_complete(&_gvn);
4437 }
4438 }
4439
4440 // Return the combined state.
4441 set_i_o( _gvn.transform(result_io) );
4442 set_all_memory( _gvn.transform(result_mem));
4443
4444 C->set_has_split_ifs(true); // Has chance for split-if optimization
4445 set_result(result_reg, result_val);
4446 return true;
4447 }
4448
4449 //----------------------inline_native_getLength--------------------------
4450 // public static native int java.lang.reflect.Array.getLength(Object array);
4451 bool LibraryCallKit::inline_native_getLength() {
4452 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4453
4454 Node* array = null_check(argument(0));
4455 // If array is dead, only null-path is taken.
4456 if (stopped()) return true;
4457
4458 // Deoptimize if it is a non-array.
4459 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4460
4461 if (non_array != nullptr) {
4462 PreserveJVMState pjvms(this);
4463 set_control(non_array);
4464 uncommon_trap(Deoptimization::Reason_intrinsic,
4465 Deoptimization::Action_maybe_recompile);
4466 }
4467
4468 // If control is dead, only non-array-path is taken.
4469 if (stopped()) return true;
4470
4471 // The works fine even if the array type is polymorphic.
4472 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4473 Node* result = load_array_length(array);
4474
4475 C->set_has_split_ifs(true); // Has chance for split-if optimization
4476 set_result(result);
4477 return true;
4478 }
4479
4480 //------------------------inline_array_copyOf----------------------------
4481 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4482 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4483 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4484 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4485
4486 // Get the arguments.
4487 Node* original = argument(0);
4488 Node* start = is_copyOfRange? argument(1): intcon(0);
4489 Node* end = is_copyOfRange? argument(2): argument(1);
4490 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4491
4492 Node* newcopy = nullptr;
4493
4494 // Set the original stack and the reexecute bit for the interpreter to reexecute
4495 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4496 { PreserveReexecuteState preexecs(this);
4497 jvms()->set_should_reexecute(true);
4498
4499 array_type_mirror = null_check(array_type_mirror);
4500 original = null_check(original);
4501
4502 // Check if a null path was taken unconditionally.
4503 if (stopped()) return true;
4504
4505 Node* orig_length = load_array_length(original);
4506
4507 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4508 klass_node = null_check(klass_node);
4509
4510 RegionNode* bailout = new RegionNode(1);
4511 record_for_igvn(bailout);
4512
4513 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4514 // Bail out if that is so.
4515 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4516 if (not_objArray != nullptr) {
4517 // Improve the klass node's type from the new optimistic assumption:
4518 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4519 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4520 Node* cast = new CastPPNode(control(), klass_node, akls);
4521 klass_node = _gvn.transform(cast);
4522 }
4523
4524 // Bail out if either start or end is negative.
4525 generate_negative_guard(start, bailout, &start);
4526 generate_negative_guard(end, bailout, &end);
4527
4528 Node* length = end;
4529 if (_gvn.type(start) != TypeInt::ZERO) {
4530 length = _gvn.transform(new SubINode(end, start));
4531 }
4532
4533 // Bail out if length is negative (i.e., if start > end).
4534 // Without this the new_array would throw
4535 // NegativeArraySizeException but IllegalArgumentException is what
4536 // should be thrown
4537 generate_negative_guard(length, bailout, &length);
4538
4539 // Bail out if start is larger than the original length
4540 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4541 generate_negative_guard(orig_tail, bailout, &orig_tail);
4542
4543 if (bailout->req() > 1) {
4544 PreserveJVMState pjvms(this);
4545 set_control(_gvn.transform(bailout));
4546 uncommon_trap(Deoptimization::Reason_intrinsic,
4547 Deoptimization::Action_maybe_recompile);
4548 }
4549
4550 if (!stopped()) {
4551 // How many elements will we copy from the original?
4552 // The answer is MinI(orig_tail, length).
4553 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4554
4555 // Generate a direct call to the right arraycopy function(s).
4556 // We know the copy is disjoint but we might not know if the
4557 // oop stores need checking.
4558 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4559 // This will fail a store-check if x contains any non-nulls.
4560
4561 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4562 // loads/stores but it is legal only if we're sure the
4563 // Arrays.copyOf would succeed. So we need all input arguments
4564 // to the copyOf to be validated, including that the copy to the
4565 // new array won't trigger an ArrayStoreException. That subtype
4566 // check can be optimized if we know something on the type of
4567 // the input array from type speculation.
4568 if (_gvn.type(klass_node)->singleton()) {
4569 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4570 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4571
4572 int test = C->static_subtype_check(superk, subk);
4573 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4574 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4575 if (t_original->speculative_type() != nullptr) {
4576 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4577 }
4578 }
4579 }
4580
4581 bool validated = false;
4582 // Reason_class_check rather than Reason_intrinsic because we
4583 // want to intrinsify even if this traps.
4584 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4585 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4586
4587 if (not_subtype_ctrl != top()) {
4588 PreserveJVMState pjvms(this);
4589 set_control(not_subtype_ctrl);
4590 uncommon_trap(Deoptimization::Reason_class_check,
4591 Deoptimization::Action_make_not_entrant);
4592 assert(stopped(), "Should be stopped");
4593 }
4594 validated = true;
4595 }
4596
4597 if (!stopped()) {
4598 newcopy = new_array(klass_node, length, 0); // no arguments to push
4599
4600 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4601 load_object_klass(original), klass_node);
4602 if (!is_copyOfRange) {
4603 ac->set_copyof(validated);
4604 } else {
4605 ac->set_copyofrange(validated);
4606 }
4607 Node* n = _gvn.transform(ac);
4608 if (n == ac) {
4609 ac->connect_outputs(this);
4610 } else {
4611 assert(validated, "shouldn't transform if all arguments not validated");
4612 set_all_memory(n);
4613 }
4614 }
4615 }
4616 } // original reexecute is set back here
4617
4618 C->set_has_split_ifs(true); // Has chance for split-if optimization
4619 if (!stopped()) {
4620 set_result(newcopy);
4621 }
4622 return true;
4623 }
4624
4625
4626 //----------------------generate_virtual_guard---------------------------
4627 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4628 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4629 RegionNode* slow_region) {
4630 ciMethod* method = callee();
4631 int vtable_index = method->vtable_index();
4632 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4633 "bad index %d", vtable_index);
4634 // Get the Method* out of the appropriate vtable entry.
4635 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4636 vtable_index*vtableEntry::size_in_bytes() +
4637 in_bytes(vtableEntry::method_offset());
4638 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4639 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4640
4641 // Compare the target method with the expected method (e.g., Object.hashCode).
4642 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4643
4644 Node* native_call = makecon(native_call_addr);
4645 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4646 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4647
4648 return generate_slow_guard(test_native, slow_region);
4649 }
4650
4651 //-----------------------generate_method_call----------------------------
4652 // Use generate_method_call to make a slow-call to the real
4653 // method if the fast path fails. An alternative would be to
4654 // use a stub like OptoRuntime::slow_arraycopy_Java.
4655 // This only works for expanding the current library call,
4656 // not another intrinsic. (E.g., don't use this for making an
4657 // arraycopy call inside of the copyOf intrinsic.)
4658 CallJavaNode*
4659 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4660 // When compiling the intrinsic method itself, do not use this technique.
4661 guarantee(callee() != C->method(), "cannot make slow-call to self");
4662
4663 ciMethod* method = callee();
4664 // ensure the JVMS we have will be correct for this call
4665 guarantee(method_id == method->intrinsic_id(), "must match");
4666
4667 const TypeFunc* tf = TypeFunc::make(method);
4668 if (res_not_null) {
4669 assert(tf->return_type() == T_OBJECT, "");
4670 const TypeTuple* range = tf->range();
4671 const Type** fields = TypeTuple::fields(range->cnt());
4672 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4673 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4674 tf = TypeFunc::make(tf->domain(), new_range);
4675 }
4676 CallJavaNode* slow_call;
4677 if (is_static) {
4678 assert(!is_virtual, "");
4679 slow_call = new CallStaticJavaNode(C, tf,
4680 SharedRuntime::get_resolve_static_call_stub(), method);
4681 } else if (is_virtual) {
4682 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4683 int vtable_index = Method::invalid_vtable_index;
4684 if (UseInlineCaches) {
4685 // Suppress the vtable call
4686 } else {
4687 // hashCode and clone are not a miranda methods,
4688 // so the vtable index is fixed.
4689 // No need to use the linkResolver to get it.
4690 vtable_index = method->vtable_index();
4691 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4692 "bad index %d", vtable_index);
4693 }
4694 slow_call = new CallDynamicJavaNode(tf,
4695 SharedRuntime::get_resolve_virtual_call_stub(),
4696 method, vtable_index);
4697 } else { // neither virtual nor static: opt_virtual
4698 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4699 slow_call = new CallStaticJavaNode(C, tf,
4700 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4701 slow_call->set_optimized_virtual(true);
4702 }
4703 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4704 // To be able to issue a direct call (optimized virtual or virtual)
4705 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4706 // about the method being invoked should be attached to the call site to
4707 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4708 slow_call->set_override_symbolic_info(true);
4709 }
4710 set_arguments_for_java_call(slow_call);
4711 set_edges_for_java_call(slow_call);
4712 return slow_call;
4713 }
4714
4715
4716 /**
4717 * Build special case code for calls to hashCode on an object. This call may
4718 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4719 * slightly different code.
4720 */
4721 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4722 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4723 assert(!(is_virtual && is_static), "either virtual, special, or static");
4724
4725 enum { _slow_path = 1, _null_path, _fast_path, _fast_path2, PATH_LIMIT };
4726
4727 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4728 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4729 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4730 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4731 Node* obj = nullptr;
4732 if (!is_static) {
4733 // Check for hashing null object
4734 obj = null_check_receiver();
4735 if (stopped()) return true; // unconditionally null
4736 result_reg->init_req(_null_path, top());
4737 result_val->init_req(_null_path, top());
4738 } else {
4739 // Do a null check, and return zero if null.
4740 // System.identityHashCode(null) == 0
4741 obj = argument(0);
4742 Node* null_ctl = top();
4743 obj = null_check_oop(obj, &null_ctl);
4744 result_reg->init_req(_null_path, null_ctl);
4745 result_val->init_req(_null_path, _gvn.intcon(0));
4746 }
4747
4748 // Unconditionally null? Then return right away.
4749 if (stopped()) {
4750 set_control( result_reg->in(_null_path));
4751 if (!stopped())
4752 set_result(result_val->in(_null_path));
4753 return true;
4754 }
4755
4756 // We only go to the fast case code if we pass a number of guards. The
4757 // paths which do not pass are accumulated in the slow_region.
4758 RegionNode* slow_region = new RegionNode(1);
4759 record_for_igvn(slow_region);
4760
4761 // If this is a virtual call, we generate a funny guard. We pull out
4762 // the vtable entry corresponding to hashCode() from the target object.
4763 // If the target method which we are calling happens to be the native
4764 // Object hashCode() method, we pass the guard. We do not need this
4765 // guard for non-virtual calls -- the caller is known to be the native
4766 // Object hashCode().
4767 if (is_virtual) {
4768 // After null check, get the object's klass.
4769 Node* obj_klass = load_object_klass(obj);
4770 generate_virtual_guard(obj_klass, slow_region);
4771 }
4772
4773 if (UseCompactObjectHeaders) {
4774 // Get the header out of the object.
4775 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4776 // The control of the load must be null. Otherwise, the load can move before
4777 // the null check after castPP removal.
4778 Node* no_ctrl = nullptr;
4779 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4780
4781 // Test the header to see if the object is in hashed or copied state.
4782 Node* hashctrl_mask = _gvn.MakeConX(markWord::hashctrl_mask_in_place);
4783 Node* masked_header = _gvn.transform(new AndXNode(header, hashctrl_mask));
4784
4785 // Take slow-path when the object has not been hashed.
4786 Node* not_hashed_val = _gvn.MakeConX(0);
4787 Node* chk_hashed = _gvn.transform(new CmpXNode(masked_header, not_hashed_val));
4788 Node* test_hashed = _gvn.transform(new BoolNode(chk_hashed, BoolTest::eq));
4789
4790 generate_slow_guard(test_hashed, slow_region);
4791
4792 // Test whether the object is hashed or hashed&copied.
4793 Node* hashed_copied = _gvn.MakeConX(markWord::hashctrl_expanded_mask_in_place | markWord::hashctrl_hashed_mask_in_place);
4794 Node* chk_copied = _gvn.transform(new CmpXNode(masked_header, hashed_copied));
4795 // If true, then object has been hashed&copied, otherwise it's only hashed.
4796 Node* test_copied = _gvn.transform(new BoolNode(chk_copied, BoolTest::eq));
4797 IfNode* if_copied = create_and_map_if(control(), test_copied, PROB_FAIR, COUNT_UNKNOWN);
4798 Node* if_true = _gvn.transform(new IfTrueNode(if_copied));
4799 Node* if_false = _gvn.transform(new IfFalseNode(if_copied));
4800
4801 // Hashed&Copied path: read hash-code out of the object.
4802 set_control(if_true);
4803 // result_val->del_req(_fast_path2);
4804 // result_reg->del_req(_fast_path2);
4805 // result_io->del_req(_fast_path2);
4806 // result_mem->del_req(_fast_path2);
4807
4808 Node* obj_klass = load_object_klass(obj);
4809 Node* hash_addr;
4810 const TypeKlassPtr* klass_t = _gvn.type(obj_klass)->isa_klassptr();
4811 bool load_offset_runtime = true;
4812
4813 if (klass_t != nullptr) {
4814 if (klass_t->klass_is_exact() && klass_t->isa_instklassptr()) {
4815 ciInstanceKlass* ciKlass = reinterpret_cast<ciInstanceKlass*>(klass_t->is_instklassptr()->exact_klass());
4816 if (!ciKlass->is_mirror_instance_klass() && !ciKlass->is_reference_instance_klass()) {
4817 // We know the InstanceKlass, load hash_offset from there at compile-time.
4818 int hash_offset = ciKlass->hash_offset_in_bytes();
4819 hash_addr = basic_plus_adr(obj, hash_offset);
4820 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4821 result_val->init_req(_fast_path2, loaded_hash);
4822 result_reg->init_req(_fast_path2, control());
4823 load_offset_runtime = false;
4824 }
4825 }
4826 }
4827
4828 //tty->print_cr("Load hash-offset at runtime: %s", BOOL_TO_STR(load_offset_runtime));
4829
4830 if (load_offset_runtime) {
4831 // We don't know if it is an array or an exact type, figure it out at run-time.
4832 // If not an ordinary instance, then we need to take slow-path.
4833 Node* kind_addr = basic_plus_adr(obj_klass, Klass::kind_offset_in_bytes());
4834 Node* kind = make_load(control(), kind_addr, TypeInt::INT, T_INT, MemNode::unordered);
4835 Node* instance_val = _gvn.intcon(Klass::InstanceKlassKind);
4836 Node* chk_inst = _gvn.transform(new CmpINode(kind, instance_val));
4837 Node* test_inst = _gvn.transform(new BoolNode(chk_inst, BoolTest::ne));
4838 generate_slow_guard(test_inst, slow_region);
4839
4840 // Otherwise it's an instance and we can read the hash_offset from the InstanceKlass.
4841 Node* hash_offset_addr = basic_plus_adr(obj_klass, InstanceKlass::hash_offset_offset_in_bytes());
4842 Node* hash_offset = make_load(control(), hash_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
4843 // hash_offset->dump();
4844 Node* hash_addr = basic_plus_adr(obj, ConvI2X(hash_offset));
4845 Compile::current()->set_has_unsafe_access(true);
4846 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4847 result_val->init_req(_fast_path2, loaded_hash);
4848 result_reg->init_req(_fast_path2, control());
4849 }
4850
4851 // Hashed-only path: recompute hash-code from object address.
4852 set_control(if_false);
4853 if (hashCode == 6) {
4854 // Our constants.
4855 Node* M = _gvn.intcon(0x337954D5);
4856 Node* A = _gvn.intcon(0xAAAAAAAA);
4857 // Split object address into lo and hi 32 bits.
4858 Node* obj_addr = _gvn.transform(new CastP2XNode(nullptr, obj));
4859 Node* x = _gvn.transform(new ConvL2INode(obj_addr));
4860 Node* upper_addr = _gvn.transform(new URShiftLNode(obj_addr, _gvn.intcon(32)));
4861 Node* y = _gvn.transform(new ConvL2INode(upper_addr));
4862
4863 Node* H0 = _gvn.transform(new XorINode(x, y));
4864 Node* L0 = _gvn.transform(new XorINode(x, A));
4865
4866 // Full multiplication of two 32 bit values L0 and M into a hi/lo result in two 32 bit values V0 and U0.
4867 Node* L0_64 = _gvn.transform(new ConvI2LNode(L0));
4868 L0_64 = _gvn.transform(new AndLNode(L0_64, _gvn.longcon(0xFFFFFFFF)));
4869 Node* M_64 = _gvn.transform(new ConvI2LNode(M));
4870 // M_64 = _gvn.transform(new AndLNode(M_64, _gvn.longcon(0xFFFFFFFF)));
4871 Node* prod64 = _gvn.transform(new MulLNode(L0_64, M_64));
4872 Node* V0 = _gvn.transform(new ConvL2INode(prod64));
4873 Node* prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4874 Node* U0 = _gvn.transform(new ConvL2INode(prod_upper));
4875
4876 Node* Q0 = _gvn.transform(new MulINode(H0, M));
4877 Node* L1 = _gvn.transform(new XorINode(Q0, U0));
4878
4879 // Full multiplication of two 32 bit values L1 and M into a hi/lo result in two 32 bit values V1 and U1.
4880 Node* L1_64 = _gvn.transform(new ConvI2LNode(L1));
4881 L1_64 = _gvn.transform(new AndLNode(L1_64, _gvn.longcon(0xFFFFFFFF)));
4882 prod64 = _gvn.transform(new MulLNode(L1_64, M_64));
4883 Node* V1 = _gvn.transform(new ConvL2INode(prod64));
4884 prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4885 Node* U1 = _gvn.transform(new ConvL2INode(prod_upper));
4886
4887 Node* P1 = _gvn.transform(new XorINode(V0, M));
4888
4889 // Right rotate P1 by distance L1.
4890 Node* distance = _gvn.transform(new AndINode(L1, _gvn.intcon(32 - 1)));
4891 Node* inverse_distance = _gvn.transform(new SubINode(_gvn.intcon(32), distance));
4892 Node* ror_part1 = _gvn.transform(new URShiftINode(P1, distance));
4893 Node* ror_part2 = _gvn.transform(new LShiftINode(P1, inverse_distance));
4894 Node* Q1 = _gvn.transform(new OrINode(ror_part1, ror_part2));
4895
4896 Node* L2 = _gvn.transform(new XorINode(Q1, U1));
4897 Node* hash = _gvn.transform(new XorINode(V1, L2));
4898 Node* hash_truncated = _gvn.transform(new AndINode(hash, _gvn.intcon(markWord::hash_mask)));
4899
4900 result_val->init_req(_fast_path, hash_truncated);
4901 } else if (hashCode == 2) {
4902 result_val->init_req(_fast_path, _gvn.intcon(1));
4903 }
4904 } else {
4905 // Get the header out of the object, use LoadMarkNode when available
4906 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4907 // The control of the load must be null. Otherwise, the load can move before
4908 // the null check after castPP removal.
4909 Node* no_ctrl = nullptr;
4910 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4911
4912 if (!UseObjectMonitorTable) {
4913 // Test the header to see if it is safe to read w.r.t. locking.
4914 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4915 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4916 if (LockingMode == LM_LIGHTWEIGHT) {
4917 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4918 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4919 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4920
4921 generate_slow_guard(test_monitor, slow_region);
4922 } else {
4923 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4924 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4925 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4926
4927 generate_slow_guard(test_not_unlocked, slow_region);
4928 }
4929 }
4930
4931 // Get the hash value and check to see that it has been properly assigned.
4932 // We depend on hash_mask being at most 32 bits and avoid the use of
4933 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4934 // vm: see markWord.hpp.
4935 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4936 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4937 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4938 // This hack lets the hash bits live anywhere in the mark object now, as long
4939 // as the shift drops the relevant bits into the low 32 bits. Note that
4940 // Java spec says that HashCode is an int so there's no point in capturing
4941 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4942 hshifted_header = ConvX2I(hshifted_header);
4943 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4944
4945 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4946 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4947 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4948
4949 generate_slow_guard(test_assigned, slow_region);
4950
4951 result_val->init_req(_fast_path, hash_val);
4952
4953 // _fast_path2 is not used here.
4954 result_val->del_req(_fast_path2);
4955 result_reg->del_req(_fast_path2);
4956 result_io->del_req(_fast_path2);
4957 result_mem->del_req(_fast_path2);
4958 }
4959
4960 Node* init_mem = reset_memory();
4961 // fill in the rest of the null path:
4962 result_io ->init_req(_null_path, i_o());
4963 result_mem->init_req(_null_path, init_mem);
4964
4965 result_reg->init_req(_fast_path, control());
4966 result_io ->init_req(_fast_path, i_o());
4967 result_mem->init_req(_fast_path, init_mem);
4968
4969 if (UseCompactObjectHeaders) {
4970 result_io->init_req(_fast_path2, i_o());
4971 result_mem->init_req(_fast_path2, init_mem);
4972 }
4973
4974 // Generate code for the slow case. We make a call to hashCode().
4975 assert(slow_region != nullptr, "must have slow_region");
4976 set_control(_gvn.transform(slow_region));
4977 if (!stopped()) {
4978 // No need for PreserveJVMState, because we're using up the present state.
4979 set_all_memory(init_mem);
4980 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4981 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4982 Node* slow_result = set_results_for_java_call(slow_call);
4983 // this->control() comes from set_results_for_java_call
4984 result_reg->init_req(_slow_path, control());
4985 result_val->init_req(_slow_path, slow_result);
4986 result_io ->set_req(_slow_path, i_o());
4987 result_mem ->set_req(_slow_path, reset_memory());
4988 }
4989
4990 // Return the combined state.
4991 set_i_o( _gvn.transform(result_io) );
4992 set_all_memory( _gvn.transform(result_mem));
4993
4994 set_result(result_reg, result_val);
4995 return true;
4996 }
4997
4998 //---------------------------inline_native_getClass----------------------------
4999 // public final native Class<?> java.lang.Object.getClass();
5000 //
5001 // Build special case code for calls to getClass on an object.
5002 bool LibraryCallKit::inline_native_getClass() {
5003 Node* obj = null_check_receiver();
5004 if (stopped()) return true;
5005 set_result(load_mirror_from_klass(load_object_klass(obj)));
5006 return true;
5007 }
5008
5009 //-----------------inline_native_Reflection_getCallerClass---------------------
5010 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5011 //
5012 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5013 //
5014 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5015 // in that it must skip particular security frames and checks for
5016 // caller sensitive methods.
5017 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5018 #ifndef PRODUCT
5019 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5020 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5021 }
5022 #endif
5023
5024 if (!jvms()->has_method()) {
5025 #ifndef PRODUCT
5026 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5027 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5028 }
5029 #endif
5030 return false;
5031 }
5032
5033 // Walk back up the JVM state to find the caller at the required
5034 // depth.
5035 JVMState* caller_jvms = jvms();
5036
5037 // Cf. JVM_GetCallerClass
5038 // NOTE: Start the loop at depth 1 because the current JVM state does
5039 // not include the Reflection.getCallerClass() frame.
5040 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5041 ciMethod* m = caller_jvms->method();
5042 switch (n) {
5043 case 0:
5044 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5045 break;
5046 case 1:
5047 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5048 if (!m->caller_sensitive()) {
5049 #ifndef PRODUCT
5050 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5051 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5052 }
5053 #endif
5054 return false; // bail-out; let JVM_GetCallerClass do the work
5055 }
5056 break;
5057 default:
5058 if (!m->is_ignored_by_security_stack_walk()) {
5059 // We have reached the desired frame; return the holder class.
5060 // Acquire method holder as java.lang.Class and push as constant.
5061 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5062 ciInstance* caller_mirror = caller_klass->java_mirror();
5063 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5064
5065 #ifndef PRODUCT
5066 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5067 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());
5068 tty->print_cr(" JVM state at this point:");
5069 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5070 ciMethod* m = jvms()->of_depth(i)->method();
5071 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5072 }
5073 }
5074 #endif
5075 return true;
5076 }
5077 break;
5078 }
5079 }
5080
5081 #ifndef PRODUCT
5082 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5083 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5084 tty->print_cr(" JVM state at this point:");
5085 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5086 ciMethod* m = jvms()->of_depth(i)->method();
5087 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5088 }
5089 }
5090 #endif
5091
5092 return false; // bail-out; let JVM_GetCallerClass do the work
5093 }
5094
5095 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5096 Node* arg = argument(0);
5097 Node* result = nullptr;
5098
5099 switch (id) {
5100 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5101 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5102 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5103 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5104 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5105 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5106
5107 case vmIntrinsics::_doubleToLongBits: {
5108 // two paths (plus control) merge in a wood
5109 RegionNode *r = new RegionNode(3);
5110 Node *phi = new PhiNode(r, TypeLong::LONG);
5111
5112 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5113 // Build the boolean node
5114 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5115
5116 // Branch either way.
5117 // NaN case is less traveled, which makes all the difference.
5118 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5119 Node *opt_isnan = _gvn.transform(ifisnan);
5120 assert( opt_isnan->is_If(), "Expect an IfNode");
5121 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5122 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5123
5124 set_control(iftrue);
5125
5126 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5127 Node *slow_result = longcon(nan_bits); // return NaN
5128 phi->init_req(1, _gvn.transform( slow_result ));
5129 r->init_req(1, iftrue);
5130
5131 // Else fall through
5132 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5133 set_control(iffalse);
5134
5135 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5136 r->init_req(2, iffalse);
5137
5138 // Post merge
5139 set_control(_gvn.transform(r));
5140 record_for_igvn(r);
5141
5142 C->set_has_split_ifs(true); // Has chance for split-if optimization
5143 result = phi;
5144 assert(result->bottom_type()->isa_long(), "must be");
5145 break;
5146 }
5147
5148 case vmIntrinsics::_floatToIntBits: {
5149 // two paths (plus control) merge in a wood
5150 RegionNode *r = new RegionNode(3);
5151 Node *phi = new PhiNode(r, TypeInt::INT);
5152
5153 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5154 // Build the boolean node
5155 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5156
5157 // Branch either way.
5158 // NaN case is less traveled, which makes all the difference.
5159 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5160 Node *opt_isnan = _gvn.transform(ifisnan);
5161 assert( opt_isnan->is_If(), "Expect an IfNode");
5162 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5163 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5164
5165 set_control(iftrue);
5166
5167 static const jint nan_bits = 0x7fc00000;
5168 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5169 phi->init_req(1, _gvn.transform( slow_result ));
5170 r->init_req(1, iftrue);
5171
5172 // Else fall through
5173 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5174 set_control(iffalse);
5175
5176 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5177 r->init_req(2, iffalse);
5178
5179 // Post merge
5180 set_control(_gvn.transform(r));
5181 record_for_igvn(r);
5182
5183 C->set_has_split_ifs(true); // Has chance for split-if optimization
5184 result = phi;
5185 assert(result->bottom_type()->isa_int(), "must be");
5186 break;
5187 }
5188
5189 default:
5190 fatal_unexpected_iid(id);
5191 break;
5192 }
5193 set_result(_gvn.transform(result));
5194 return true;
5195 }
5196
5197 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5198 Node* arg = argument(0);
5199 Node* result = nullptr;
5200
5201 switch (id) {
5202 case vmIntrinsics::_floatIsInfinite:
5203 result = new IsInfiniteFNode(arg);
5204 break;
5205 case vmIntrinsics::_floatIsFinite:
5206 result = new IsFiniteFNode(arg);
5207 break;
5208 case vmIntrinsics::_doubleIsInfinite:
5209 result = new IsInfiniteDNode(arg);
5210 break;
5211 case vmIntrinsics::_doubleIsFinite:
5212 result = new IsFiniteDNode(arg);
5213 break;
5214 default:
5215 fatal_unexpected_iid(id);
5216 break;
5217 }
5218 set_result(_gvn.transform(result));
5219 return true;
5220 }
5221
5222 //----------------------inline_unsafe_copyMemory-------------------------
5223 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5224
5225 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5226 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5227 const Type* base_t = gvn.type(base);
5228
5229 bool in_native = (base_t == TypePtr::NULL_PTR);
5230 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5231 bool is_mixed = !in_heap && !in_native;
5232
5233 if (is_mixed) {
5234 return true; // mixed accesses can touch both on-heap and off-heap memory
5235 }
5236 if (in_heap) {
5237 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5238 if (!is_prim_array) {
5239 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5240 // there's not enough type information available to determine proper memory slice for it.
5241 return true;
5242 }
5243 }
5244 return false;
5245 }
5246
5247 bool LibraryCallKit::inline_unsafe_copyMemory() {
5248 if (callee()->is_static()) return false; // caller must have the capability!
5249 null_check_receiver(); // null-check receiver
5250 if (stopped()) return true;
5251
5252 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5253
5254 Node* src_base = argument(1); // type: oop
5255 Node* src_off = ConvL2X(argument(2)); // type: long
5256 Node* dst_base = argument(4); // type: oop
5257 Node* dst_off = ConvL2X(argument(5)); // type: long
5258 Node* size = ConvL2X(argument(7)); // type: long
5259
5260 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5261 "fieldOffset must be byte-scaled");
5262
5263 Node* src_addr = make_unsafe_address(src_base, src_off);
5264 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5265
5266 Node* thread = _gvn.transform(new ThreadLocalNode());
5267 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5268 BasicType doing_unsafe_access_bt = T_BYTE;
5269 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5270
5271 // update volatile field
5272 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5273
5274 int flags = RC_LEAF | RC_NO_FP;
5275
5276 const TypePtr* dst_type = TypePtr::BOTTOM;
5277
5278 // Adjust memory effects of the runtime call based on input values.
5279 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5280 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5281 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5282
5283 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5284 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5285 flags |= RC_NARROW_MEM; // narrow in memory
5286 }
5287 }
5288
5289 // Call it. Note that the length argument is not scaled.
5290 make_runtime_call(flags,
5291 OptoRuntime::fast_arraycopy_Type(),
5292 StubRoutines::unsafe_arraycopy(),
5293 "unsafe_arraycopy",
5294 dst_type,
5295 src_addr, dst_addr, size XTOP);
5296
5297 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5298
5299 return true;
5300 }
5301
5302 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5303 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5304 bool LibraryCallKit::inline_unsafe_setMemory() {
5305 if (callee()->is_static()) return false; // caller must have the capability!
5306 null_check_receiver(); // null-check receiver
5307 if (stopped()) return true;
5308
5309 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5310
5311 Node* dst_base = argument(1); // type: oop
5312 Node* dst_off = ConvL2X(argument(2)); // type: long
5313 Node* size = ConvL2X(argument(4)); // type: long
5314 Node* byte = argument(6); // type: byte
5315
5316 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5317 "fieldOffset must be byte-scaled");
5318
5319 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5320
5321 Node* thread = _gvn.transform(new ThreadLocalNode());
5322 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5323 BasicType doing_unsafe_access_bt = T_BYTE;
5324 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5325
5326 // update volatile field
5327 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5328
5329 int flags = RC_LEAF | RC_NO_FP;
5330
5331 const TypePtr* dst_type = TypePtr::BOTTOM;
5332
5333 // Adjust memory effects of the runtime call based on input values.
5334 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5335 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5336
5337 flags |= RC_NARROW_MEM; // narrow in memory
5338 }
5339
5340 // Call it. Note that the length argument is not scaled.
5341 make_runtime_call(flags,
5342 OptoRuntime::unsafe_setmemory_Type(),
5343 StubRoutines::unsafe_setmemory(),
5344 "unsafe_setmemory",
5345 dst_type,
5346 dst_addr, size XTOP, byte);
5347
5348 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5349
5350 return true;
5351 }
5352
5353 #undef XTOP
5354
5355 //------------------------clone_coping-----------------------------------
5356 // Helper function for inline_native_clone.
5357 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5358 assert(obj_size != nullptr, "");
5359 Node* raw_obj = alloc_obj->in(1);
5360 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5361
5362 AllocateNode* alloc = nullptr;
5363 if (ReduceBulkZeroing &&
5364 // If we are implementing an array clone without knowing its source type
5365 // (can happen when compiling the array-guarded branch of a reflective
5366 // Object.clone() invocation), initialize the array within the allocation.
5367 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5368 // to a runtime clone call that assumes fully initialized source arrays.
5369 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5370 // We will be completely responsible for initializing this object -
5371 // mark Initialize node as complete.
5372 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5373 // The object was just allocated - there should be no any stores!
5374 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5375 // Mark as complete_with_arraycopy so that on AllocateNode
5376 // expansion, we know this AllocateNode is initialized by an array
5377 // copy and a StoreStore barrier exists after the array copy.
5378 alloc->initialization()->set_complete_with_arraycopy();
5379 }
5380
5381 Node* size = _gvn.transform(obj_size);
5382 access_clone(obj, alloc_obj, size, is_array);
5383
5384 // Do not let reads from the cloned object float above the arraycopy.
5385 if (alloc != nullptr) {
5386 // Do not let stores that initialize this object be reordered with
5387 // a subsequent store that would make this object accessible by
5388 // other threads.
5389 // Record what AllocateNode this StoreStore protects so that
5390 // escape analysis can go from the MemBarStoreStoreNode to the
5391 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5392 // based on the escape status of the AllocateNode.
5393 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5394 } else {
5395 insert_mem_bar(Op_MemBarCPUOrder);
5396 }
5397 }
5398
5399 //------------------------inline_native_clone----------------------------
5400 // protected native Object java.lang.Object.clone();
5401 //
5402 // Here are the simple edge cases:
5403 // null receiver => normal trap
5404 // virtual and clone was overridden => slow path to out-of-line clone
5405 // not cloneable or finalizer => slow path to out-of-line Object.clone
5406 //
5407 // The general case has two steps, allocation and copying.
5408 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5409 //
5410 // Copying also has two cases, oop arrays and everything else.
5411 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5412 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5413 //
5414 // These steps fold up nicely if and when the cloned object's klass
5415 // can be sharply typed as an object array, a type array, or an instance.
5416 //
5417 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5418 PhiNode* result_val;
5419
5420 // Set the reexecute bit for the interpreter to reexecute
5421 // the bytecode that invokes Object.clone if deoptimization happens.
5422 { PreserveReexecuteState preexecs(this);
5423 jvms()->set_should_reexecute(true);
5424
5425 Node* obj = null_check_receiver();
5426 if (stopped()) return true;
5427
5428 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5429
5430 // If we are going to clone an instance, we need its exact type to
5431 // know the number and types of fields to convert the clone to
5432 // loads/stores. Maybe a speculative type can help us.
5433 if (!obj_type->klass_is_exact() &&
5434 obj_type->speculative_type() != nullptr &&
5435 obj_type->speculative_type()->is_instance_klass()) {
5436 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5437 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5438 !spec_ik->has_injected_fields()) {
5439 if (!obj_type->isa_instptr() ||
5440 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5441 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5442 }
5443 }
5444 }
5445
5446 // Conservatively insert a memory barrier on all memory slices.
5447 // Do not let writes into the original float below the clone.
5448 insert_mem_bar(Op_MemBarCPUOrder);
5449
5450 // paths into result_reg:
5451 enum {
5452 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5453 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5454 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5455 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5456 PATH_LIMIT
5457 };
5458 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5459 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5460 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5461 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5462 record_for_igvn(result_reg);
5463
5464 Node* obj_klass = load_object_klass(obj);
5465 Node* array_obj = obj;
5466 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5467 if (array_ctl != nullptr) {
5468 // It's an array.
5469 PreserveJVMState pjvms(this);
5470 set_control(array_ctl);
5471 Node* obj_length = load_array_length(array_obj);
5472 Node* array_size = nullptr; // Size of the array without object alignment padding.
5473 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5474
5475 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5476 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5477 // If it is an oop array, it requires very special treatment,
5478 // because gc barriers are required when accessing the array.
5479 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5480 if (is_obja != nullptr) {
5481 PreserveJVMState pjvms2(this);
5482 set_control(is_obja);
5483 // Generate a direct call to the right arraycopy function(s).
5484 // Clones are always tightly coupled.
5485 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5486 ac->set_clone_oop_array();
5487 Node* n = _gvn.transform(ac);
5488 assert(n == ac, "cannot disappear");
5489 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5490
5491 result_reg->init_req(_objArray_path, control());
5492 result_val->init_req(_objArray_path, alloc_obj);
5493 result_i_o ->set_req(_objArray_path, i_o());
5494 result_mem ->set_req(_objArray_path, reset_memory());
5495 }
5496 }
5497 // Otherwise, there are no barriers to worry about.
5498 // (We can dispense with card marks if we know the allocation
5499 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5500 // causes the non-eden paths to take compensating steps to
5501 // simulate a fresh allocation, so that no further
5502 // card marks are required in compiled code to initialize
5503 // the object.)
5504
5505 if (!stopped()) {
5506 copy_to_clone(array_obj, alloc_obj, array_size, true);
5507
5508 // Present the results of the copy.
5509 result_reg->init_req(_array_path, control());
5510 result_val->init_req(_array_path, alloc_obj);
5511 result_i_o ->set_req(_array_path, i_o());
5512 result_mem ->set_req(_array_path, reset_memory());
5513 }
5514 }
5515
5516 // We only go to the instance fast case code if we pass a number of guards.
5517 // The paths which do not pass are accumulated in the slow_region.
5518 RegionNode* slow_region = new RegionNode(1);
5519 record_for_igvn(slow_region);
5520 if (!stopped()) {
5521 // It's an instance (we did array above). Make the slow-path tests.
5522 // If this is a virtual call, we generate a funny guard. We grab
5523 // the vtable entry corresponding to clone() from the target object.
5524 // If the target method which we are calling happens to be the
5525 // Object clone() method, we pass the guard. We do not need this
5526 // guard for non-virtual calls; the caller is known to be the native
5527 // Object clone().
5528 if (is_virtual) {
5529 generate_virtual_guard(obj_klass, slow_region);
5530 }
5531
5532 // The object must be easily cloneable and must not have a finalizer.
5533 // Both of these conditions may be checked in a single test.
5534 // We could optimize the test further, but we don't care.
5535 generate_misc_flags_guard(obj_klass,
5536 // Test both conditions:
5537 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5538 // Must be cloneable but not finalizer:
5539 KlassFlags::_misc_is_cloneable_fast,
5540 slow_region);
5541 }
5542
5543 if (!stopped()) {
5544 // It's an instance, and it passed the slow-path tests.
5545 PreserveJVMState pjvms(this);
5546 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5547 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5548 // is reexecuted if deoptimization occurs and there could be problems when merging
5549 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5550 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5551
5552 copy_to_clone(obj, alloc_obj, obj_size, false);
5553
5554 // Present the results of the slow call.
5555 result_reg->init_req(_instance_path, control());
5556 result_val->init_req(_instance_path, alloc_obj);
5557 result_i_o ->set_req(_instance_path, i_o());
5558 result_mem ->set_req(_instance_path, reset_memory());
5559 }
5560
5561 // Generate code for the slow case. We make a call to clone().
5562 set_control(_gvn.transform(slow_region));
5563 if (!stopped()) {
5564 PreserveJVMState pjvms(this);
5565 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5566 // We need to deoptimize on exception (see comment above)
5567 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5568 // this->control() comes from set_results_for_java_call
5569 result_reg->init_req(_slow_path, control());
5570 result_val->init_req(_slow_path, slow_result);
5571 result_i_o ->set_req(_slow_path, i_o());
5572 result_mem ->set_req(_slow_path, reset_memory());
5573 }
5574
5575 // Return the combined state.
5576 set_control( _gvn.transform(result_reg));
5577 set_i_o( _gvn.transform(result_i_o));
5578 set_all_memory( _gvn.transform(result_mem));
5579 } // original reexecute is set back here
5580
5581 set_result(_gvn.transform(result_val));
5582 return true;
5583 }
5584
5585 // If we have a tightly coupled allocation, the arraycopy may take care
5586 // of the array initialization. If one of the guards we insert between
5587 // the allocation and the arraycopy causes a deoptimization, an
5588 // uninitialized array will escape the compiled method. To prevent that
5589 // we set the JVM state for uncommon traps between the allocation and
5590 // the arraycopy to the state before the allocation so, in case of
5591 // deoptimization, we'll reexecute the allocation and the
5592 // initialization.
5593 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5594 if (alloc != nullptr) {
5595 ciMethod* trap_method = alloc->jvms()->method();
5596 int trap_bci = alloc->jvms()->bci();
5597
5598 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5599 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5600 // Make sure there's no store between the allocation and the
5601 // arraycopy otherwise visible side effects could be rexecuted
5602 // in case of deoptimization and cause incorrect execution.
5603 bool no_interfering_store = true;
5604 Node* mem = alloc->in(TypeFunc::Memory);
5605 if (mem->is_MergeMem()) {
5606 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5607 Node* n = mms.memory();
5608 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5609 assert(n->is_Store(), "what else?");
5610 no_interfering_store = false;
5611 break;
5612 }
5613 }
5614 } else {
5615 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5616 Node* n = mms.memory();
5617 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5618 assert(n->is_Store(), "what else?");
5619 no_interfering_store = false;
5620 break;
5621 }
5622 }
5623 }
5624
5625 if (no_interfering_store) {
5626 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5627
5628 JVMState* saved_jvms = jvms();
5629 saved_reexecute_sp = _reexecute_sp;
5630
5631 set_jvms(sfpt->jvms());
5632 _reexecute_sp = jvms()->sp();
5633
5634 return saved_jvms;
5635 }
5636 }
5637 }
5638 return nullptr;
5639 }
5640
5641 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5642 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5643 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5644 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5645 uint size = alloc->req();
5646 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5647 old_jvms->set_map(sfpt);
5648 for (uint i = 0; i < size; i++) {
5649 sfpt->init_req(i, alloc->in(i));
5650 }
5651 // re-push array length for deoptimization
5652 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5653 old_jvms->set_sp(old_jvms->sp()+1);
5654 old_jvms->set_monoff(old_jvms->monoff()+1);
5655 old_jvms->set_scloff(old_jvms->scloff()+1);
5656 old_jvms->set_endoff(old_jvms->endoff()+1);
5657 old_jvms->set_should_reexecute(true);
5658
5659 sfpt->set_i_o(map()->i_o());
5660 sfpt->set_memory(map()->memory());
5661 sfpt->set_control(map()->control());
5662 return sfpt;
5663 }
5664
5665 // In case of a deoptimization, we restart execution at the
5666 // allocation, allocating a new array. We would leave an uninitialized
5667 // array in the heap that GCs wouldn't expect. Move the allocation
5668 // after the traps so we don't allocate the array if we
5669 // deoptimize. This is possible because tightly_coupled_allocation()
5670 // guarantees there's no observer of the allocated array at this point
5671 // and the control flow is simple enough.
5672 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5673 int saved_reexecute_sp, uint new_idx) {
5674 if (saved_jvms_before_guards != nullptr && !stopped()) {
5675 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5676
5677 assert(alloc != nullptr, "only with a tightly coupled allocation");
5678 // restore JVM state to the state at the arraycopy
5679 saved_jvms_before_guards->map()->set_control(map()->control());
5680 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5681 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5682 // If we've improved the types of some nodes (null check) while
5683 // emitting the guards, propagate them to the current state
5684 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5685 set_jvms(saved_jvms_before_guards);
5686 _reexecute_sp = saved_reexecute_sp;
5687
5688 // Remove the allocation from above the guards
5689 CallProjections callprojs;
5690 alloc->extract_projections(&callprojs, true);
5691 InitializeNode* init = alloc->initialization();
5692 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5693 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5694 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5695
5696 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5697 // the allocation (i.e. is only valid if the allocation succeeds):
5698 // 1) replace CastIINode with AllocateArrayNode's length here
5699 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5700 //
5701 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5702 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5703 Node* init_control = init->proj_out(TypeFunc::Control);
5704 Node* alloc_length = alloc->Ideal_length();
5705 #ifdef ASSERT
5706 Node* prev_cast = nullptr;
5707 #endif
5708 for (uint i = 0; i < init_control->outcnt(); i++) {
5709 Node* init_out = init_control->raw_out(i);
5710 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5711 #ifdef ASSERT
5712 if (prev_cast == nullptr) {
5713 prev_cast = init_out;
5714 } else {
5715 if (prev_cast->cmp(*init_out) == false) {
5716 prev_cast->dump();
5717 init_out->dump();
5718 assert(false, "not equal CastIINode");
5719 }
5720 }
5721 #endif
5722 C->gvn_replace_by(init_out, alloc_length);
5723 }
5724 }
5725 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5726
5727 // move the allocation here (after the guards)
5728 _gvn.hash_delete(alloc);
5729 alloc->set_req(TypeFunc::Control, control());
5730 alloc->set_req(TypeFunc::I_O, i_o());
5731 Node *mem = reset_memory();
5732 set_all_memory(mem);
5733 alloc->set_req(TypeFunc::Memory, mem);
5734 set_control(init->proj_out_or_null(TypeFunc::Control));
5735 set_i_o(callprojs.fallthrough_ioproj);
5736
5737 // Update memory as done in GraphKit::set_output_for_allocation()
5738 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5739 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5740 if (ary_type->isa_aryptr() && length_type != nullptr) {
5741 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5742 }
5743 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5744 int elemidx = C->get_alias_index(telemref);
5745 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5746 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5747
5748 Node* allocx = _gvn.transform(alloc);
5749 assert(allocx == alloc, "where has the allocation gone?");
5750 assert(dest->is_CheckCastPP(), "not an allocation result?");
5751
5752 _gvn.hash_delete(dest);
5753 dest->set_req(0, control());
5754 Node* destx = _gvn.transform(dest);
5755 assert(destx == dest, "where has the allocation result gone?");
5756
5757 array_ideal_length(alloc, ary_type, true);
5758 }
5759 }
5760
5761 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5762 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5763 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5764 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5765 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5766 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5767 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5768 JVMState* saved_jvms_before_guards) {
5769 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5770 // There is at least one unrelated uncommon trap which needs to be replaced.
5771 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5772
5773 JVMState* saved_jvms = jvms();
5774 const int saved_reexecute_sp = _reexecute_sp;
5775 set_jvms(sfpt->jvms());
5776 _reexecute_sp = jvms()->sp();
5777
5778 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5779
5780 // Restore state
5781 set_jvms(saved_jvms);
5782 _reexecute_sp = saved_reexecute_sp;
5783 }
5784 }
5785
5786 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5787 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5788 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5789 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5790 while (if_proj->is_IfProj()) {
5791 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5792 if (uncommon_trap != nullptr) {
5793 create_new_uncommon_trap(uncommon_trap);
5794 }
5795 assert(if_proj->in(0)->is_If(), "must be If");
5796 if_proj = if_proj->in(0)->in(0);
5797 }
5798 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5799 "must have reached control projection of init node");
5800 }
5801
5802 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5803 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5804 assert(trap_request != 0, "no valid UCT trap request");
5805 PreserveJVMState pjvms(this);
5806 set_control(uncommon_trap_call->in(0));
5807 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5808 Deoptimization::trap_request_action(trap_request));
5809 assert(stopped(), "Should be stopped");
5810 _gvn.hash_delete(uncommon_trap_call);
5811 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5812 }
5813
5814 // Common checks for array sorting intrinsics arguments.
5815 // Returns `true` if checks passed.
5816 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5817 // check address of the class
5818 if (elementType == nullptr || elementType->is_top()) {
5819 return false; // dead path
5820 }
5821 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5822 if (elem_klass == nullptr) {
5823 return false; // dead path
5824 }
5825 // java_mirror_type() returns non-null for compile-time Class constants only
5826 ciType* elem_type = elem_klass->java_mirror_type();
5827 if (elem_type == nullptr) {
5828 return false;
5829 }
5830 bt = elem_type->basic_type();
5831 // Disable the intrinsic if the CPU does not support SIMD sort
5832 if (!Matcher::supports_simd_sort(bt)) {
5833 return false;
5834 }
5835 // check address of the array
5836 if (obj == nullptr || obj->is_top()) {
5837 return false; // dead path
5838 }
5839 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5840 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5841 return false; // failed input validation
5842 }
5843 return true;
5844 }
5845
5846 //------------------------------inline_array_partition-----------------------
5847 bool LibraryCallKit::inline_array_partition() {
5848 address stubAddr = StubRoutines::select_array_partition_function();
5849 if (stubAddr == nullptr) {
5850 return false; // Intrinsic's stub is not implemented on this platform
5851 }
5852 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5853
5854 // no receiver because it is a static method
5855 Node* elementType = argument(0);
5856 Node* obj = argument(1);
5857 Node* offset = argument(2); // long
5858 Node* fromIndex = argument(4);
5859 Node* toIndex = argument(5);
5860 Node* indexPivot1 = argument(6);
5861 Node* indexPivot2 = argument(7);
5862 // PartitionOperation: argument(8) is ignored
5863
5864 Node* pivotIndices = nullptr;
5865 BasicType bt = T_ILLEGAL;
5866
5867 if (!check_array_sort_arguments(elementType, obj, bt)) {
5868 return false;
5869 }
5870 null_check(obj);
5871 // If obj is dead, only null-path is taken.
5872 if (stopped()) {
5873 return true;
5874 }
5875 // Set the original stack and the reexecute bit for the interpreter to reexecute
5876 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5877 { PreserveReexecuteState preexecs(this);
5878 jvms()->set_should_reexecute(true);
5879
5880 Node* obj_adr = make_unsafe_address(obj, offset);
5881
5882 // create the pivotIndices array of type int and size = 2
5883 Node* size = intcon(2);
5884 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5885 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5886 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5887 guarantee(alloc != nullptr, "created above");
5888 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5889
5890 // pass the basic type enum to the stub
5891 Node* elemType = intcon(bt);
5892
5893 // Call the stub
5894 const char *stubName = "array_partition_stub";
5895 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5896 stubAddr, stubName, TypePtr::BOTTOM,
5897 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5898 indexPivot1, indexPivot2);
5899
5900 } // original reexecute is set back here
5901
5902 if (!stopped()) {
5903 set_result(pivotIndices);
5904 }
5905
5906 return true;
5907 }
5908
5909
5910 //------------------------------inline_array_sort-----------------------
5911 bool LibraryCallKit::inline_array_sort() {
5912 address stubAddr = StubRoutines::select_arraysort_function();
5913 if (stubAddr == nullptr) {
5914 return false; // Intrinsic's stub is not implemented on this platform
5915 }
5916 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5917
5918 // no receiver because it is a static method
5919 Node* elementType = argument(0);
5920 Node* obj = argument(1);
5921 Node* offset = argument(2); // long
5922 Node* fromIndex = argument(4);
5923 Node* toIndex = argument(5);
5924 // SortOperation: argument(6) is ignored
5925
5926 BasicType bt = T_ILLEGAL;
5927
5928 if (!check_array_sort_arguments(elementType, obj, bt)) {
5929 return false;
5930 }
5931 null_check(obj);
5932 // If obj is dead, only null-path is taken.
5933 if (stopped()) {
5934 return true;
5935 }
5936 Node* obj_adr = make_unsafe_address(obj, offset);
5937
5938 // pass the basic type enum to the stub
5939 Node* elemType = intcon(bt);
5940
5941 // Call the stub.
5942 const char *stubName = "arraysort_stub";
5943 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5944 stubAddr, stubName, TypePtr::BOTTOM,
5945 obj_adr, elemType, fromIndex, toIndex);
5946
5947 return true;
5948 }
5949
5950
5951 //------------------------------inline_arraycopy-----------------------
5952 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5953 // Object dest, int destPos,
5954 // int length);
5955 bool LibraryCallKit::inline_arraycopy() {
5956 // Get the arguments.
5957 Node* src = argument(0); // type: oop
5958 Node* src_offset = argument(1); // type: int
5959 Node* dest = argument(2); // type: oop
5960 Node* dest_offset = argument(3); // type: int
5961 Node* length = argument(4); // type: int
5962
5963 uint new_idx = C->unique();
5964
5965 // Check for allocation before we add nodes that would confuse
5966 // tightly_coupled_allocation()
5967 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5968
5969 int saved_reexecute_sp = -1;
5970 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5971 // See arraycopy_restore_alloc_state() comment
5972 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5973 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5974 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5975 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5976
5977 // The following tests must be performed
5978 // (1) src and dest are arrays.
5979 // (2) src and dest arrays must have elements of the same BasicType
5980 // (3) src and dest must not be null.
5981 // (4) src_offset must not be negative.
5982 // (5) dest_offset must not be negative.
5983 // (6) length must not be negative.
5984 // (7) src_offset + length must not exceed length of src.
5985 // (8) dest_offset + length must not exceed length of dest.
5986 // (9) each element of an oop array must be assignable
5987
5988 // (3) src and dest must not be null.
5989 // always do this here because we need the JVM state for uncommon traps
5990 Node* null_ctl = top();
5991 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5992 assert(null_ctl->is_top(), "no null control here");
5993 dest = null_check(dest, T_ARRAY);
5994
5995 if (!can_emit_guards) {
5996 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5997 // guards but the arraycopy node could still take advantage of a
5998 // tightly allocated allocation. tightly_coupled_allocation() is
5999 // called again to make sure it takes the null check above into
6000 // account: the null check is mandatory and if it caused an
6001 // uncommon trap to be emitted then the allocation can't be
6002 // considered tightly coupled in this context.
6003 alloc = tightly_coupled_allocation(dest);
6004 }
6005
6006 bool validated = false;
6007
6008 const Type* src_type = _gvn.type(src);
6009 const Type* dest_type = _gvn.type(dest);
6010 const TypeAryPtr* top_src = src_type->isa_aryptr();
6011 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6012
6013 // Do we have the type of src?
6014 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6015 // Do we have the type of dest?
6016 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6017 // Is the type for src from speculation?
6018 bool src_spec = false;
6019 // Is the type for dest from speculation?
6020 bool dest_spec = false;
6021
6022 if ((!has_src || !has_dest) && can_emit_guards) {
6023 // We don't have sufficient type information, let's see if
6024 // speculative types can help. We need to have types for both src
6025 // and dest so that it pays off.
6026
6027 // Do we already have or could we have type information for src
6028 bool could_have_src = has_src;
6029 // Do we already have or could we have type information for dest
6030 bool could_have_dest = has_dest;
6031
6032 ciKlass* src_k = nullptr;
6033 if (!has_src) {
6034 src_k = src_type->speculative_type_not_null();
6035 if (src_k != nullptr && src_k->is_array_klass()) {
6036 could_have_src = true;
6037 }
6038 }
6039
6040 ciKlass* dest_k = nullptr;
6041 if (!has_dest) {
6042 dest_k = dest_type->speculative_type_not_null();
6043 if (dest_k != nullptr && dest_k->is_array_klass()) {
6044 could_have_dest = true;
6045 }
6046 }
6047
6048 if (could_have_src && could_have_dest) {
6049 // This is going to pay off so emit the required guards
6050 if (!has_src) {
6051 src = maybe_cast_profiled_obj(src, src_k, true);
6052 src_type = _gvn.type(src);
6053 top_src = src_type->isa_aryptr();
6054 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6055 src_spec = true;
6056 }
6057 if (!has_dest) {
6058 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6059 dest_type = _gvn.type(dest);
6060 top_dest = dest_type->isa_aryptr();
6061 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6062 dest_spec = true;
6063 }
6064 }
6065 }
6066
6067 if (has_src && has_dest && can_emit_guards) {
6068 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6069 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6070 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6071 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6072
6073 if (src_elem == dest_elem && src_elem == T_OBJECT) {
6074 // If both arrays are object arrays then having the exact types
6075 // for both will remove the need for a subtype check at runtime
6076 // before the call and may make it possible to pick a faster copy
6077 // routine (without a subtype check on every element)
6078 // Do we have the exact type of src?
6079 bool could_have_src = src_spec;
6080 // Do we have the exact type of dest?
6081 bool could_have_dest = dest_spec;
6082 ciKlass* src_k = nullptr;
6083 ciKlass* dest_k = nullptr;
6084 if (!src_spec) {
6085 src_k = src_type->speculative_type_not_null();
6086 if (src_k != nullptr && src_k->is_array_klass()) {
6087 could_have_src = true;
6088 }
6089 }
6090 if (!dest_spec) {
6091 dest_k = dest_type->speculative_type_not_null();
6092 if (dest_k != nullptr && dest_k->is_array_klass()) {
6093 could_have_dest = true;
6094 }
6095 }
6096 if (could_have_src && could_have_dest) {
6097 // If we can have both exact types, emit the missing guards
6098 if (could_have_src && !src_spec) {
6099 src = maybe_cast_profiled_obj(src, src_k, true);
6100 }
6101 if (could_have_dest && !dest_spec) {
6102 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6103 }
6104 }
6105 }
6106 }
6107
6108 ciMethod* trap_method = method();
6109 int trap_bci = bci();
6110 if (saved_jvms_before_guards != nullptr) {
6111 trap_method = alloc->jvms()->method();
6112 trap_bci = alloc->jvms()->bci();
6113 }
6114
6115 bool negative_length_guard_generated = false;
6116
6117 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6118 can_emit_guards &&
6119 !src->is_top() && !dest->is_top()) {
6120 // validate arguments: enables transformation the ArrayCopyNode
6121 validated = true;
6122
6123 RegionNode* slow_region = new RegionNode(1);
6124 record_for_igvn(slow_region);
6125
6126 // (1) src and dest are arrays.
6127 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6128 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6129
6130 // (2) src and dest arrays must have elements of the same BasicType
6131 // done at macro expansion or at Ideal transformation time
6132
6133 // (4) src_offset must not be negative.
6134 generate_negative_guard(src_offset, slow_region);
6135
6136 // (5) dest_offset must not be negative.
6137 generate_negative_guard(dest_offset, slow_region);
6138
6139 // (7) src_offset + length must not exceed length of src.
6140 generate_limit_guard(src_offset, length,
6141 load_array_length(src),
6142 slow_region);
6143
6144 // (8) dest_offset + length must not exceed length of dest.
6145 generate_limit_guard(dest_offset, length,
6146 load_array_length(dest),
6147 slow_region);
6148
6149 // (6) length must not be negative.
6150 // This is also checked in generate_arraycopy() during macro expansion, but
6151 // we also have to check it here for the case where the ArrayCopyNode will
6152 // be eliminated by Escape Analysis.
6153 if (EliminateAllocations) {
6154 generate_negative_guard(length, slow_region);
6155 negative_length_guard_generated = true;
6156 }
6157
6158 // (9) each element of an oop array must be assignable
6159 Node* dest_klass = load_object_klass(dest);
6160 if (src != dest) {
6161 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6162
6163 if (not_subtype_ctrl != top()) {
6164 PreserveJVMState pjvms(this);
6165 set_control(not_subtype_ctrl);
6166 uncommon_trap(Deoptimization::Reason_intrinsic,
6167 Deoptimization::Action_make_not_entrant);
6168 assert(stopped(), "Should be stopped");
6169 }
6170 }
6171 {
6172 PreserveJVMState pjvms(this);
6173 set_control(_gvn.transform(slow_region));
6174 uncommon_trap(Deoptimization::Reason_intrinsic,
6175 Deoptimization::Action_make_not_entrant);
6176 assert(stopped(), "Should be stopped");
6177 }
6178
6179 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6180 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6181 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6182 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6183 }
6184
6185 if (stopped()) {
6186 return true;
6187 }
6188
6189 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6190 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6191 // so the compiler has a chance to eliminate them: during macro expansion,
6192 // we have to set their control (CastPP nodes are eliminated).
6193 load_object_klass(src), load_object_klass(dest),
6194 load_array_length(src), load_array_length(dest));
6195
6196 ac->set_arraycopy(validated);
6197
6198 Node* n = _gvn.transform(ac);
6199 if (n == ac) {
6200 ac->connect_outputs(this);
6201 } else {
6202 assert(validated, "shouldn't transform if all arguments not validated");
6203 set_all_memory(n);
6204 }
6205 clear_upper_avx();
6206
6207
6208 return true;
6209 }
6210
6211
6212 // Helper function which determines if an arraycopy immediately follows
6213 // an allocation, with no intervening tests or other escapes for the object.
6214 AllocateArrayNode*
6215 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6216 if (stopped()) return nullptr; // no fast path
6217 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6218
6219 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6220 if (alloc == nullptr) return nullptr;
6221
6222 Node* rawmem = memory(Compile::AliasIdxRaw);
6223 // Is the allocation's memory state untouched?
6224 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6225 // Bail out if there have been raw-memory effects since the allocation.
6226 // (Example: There might have been a call or safepoint.)
6227 return nullptr;
6228 }
6229 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6230 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6231 return nullptr;
6232 }
6233
6234 // There must be no unexpected observers of this allocation.
6235 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6236 Node* obs = ptr->fast_out(i);
6237 if (obs != this->map()) {
6238 return nullptr;
6239 }
6240 }
6241
6242 // This arraycopy must unconditionally follow the allocation of the ptr.
6243 Node* alloc_ctl = ptr->in(0);
6244 Node* ctl = control();
6245 while (ctl != alloc_ctl) {
6246 // There may be guards which feed into the slow_region.
6247 // Any other control flow means that we might not get a chance
6248 // to finish initializing the allocated object.
6249 // Various low-level checks bottom out in uncommon traps. These
6250 // are considered safe since we've already checked above that
6251 // there is no unexpected observer of this allocation.
6252 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6253 assert(ctl->in(0)->is_If(), "must be If");
6254 ctl = ctl->in(0)->in(0);
6255 } else {
6256 return nullptr;
6257 }
6258 }
6259
6260 // If we get this far, we have an allocation which immediately
6261 // precedes the arraycopy, and we can take over zeroing the new object.
6262 // The arraycopy will finish the initialization, and provide
6263 // a new control state to which we will anchor the destination pointer.
6264
6265 return alloc;
6266 }
6267
6268 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6269 if (node->is_IfProj()) {
6270 Node* other_proj = node->as_IfProj()->other_if_proj();
6271 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6272 Node* obs = other_proj->fast_out(j);
6273 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6274 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6275 return obs->as_CallStaticJava();
6276 }
6277 }
6278 }
6279 return nullptr;
6280 }
6281
6282 //-------------inline_encodeISOArray-----------------------------------
6283 // encode char[] to byte[] in ISO_8859_1 or ASCII
6284 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6285 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6286 // no receiver since it is static method
6287 Node *src = argument(0);
6288 Node *src_offset = argument(1);
6289 Node *dst = argument(2);
6290 Node *dst_offset = argument(3);
6291 Node *length = argument(4);
6292
6293 src = must_be_not_null(src, true);
6294 dst = must_be_not_null(dst, true);
6295
6296 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6297 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6298 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6299 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6300 // failed array check
6301 return false;
6302 }
6303
6304 // Figure out the size and type of the elements we will be copying.
6305 BasicType src_elem = src_type->elem()->array_element_basic_type();
6306 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6307 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6308 return false;
6309 }
6310
6311 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6312 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6313 // 'src_start' points to src array + scaled offset
6314 // 'dst_start' points to dst array + scaled offset
6315
6316 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6317 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6318 enc = _gvn.transform(enc);
6319 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6320 set_memory(res_mem, mtype);
6321 set_result(enc);
6322 clear_upper_avx();
6323
6324 return true;
6325 }
6326
6327 //-------------inline_multiplyToLen-----------------------------------
6328 bool LibraryCallKit::inline_multiplyToLen() {
6329 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6330
6331 address stubAddr = StubRoutines::multiplyToLen();
6332 if (stubAddr == nullptr) {
6333 return false; // Intrinsic's stub is not implemented on this platform
6334 }
6335 const char* stubName = "multiplyToLen";
6336
6337 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6338
6339 // no receiver because it is a static method
6340 Node* x = argument(0);
6341 Node* xlen = argument(1);
6342 Node* y = argument(2);
6343 Node* ylen = argument(3);
6344 Node* z = argument(4);
6345
6346 x = must_be_not_null(x, true);
6347 y = must_be_not_null(y, true);
6348
6349 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6350 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6351 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6352 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6353 // failed array check
6354 return false;
6355 }
6356
6357 BasicType x_elem = x_type->elem()->array_element_basic_type();
6358 BasicType y_elem = y_type->elem()->array_element_basic_type();
6359 if (x_elem != T_INT || y_elem != T_INT) {
6360 return false;
6361 }
6362
6363 Node* x_start = array_element_address(x, intcon(0), x_elem);
6364 Node* y_start = array_element_address(y, intcon(0), y_elem);
6365 // 'x_start' points to x array + scaled xlen
6366 // 'y_start' points to y array + scaled ylen
6367
6368 Node* z_start = array_element_address(z, intcon(0), T_INT);
6369
6370 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6371 OptoRuntime::multiplyToLen_Type(),
6372 stubAddr, stubName, TypePtr::BOTTOM,
6373 x_start, xlen, y_start, ylen, z_start);
6374
6375 C->set_has_split_ifs(true); // Has chance for split-if optimization
6376 set_result(z);
6377 return true;
6378 }
6379
6380 //-------------inline_squareToLen------------------------------------
6381 bool LibraryCallKit::inline_squareToLen() {
6382 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6383
6384 address stubAddr = StubRoutines::squareToLen();
6385 if (stubAddr == nullptr) {
6386 return false; // Intrinsic's stub is not implemented on this platform
6387 }
6388 const char* stubName = "squareToLen";
6389
6390 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6391
6392 Node* x = argument(0);
6393 Node* len = argument(1);
6394 Node* z = argument(2);
6395 Node* zlen = argument(3);
6396
6397 x = must_be_not_null(x, true);
6398 z = must_be_not_null(z, true);
6399
6400 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6401 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6402 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6403 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6404 // failed array check
6405 return false;
6406 }
6407
6408 BasicType x_elem = x_type->elem()->array_element_basic_type();
6409 BasicType z_elem = z_type->elem()->array_element_basic_type();
6410 if (x_elem != T_INT || z_elem != T_INT) {
6411 return false;
6412 }
6413
6414
6415 Node* x_start = array_element_address(x, intcon(0), x_elem);
6416 Node* z_start = array_element_address(z, intcon(0), z_elem);
6417
6418 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6419 OptoRuntime::squareToLen_Type(),
6420 stubAddr, stubName, TypePtr::BOTTOM,
6421 x_start, len, z_start, zlen);
6422
6423 set_result(z);
6424 return true;
6425 }
6426
6427 //-------------inline_mulAdd------------------------------------------
6428 bool LibraryCallKit::inline_mulAdd() {
6429 assert(UseMulAddIntrinsic, "not implemented on this platform");
6430
6431 address stubAddr = StubRoutines::mulAdd();
6432 if (stubAddr == nullptr) {
6433 return false; // Intrinsic's stub is not implemented on this platform
6434 }
6435 const char* stubName = "mulAdd";
6436
6437 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6438
6439 Node* out = argument(0);
6440 Node* in = argument(1);
6441 Node* offset = argument(2);
6442 Node* len = argument(3);
6443 Node* k = argument(4);
6444
6445 in = must_be_not_null(in, true);
6446 out = must_be_not_null(out, true);
6447
6448 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6449 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6450 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6451 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6452 // failed array check
6453 return false;
6454 }
6455
6456 BasicType out_elem = out_type->elem()->array_element_basic_type();
6457 BasicType in_elem = in_type->elem()->array_element_basic_type();
6458 if (out_elem != T_INT || in_elem != T_INT) {
6459 return false;
6460 }
6461
6462 Node* outlen = load_array_length(out);
6463 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6464 Node* out_start = array_element_address(out, intcon(0), out_elem);
6465 Node* in_start = array_element_address(in, intcon(0), in_elem);
6466
6467 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6468 OptoRuntime::mulAdd_Type(),
6469 stubAddr, stubName, TypePtr::BOTTOM,
6470 out_start,in_start, new_offset, len, k);
6471 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6472 set_result(result);
6473 return true;
6474 }
6475
6476 //-------------inline_montgomeryMultiply-----------------------------------
6477 bool LibraryCallKit::inline_montgomeryMultiply() {
6478 address stubAddr = StubRoutines::montgomeryMultiply();
6479 if (stubAddr == nullptr) {
6480 return false; // Intrinsic's stub is not implemented on this platform
6481 }
6482
6483 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6484 const char* stubName = "montgomery_multiply";
6485
6486 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6487
6488 Node* a = argument(0);
6489 Node* b = argument(1);
6490 Node* n = argument(2);
6491 Node* len = argument(3);
6492 Node* inv = argument(4);
6493 Node* m = argument(6);
6494
6495 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6496 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6497 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6498 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6499 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6500 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6501 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6502 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6503 // failed array check
6504 return false;
6505 }
6506
6507 BasicType a_elem = a_type->elem()->array_element_basic_type();
6508 BasicType b_elem = b_type->elem()->array_element_basic_type();
6509 BasicType n_elem = n_type->elem()->array_element_basic_type();
6510 BasicType m_elem = m_type->elem()->array_element_basic_type();
6511 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6512 return false;
6513 }
6514
6515 // Make the call
6516 {
6517 Node* a_start = array_element_address(a, intcon(0), a_elem);
6518 Node* b_start = array_element_address(b, intcon(0), b_elem);
6519 Node* n_start = array_element_address(n, intcon(0), n_elem);
6520 Node* m_start = array_element_address(m, intcon(0), m_elem);
6521
6522 Node* call = make_runtime_call(RC_LEAF,
6523 OptoRuntime::montgomeryMultiply_Type(),
6524 stubAddr, stubName, TypePtr::BOTTOM,
6525 a_start, b_start, n_start, len, inv, top(),
6526 m_start);
6527 set_result(m);
6528 }
6529
6530 return true;
6531 }
6532
6533 bool LibraryCallKit::inline_montgomerySquare() {
6534 address stubAddr = StubRoutines::montgomerySquare();
6535 if (stubAddr == nullptr) {
6536 return false; // Intrinsic's stub is not implemented on this platform
6537 }
6538
6539 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6540 const char* stubName = "montgomery_square";
6541
6542 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6543
6544 Node* a = argument(0);
6545 Node* n = argument(1);
6546 Node* len = argument(2);
6547 Node* inv = argument(3);
6548 Node* m = argument(5);
6549
6550 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6551 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6552 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6553 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6554 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6555 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6556 // failed array check
6557 return false;
6558 }
6559
6560 BasicType a_elem = a_type->elem()->array_element_basic_type();
6561 BasicType n_elem = n_type->elem()->array_element_basic_type();
6562 BasicType m_elem = m_type->elem()->array_element_basic_type();
6563 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6564 return false;
6565 }
6566
6567 // Make the call
6568 {
6569 Node* a_start = array_element_address(a, intcon(0), a_elem);
6570 Node* n_start = array_element_address(n, intcon(0), n_elem);
6571 Node* m_start = array_element_address(m, intcon(0), m_elem);
6572
6573 Node* call = make_runtime_call(RC_LEAF,
6574 OptoRuntime::montgomerySquare_Type(),
6575 stubAddr, stubName, TypePtr::BOTTOM,
6576 a_start, n_start, len, inv, top(),
6577 m_start);
6578 set_result(m);
6579 }
6580
6581 return true;
6582 }
6583
6584 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6585 address stubAddr = nullptr;
6586 const char* stubName = nullptr;
6587
6588 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6589 if (stubAddr == nullptr) {
6590 return false; // Intrinsic's stub is not implemented on this platform
6591 }
6592
6593 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6594
6595 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6596
6597 Node* newArr = argument(0);
6598 Node* oldArr = argument(1);
6599 Node* newIdx = argument(2);
6600 Node* shiftCount = argument(3);
6601 Node* numIter = argument(4);
6602
6603 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6604 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6605 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6606 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6607 return false;
6608 }
6609
6610 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6611 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6612 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6613 return false;
6614 }
6615
6616 // Make the call
6617 {
6618 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6619 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6620
6621 Node* call = make_runtime_call(RC_LEAF,
6622 OptoRuntime::bigIntegerShift_Type(),
6623 stubAddr,
6624 stubName,
6625 TypePtr::BOTTOM,
6626 newArr_start,
6627 oldArr_start,
6628 newIdx,
6629 shiftCount,
6630 numIter);
6631 }
6632
6633 return true;
6634 }
6635
6636 //-------------inline_vectorizedMismatch------------------------------
6637 bool LibraryCallKit::inline_vectorizedMismatch() {
6638 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6639
6640 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6641 Node* obja = argument(0); // Object
6642 Node* aoffset = argument(1); // long
6643 Node* objb = argument(3); // Object
6644 Node* boffset = argument(4); // long
6645 Node* length = argument(6); // int
6646 Node* scale = argument(7); // int
6647
6648 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6649 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6650 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6651 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6652 scale == top()) {
6653 return false; // failed input validation
6654 }
6655
6656 Node* obja_adr = make_unsafe_address(obja, aoffset);
6657 Node* objb_adr = make_unsafe_address(objb, boffset);
6658
6659 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6660 //
6661 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6662 // if (length <= inline_limit) {
6663 // inline_path:
6664 // vmask = VectorMaskGen length
6665 // vload1 = LoadVectorMasked obja, vmask
6666 // vload2 = LoadVectorMasked objb, vmask
6667 // result1 = VectorCmpMasked vload1, vload2, vmask
6668 // } else {
6669 // call_stub_path:
6670 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6671 // }
6672 // exit_block:
6673 // return Phi(result1, result2);
6674 //
6675 enum { inline_path = 1, // input is small enough to process it all at once
6676 stub_path = 2, // input is too large; call into the VM
6677 PATH_LIMIT = 3
6678 };
6679
6680 Node* exit_block = new RegionNode(PATH_LIMIT);
6681 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6682 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6683
6684 Node* call_stub_path = control();
6685
6686 BasicType elem_bt = T_ILLEGAL;
6687
6688 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6689 if (scale_t->is_con()) {
6690 switch (scale_t->get_con()) {
6691 case 0: elem_bt = T_BYTE; break;
6692 case 1: elem_bt = T_SHORT; break;
6693 case 2: elem_bt = T_INT; break;
6694 case 3: elem_bt = T_LONG; break;
6695
6696 default: elem_bt = T_ILLEGAL; break; // not supported
6697 }
6698 }
6699
6700 int inline_limit = 0;
6701 bool do_partial_inline = false;
6702
6703 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6704 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6705 do_partial_inline = inline_limit >= 16;
6706 }
6707
6708 if (do_partial_inline) {
6709 assert(elem_bt != T_ILLEGAL, "sanity");
6710
6711 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6712 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6713 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6714
6715 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6716 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6717 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6718
6719 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6720
6721 if (!stopped()) {
6722 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6723
6724 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6725 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6726 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6727 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6728
6729 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6730 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6731 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6732 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6733
6734 exit_block->init_req(inline_path, control());
6735 memory_phi->init_req(inline_path, map()->memory());
6736 result_phi->init_req(inline_path, result);
6737
6738 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6739 clear_upper_avx();
6740 }
6741 }
6742 }
6743
6744 if (call_stub_path != nullptr) {
6745 set_control(call_stub_path);
6746
6747 Node* call = make_runtime_call(RC_LEAF,
6748 OptoRuntime::vectorizedMismatch_Type(),
6749 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6750 obja_adr, objb_adr, length, scale);
6751
6752 exit_block->init_req(stub_path, control());
6753 memory_phi->init_req(stub_path, map()->memory());
6754 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6755 }
6756
6757 exit_block = _gvn.transform(exit_block);
6758 memory_phi = _gvn.transform(memory_phi);
6759 result_phi = _gvn.transform(result_phi);
6760
6761 record_for_igvn(exit_block);
6762 record_for_igvn(memory_phi);
6763 record_for_igvn(result_phi);
6764
6765 set_control(exit_block);
6766 set_all_memory(memory_phi);
6767 set_result(result_phi);
6768
6769 return true;
6770 }
6771
6772 //------------------------------inline_vectorizedHashcode----------------------------
6773 bool LibraryCallKit::inline_vectorizedHashCode() {
6774 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6775
6776 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6777 Node* array = argument(0);
6778 Node* offset = argument(1);
6779 Node* length = argument(2);
6780 Node* initialValue = argument(3);
6781 Node* basic_type = argument(4);
6782
6783 if (basic_type == top()) {
6784 return false; // failed input validation
6785 }
6786
6787 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6788 if (!basic_type_t->is_con()) {
6789 return false; // Only intrinsify if mode argument is constant
6790 }
6791
6792 array = must_be_not_null(array, true);
6793
6794 BasicType bt = (BasicType)basic_type_t->get_con();
6795
6796 // Resolve address of first element
6797 Node* array_start = array_element_address(array, offset, bt);
6798
6799 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6800 array_start, length, initialValue, basic_type)));
6801 clear_upper_avx();
6802
6803 return true;
6804 }
6805
6806 /**
6807 * Calculate CRC32 for byte.
6808 * int java.util.zip.CRC32.update(int crc, int b)
6809 */
6810 bool LibraryCallKit::inline_updateCRC32() {
6811 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6812 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6813 // no receiver since it is static method
6814 Node* crc = argument(0); // type: int
6815 Node* b = argument(1); // type: int
6816
6817 /*
6818 * int c = ~ crc;
6819 * b = timesXtoThe32[(b ^ c) & 0xFF];
6820 * b = b ^ (c >>> 8);
6821 * crc = ~b;
6822 */
6823
6824 Node* M1 = intcon(-1);
6825 crc = _gvn.transform(new XorINode(crc, M1));
6826 Node* result = _gvn.transform(new XorINode(crc, b));
6827 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6828
6829 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6830 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6831 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6832 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6833
6834 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6835 result = _gvn.transform(new XorINode(crc, result));
6836 result = _gvn.transform(new XorINode(result, M1));
6837 set_result(result);
6838 return true;
6839 }
6840
6841 /**
6842 * Calculate CRC32 for byte[] array.
6843 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6844 */
6845 bool LibraryCallKit::inline_updateBytesCRC32() {
6846 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6847 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6848 // no receiver since it is static method
6849 Node* crc = argument(0); // type: int
6850 Node* src = argument(1); // type: oop
6851 Node* offset = argument(2); // type: int
6852 Node* length = argument(3); // type: int
6853
6854 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6855 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6856 // failed array check
6857 return false;
6858 }
6859
6860 // Figure out the size and type of the elements we will be copying.
6861 BasicType src_elem = src_type->elem()->array_element_basic_type();
6862 if (src_elem != T_BYTE) {
6863 return false;
6864 }
6865
6866 // 'src_start' points to src array + scaled offset
6867 src = must_be_not_null(src, true);
6868 Node* src_start = array_element_address(src, offset, src_elem);
6869
6870 // We assume that range check is done by caller.
6871 // TODO: generate range check (offset+length < src.length) in debug VM.
6872
6873 // Call the stub.
6874 address stubAddr = StubRoutines::updateBytesCRC32();
6875 const char *stubName = "updateBytesCRC32";
6876
6877 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6878 stubAddr, stubName, TypePtr::BOTTOM,
6879 crc, src_start, length);
6880 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6881 set_result(result);
6882 return true;
6883 }
6884
6885 /**
6886 * Calculate CRC32 for ByteBuffer.
6887 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6888 */
6889 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6890 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6891 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6892 // no receiver since it is static method
6893 Node* crc = argument(0); // type: int
6894 Node* src = argument(1); // type: long
6895 Node* offset = argument(3); // type: int
6896 Node* length = argument(4); // type: int
6897
6898 src = ConvL2X(src); // adjust Java long to machine word
6899 Node* base = _gvn.transform(new CastX2PNode(src));
6900 offset = ConvI2X(offset);
6901
6902 // 'src_start' points to src array + scaled offset
6903 Node* src_start = basic_plus_adr(top(), base, offset);
6904
6905 // Call the stub.
6906 address stubAddr = StubRoutines::updateBytesCRC32();
6907 const char *stubName = "updateBytesCRC32";
6908
6909 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6910 stubAddr, stubName, TypePtr::BOTTOM,
6911 crc, src_start, length);
6912 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6913 set_result(result);
6914 return true;
6915 }
6916
6917 //------------------------------get_table_from_crc32c_class-----------------------
6918 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6919 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6920 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6921
6922 return table;
6923 }
6924
6925 //------------------------------inline_updateBytesCRC32C-----------------------
6926 //
6927 // Calculate CRC32C for byte[] array.
6928 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6929 //
6930 bool LibraryCallKit::inline_updateBytesCRC32C() {
6931 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6932 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6933 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6934 // no receiver since it is a static method
6935 Node* crc = argument(0); // type: int
6936 Node* src = argument(1); // type: oop
6937 Node* offset = argument(2); // type: int
6938 Node* end = argument(3); // type: int
6939
6940 Node* length = _gvn.transform(new SubINode(end, offset));
6941
6942 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6943 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6944 // failed array check
6945 return false;
6946 }
6947
6948 // Figure out the size and type of the elements we will be copying.
6949 BasicType src_elem = src_type->elem()->array_element_basic_type();
6950 if (src_elem != T_BYTE) {
6951 return false;
6952 }
6953
6954 // 'src_start' points to src array + scaled offset
6955 src = must_be_not_null(src, true);
6956 Node* src_start = array_element_address(src, offset, src_elem);
6957
6958 // static final int[] byteTable in class CRC32C
6959 Node* table = get_table_from_crc32c_class(callee()->holder());
6960 table = must_be_not_null(table, true);
6961 Node* table_start = array_element_address(table, intcon(0), T_INT);
6962
6963 // We assume that range check is done by caller.
6964 // TODO: generate range check (offset+length < src.length) in debug VM.
6965
6966 // Call the stub.
6967 address stubAddr = StubRoutines::updateBytesCRC32C();
6968 const char *stubName = "updateBytesCRC32C";
6969
6970 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6971 stubAddr, stubName, TypePtr::BOTTOM,
6972 crc, src_start, length, table_start);
6973 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6974 set_result(result);
6975 return true;
6976 }
6977
6978 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6979 //
6980 // Calculate CRC32C for DirectByteBuffer.
6981 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6982 //
6983 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6984 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6985 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6986 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6987 // no receiver since it is a static method
6988 Node* crc = argument(0); // type: int
6989 Node* src = argument(1); // type: long
6990 Node* offset = argument(3); // type: int
6991 Node* end = argument(4); // type: int
6992
6993 Node* length = _gvn.transform(new SubINode(end, offset));
6994
6995 src = ConvL2X(src); // adjust Java long to machine word
6996 Node* base = _gvn.transform(new CastX2PNode(src));
6997 offset = ConvI2X(offset);
6998
6999 // 'src_start' points to src array + scaled offset
7000 Node* src_start = basic_plus_adr(top(), base, offset);
7001
7002 // static final int[] byteTable in class CRC32C
7003 Node* table = get_table_from_crc32c_class(callee()->holder());
7004 table = must_be_not_null(table, true);
7005 Node* table_start = array_element_address(table, intcon(0), T_INT);
7006
7007 // Call the stub.
7008 address stubAddr = StubRoutines::updateBytesCRC32C();
7009 const char *stubName = "updateBytesCRC32C";
7010
7011 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7012 stubAddr, stubName, TypePtr::BOTTOM,
7013 crc, src_start, length, table_start);
7014 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7015 set_result(result);
7016 return true;
7017 }
7018
7019 //------------------------------inline_updateBytesAdler32----------------------
7020 //
7021 // Calculate Adler32 checksum for byte[] array.
7022 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7023 //
7024 bool LibraryCallKit::inline_updateBytesAdler32() {
7025 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7026 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7027 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7028 // no receiver since it is static method
7029 Node* crc = argument(0); // type: int
7030 Node* src = argument(1); // type: oop
7031 Node* offset = argument(2); // type: int
7032 Node* length = argument(3); // type: int
7033
7034 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7035 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7036 // failed array check
7037 return false;
7038 }
7039
7040 // Figure out the size and type of the elements we will be copying.
7041 BasicType src_elem = src_type->elem()->array_element_basic_type();
7042 if (src_elem != T_BYTE) {
7043 return false;
7044 }
7045
7046 // 'src_start' points to src array + scaled offset
7047 Node* src_start = array_element_address(src, offset, src_elem);
7048
7049 // We assume that range check is done by caller.
7050 // TODO: generate range check (offset+length < src.length) in debug VM.
7051
7052 // Call the stub.
7053 address stubAddr = StubRoutines::updateBytesAdler32();
7054 const char *stubName = "updateBytesAdler32";
7055
7056 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7057 stubAddr, stubName, TypePtr::BOTTOM,
7058 crc, src_start, length);
7059 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7060 set_result(result);
7061 return true;
7062 }
7063
7064 //------------------------------inline_updateByteBufferAdler32---------------
7065 //
7066 // Calculate Adler32 checksum for DirectByteBuffer.
7067 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7068 //
7069 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7070 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7071 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7072 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7073 // no receiver since it is static method
7074 Node* crc = argument(0); // type: int
7075 Node* src = argument(1); // type: long
7076 Node* offset = argument(3); // type: int
7077 Node* length = argument(4); // type: int
7078
7079 src = ConvL2X(src); // adjust Java long to machine word
7080 Node* base = _gvn.transform(new CastX2PNode(src));
7081 offset = ConvI2X(offset);
7082
7083 // 'src_start' points to src array + scaled offset
7084 Node* src_start = basic_plus_adr(top(), base, offset);
7085
7086 // Call the stub.
7087 address stubAddr = StubRoutines::updateBytesAdler32();
7088 const char *stubName = "updateBytesAdler32";
7089
7090 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7091 stubAddr, stubName, TypePtr::BOTTOM,
7092 crc, src_start, length);
7093
7094 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7095 set_result(result);
7096 return true;
7097 }
7098
7099 //----------------------------inline_reference_get0----------------------------
7100 // public T java.lang.ref.Reference.get();
7101 bool LibraryCallKit::inline_reference_get0() {
7102 const int referent_offset = java_lang_ref_Reference::referent_offset();
7103
7104 // Get the argument:
7105 Node* reference_obj = null_check_receiver();
7106 if (stopped()) return true;
7107
7108 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7109 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7110 decorators, /*is_static*/ false, nullptr);
7111 if (result == nullptr) return false;
7112
7113 // Add memory barrier to prevent commoning reads from this field
7114 // across safepoint since GC can change its value.
7115 insert_mem_bar(Op_MemBarCPUOrder);
7116
7117 set_result(result);
7118 return true;
7119 }
7120
7121 //----------------------------inline_reference_refersTo0----------------------------
7122 // bool java.lang.ref.Reference.refersTo0();
7123 // bool java.lang.ref.PhantomReference.refersTo0();
7124 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7125 // Get arguments:
7126 Node* reference_obj = null_check_receiver();
7127 Node* other_obj = argument(1);
7128 if (stopped()) return true;
7129
7130 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7131 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7132 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7133 decorators, /*is_static*/ false, nullptr);
7134 if (referent == nullptr) return false;
7135
7136 // Add memory barrier to prevent commoning reads from this field
7137 // across safepoint since GC can change its value.
7138 insert_mem_bar(Op_MemBarCPUOrder);
7139
7140 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7141 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7142 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7143
7144 RegionNode* region = new RegionNode(3);
7145 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7146
7147 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7148 region->init_req(1, if_true);
7149 phi->init_req(1, intcon(1));
7150
7151 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7152 region->init_req(2, if_false);
7153 phi->init_req(2, intcon(0));
7154
7155 set_control(_gvn.transform(region));
7156 record_for_igvn(region);
7157 set_result(_gvn.transform(phi));
7158 return true;
7159 }
7160
7161 //----------------------------inline_reference_clear0----------------------------
7162 // void java.lang.ref.Reference.clear0();
7163 // void java.lang.ref.PhantomReference.clear0();
7164 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7165 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7166
7167 // Get arguments
7168 Node* reference_obj = null_check_receiver();
7169 if (stopped()) return true;
7170
7171 // Common access parameters
7172 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7173 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7174 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7175 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7176 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7177
7178 Node* referent = access_load_at(reference_obj,
7179 referent_field_addr,
7180 referent_field_addr_type,
7181 val_type,
7182 T_OBJECT,
7183 decorators);
7184
7185 IdealKit ideal(this);
7186 #define __ ideal.
7187 __ if_then(referent, BoolTest::ne, null());
7188 sync_kit(ideal);
7189 access_store_at(reference_obj,
7190 referent_field_addr,
7191 referent_field_addr_type,
7192 null(),
7193 val_type,
7194 T_OBJECT,
7195 decorators);
7196 __ sync_kit(this);
7197 __ end_if();
7198 final_sync(ideal);
7199 #undef __
7200
7201 return true;
7202 }
7203
7204 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7205 DecoratorSet decorators, bool is_static,
7206 ciInstanceKlass* fromKls) {
7207 if (fromKls == nullptr) {
7208 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7209 assert(tinst != nullptr, "obj is null");
7210 assert(tinst->is_loaded(), "obj is not loaded");
7211 fromKls = tinst->instance_klass();
7212 } else {
7213 assert(is_static, "only for static field access");
7214 }
7215 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7216 ciSymbol::make(fieldTypeString),
7217 is_static);
7218
7219 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7220 if (field == nullptr) return (Node *) nullptr;
7221
7222 if (is_static) {
7223 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7224 fromObj = makecon(tip);
7225 }
7226
7227 // Next code copied from Parse::do_get_xxx():
7228
7229 // Compute address and memory type.
7230 int offset = field->offset_in_bytes();
7231 bool is_vol = field->is_volatile();
7232 ciType* field_klass = field->type();
7233 assert(field_klass->is_loaded(), "should be loaded");
7234 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7235 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7236 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7237 "slice of address and input slice don't match");
7238 BasicType bt = field->layout_type();
7239
7240 // Build the resultant type of the load
7241 const Type *type;
7242 if (bt == T_OBJECT) {
7243 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7244 } else {
7245 type = Type::get_const_basic_type(bt);
7246 }
7247
7248 if (is_vol) {
7249 decorators |= MO_SEQ_CST;
7250 }
7251
7252 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7253 }
7254
7255 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7256 bool is_exact /* true */, bool is_static /* false */,
7257 ciInstanceKlass * fromKls /* nullptr */) {
7258 if (fromKls == nullptr) {
7259 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7260 assert(tinst != nullptr, "obj is null");
7261 assert(tinst->is_loaded(), "obj is not loaded");
7262 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7263 fromKls = tinst->instance_klass();
7264 }
7265 else {
7266 assert(is_static, "only for static field access");
7267 }
7268 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7269 ciSymbol::make(fieldTypeString),
7270 is_static);
7271
7272 assert(field != nullptr, "undefined field");
7273 assert(!field->is_volatile(), "not defined for volatile fields");
7274
7275 if (is_static) {
7276 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7277 fromObj = makecon(tip);
7278 }
7279
7280 // Next code copied from Parse::do_get_xxx():
7281
7282 // Compute address and memory type.
7283 int offset = field->offset_in_bytes();
7284 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7285
7286 return adr;
7287 }
7288
7289 //------------------------------inline_aescrypt_Block-----------------------
7290 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7291 address stubAddr = nullptr;
7292 const char *stubName;
7293 assert(UseAES, "need AES instruction support");
7294
7295 switch(id) {
7296 case vmIntrinsics::_aescrypt_encryptBlock:
7297 stubAddr = StubRoutines::aescrypt_encryptBlock();
7298 stubName = "aescrypt_encryptBlock";
7299 break;
7300 case vmIntrinsics::_aescrypt_decryptBlock:
7301 stubAddr = StubRoutines::aescrypt_decryptBlock();
7302 stubName = "aescrypt_decryptBlock";
7303 break;
7304 default:
7305 break;
7306 }
7307 if (stubAddr == nullptr) return false;
7308
7309 Node* aescrypt_object = argument(0);
7310 Node* src = argument(1);
7311 Node* src_offset = argument(2);
7312 Node* dest = argument(3);
7313 Node* dest_offset = argument(4);
7314
7315 src = must_be_not_null(src, true);
7316 dest = must_be_not_null(dest, true);
7317
7318 // (1) src and dest are arrays.
7319 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7320 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7321 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7322 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7323
7324 // for the quick and dirty code we will skip all the checks.
7325 // we are just trying to get the call to be generated.
7326 Node* src_start = src;
7327 Node* dest_start = dest;
7328 if (src_offset != nullptr || dest_offset != nullptr) {
7329 assert(src_offset != nullptr && dest_offset != nullptr, "");
7330 src_start = array_element_address(src, src_offset, T_BYTE);
7331 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7332 }
7333
7334 // now need to get the start of its expanded key array
7335 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7336 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7337 if (k_start == nullptr) return false;
7338
7339 // Call the stub.
7340 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7341 stubAddr, stubName, TypePtr::BOTTOM,
7342 src_start, dest_start, k_start);
7343
7344 return true;
7345 }
7346
7347 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7348 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7349 address stubAddr = nullptr;
7350 const char *stubName = nullptr;
7351
7352 assert(UseAES, "need AES instruction support");
7353
7354 switch(id) {
7355 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7356 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7357 stubName = "cipherBlockChaining_encryptAESCrypt";
7358 break;
7359 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7360 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7361 stubName = "cipherBlockChaining_decryptAESCrypt";
7362 break;
7363 default:
7364 break;
7365 }
7366 if (stubAddr == nullptr) return false;
7367
7368 Node* cipherBlockChaining_object = argument(0);
7369 Node* src = argument(1);
7370 Node* src_offset = argument(2);
7371 Node* len = argument(3);
7372 Node* dest = argument(4);
7373 Node* dest_offset = argument(5);
7374
7375 src = must_be_not_null(src, false);
7376 dest = must_be_not_null(dest, false);
7377
7378 // (1) src and dest are arrays.
7379 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7380 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7381 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7382 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7383
7384 // checks are the responsibility of the caller
7385 Node* src_start = src;
7386 Node* dest_start = dest;
7387 if (src_offset != nullptr || dest_offset != nullptr) {
7388 assert(src_offset != nullptr && dest_offset != nullptr, "");
7389 src_start = array_element_address(src, src_offset, T_BYTE);
7390 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7391 }
7392
7393 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7394 // (because of the predicated logic executed earlier).
7395 // so we cast it here safely.
7396 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7397
7398 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7399 if (embeddedCipherObj == nullptr) return false;
7400
7401 // cast it to what we know it will be at runtime
7402 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7403 assert(tinst != nullptr, "CBC obj is null");
7404 assert(tinst->is_loaded(), "CBC obj is not loaded");
7405 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7406 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7407
7408 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7409 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7410 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7411 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7412 aescrypt_object = _gvn.transform(aescrypt_object);
7413
7414 // we need to get the start of the aescrypt_object's expanded key array
7415 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7416 if (k_start == nullptr) return false;
7417
7418 // similarly, get the start address of the r vector
7419 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7420 if (objRvec == nullptr) return false;
7421 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7422
7423 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7424 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7425 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7426 stubAddr, stubName, TypePtr::BOTTOM,
7427 src_start, dest_start, k_start, r_start, len);
7428
7429 // return cipher length (int)
7430 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7431 set_result(retvalue);
7432 return true;
7433 }
7434
7435 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7436 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7437 address stubAddr = nullptr;
7438 const char *stubName = nullptr;
7439
7440 assert(UseAES, "need AES instruction support");
7441
7442 switch (id) {
7443 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7444 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7445 stubName = "electronicCodeBook_encryptAESCrypt";
7446 break;
7447 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7448 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7449 stubName = "electronicCodeBook_decryptAESCrypt";
7450 break;
7451 default:
7452 break;
7453 }
7454
7455 if (stubAddr == nullptr) return false;
7456
7457 Node* electronicCodeBook_object = argument(0);
7458 Node* src = argument(1);
7459 Node* src_offset = argument(2);
7460 Node* len = argument(3);
7461 Node* dest = argument(4);
7462 Node* dest_offset = argument(5);
7463
7464 // (1) src and dest are arrays.
7465 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7466 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7467 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7468 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7469
7470 // checks are the responsibility of the caller
7471 Node* src_start = src;
7472 Node* dest_start = dest;
7473 if (src_offset != nullptr || dest_offset != nullptr) {
7474 assert(src_offset != nullptr && dest_offset != nullptr, "");
7475 src_start = array_element_address(src, src_offset, T_BYTE);
7476 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7477 }
7478
7479 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7480 // (because of the predicated logic executed earlier).
7481 // so we cast it here safely.
7482 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7483
7484 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7485 if (embeddedCipherObj == nullptr) return false;
7486
7487 // cast it to what we know it will be at runtime
7488 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7489 assert(tinst != nullptr, "ECB obj is null");
7490 assert(tinst->is_loaded(), "ECB obj is not loaded");
7491 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7492 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7493
7494 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7495 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7496 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7497 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7498 aescrypt_object = _gvn.transform(aescrypt_object);
7499
7500 // we need to get the start of the aescrypt_object's expanded key array
7501 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7502 if (k_start == nullptr) return false;
7503
7504 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7505 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7506 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7507 stubAddr, stubName, TypePtr::BOTTOM,
7508 src_start, dest_start, k_start, len);
7509
7510 // return cipher length (int)
7511 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7512 set_result(retvalue);
7513 return true;
7514 }
7515
7516 //------------------------------inline_counterMode_AESCrypt-----------------------
7517 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7518 assert(UseAES, "need AES instruction support");
7519 if (!UseAESCTRIntrinsics) return false;
7520
7521 address stubAddr = nullptr;
7522 const char *stubName = nullptr;
7523 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7524 stubAddr = StubRoutines::counterMode_AESCrypt();
7525 stubName = "counterMode_AESCrypt";
7526 }
7527 if (stubAddr == nullptr) return false;
7528
7529 Node* counterMode_object = argument(0);
7530 Node* src = argument(1);
7531 Node* src_offset = argument(2);
7532 Node* len = argument(3);
7533 Node* dest = argument(4);
7534 Node* dest_offset = argument(5);
7535
7536 // (1) src and dest are arrays.
7537 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7538 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7539 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7540 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7541
7542 // checks are the responsibility of the caller
7543 Node* src_start = src;
7544 Node* dest_start = dest;
7545 if (src_offset != nullptr || dest_offset != nullptr) {
7546 assert(src_offset != nullptr && dest_offset != nullptr, "");
7547 src_start = array_element_address(src, src_offset, T_BYTE);
7548 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7549 }
7550
7551 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7552 // (because of the predicated logic executed earlier).
7553 // so we cast it here safely.
7554 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7555 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7556 if (embeddedCipherObj == nullptr) return false;
7557 // cast it to what we know it will be at runtime
7558 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7559 assert(tinst != nullptr, "CTR obj is null");
7560 assert(tinst->is_loaded(), "CTR obj is not loaded");
7561 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7562 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7563 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7564 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7565 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7566 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7567 aescrypt_object = _gvn.transform(aescrypt_object);
7568 // we need to get the start of the aescrypt_object's expanded key array
7569 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7570 if (k_start == nullptr) return false;
7571 // similarly, get the start address of the r vector
7572 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7573 if (obj_counter == nullptr) return false;
7574 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7575
7576 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7577 if (saved_encCounter == nullptr) return false;
7578 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7579 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7580
7581 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7582 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7583 OptoRuntime::counterMode_aescrypt_Type(),
7584 stubAddr, stubName, TypePtr::BOTTOM,
7585 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7586
7587 // return cipher length (int)
7588 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7589 set_result(retvalue);
7590 return true;
7591 }
7592
7593 //------------------------------get_key_start_from_aescrypt_object-----------------------
7594 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7595 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7596 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7597 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7598 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7599 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7600 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7601 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7602 if (objSessionK == nullptr) {
7603 return (Node *) nullptr;
7604 }
7605 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7606 #else
7607 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7608 #endif // PPC64
7609 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7610 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7611
7612 // now have the array, need to get the start address of the K array
7613 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7614 return k_start;
7615 }
7616
7617 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7618 // Return node representing slow path of predicate check.
7619 // the pseudo code we want to emulate with this predicate is:
7620 // for encryption:
7621 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7622 // for decryption:
7623 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7624 // note cipher==plain is more conservative than the original java code but that's OK
7625 //
7626 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7627 // The receiver was checked for null already.
7628 Node* objCBC = argument(0);
7629
7630 Node* src = argument(1);
7631 Node* dest = argument(4);
7632
7633 // Load embeddedCipher field of CipherBlockChaining object.
7634 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7635
7636 // get AESCrypt klass for instanceOf check
7637 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7638 // will have same classloader as CipherBlockChaining object
7639 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7640 assert(tinst != nullptr, "CBCobj is null");
7641 assert(tinst->is_loaded(), "CBCobj is not loaded");
7642
7643 // we want to do an instanceof comparison against the AESCrypt class
7644 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7645 if (!klass_AESCrypt->is_loaded()) {
7646 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7647 Node* ctrl = control();
7648 set_control(top()); // no regular fast path
7649 return ctrl;
7650 }
7651
7652 src = must_be_not_null(src, true);
7653 dest = must_be_not_null(dest, true);
7654
7655 // Resolve oops to stable for CmpP below.
7656 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7657
7658 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7659 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7660 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7661
7662 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7663
7664 // for encryption, we are done
7665 if (!decrypting)
7666 return instof_false; // even if it is null
7667
7668 // for decryption, we need to add a further check to avoid
7669 // taking the intrinsic path when cipher and plain are the same
7670 // see the original java code for why.
7671 RegionNode* region = new RegionNode(3);
7672 region->init_req(1, instof_false);
7673
7674 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7675 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7676 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7677 region->init_req(2, src_dest_conjoint);
7678
7679 record_for_igvn(region);
7680 return _gvn.transform(region);
7681 }
7682
7683 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7684 // Return node representing slow path of predicate check.
7685 // the pseudo code we want to emulate with this predicate is:
7686 // for encryption:
7687 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7688 // for decryption:
7689 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7690 // note cipher==plain is more conservative than the original java code but that's OK
7691 //
7692 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7693 // The receiver was checked for null already.
7694 Node* objECB = argument(0);
7695
7696 // Load embeddedCipher field of ElectronicCodeBook object.
7697 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7698
7699 // get AESCrypt klass for instanceOf check
7700 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7701 // will have same classloader as ElectronicCodeBook object
7702 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7703 assert(tinst != nullptr, "ECBobj is null");
7704 assert(tinst->is_loaded(), "ECBobj is not loaded");
7705
7706 // we want to do an instanceof comparison against the AESCrypt class
7707 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7708 if (!klass_AESCrypt->is_loaded()) {
7709 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7710 Node* ctrl = control();
7711 set_control(top()); // no regular fast path
7712 return ctrl;
7713 }
7714 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7715
7716 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7717 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7718 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7719
7720 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7721
7722 // for encryption, we are done
7723 if (!decrypting)
7724 return instof_false; // even if it is null
7725
7726 // for decryption, we need to add a further check to avoid
7727 // taking the intrinsic path when cipher and plain are the same
7728 // see the original java code for why.
7729 RegionNode* region = new RegionNode(3);
7730 region->init_req(1, instof_false);
7731 Node* src = argument(1);
7732 Node* dest = argument(4);
7733 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7734 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7735 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7736 region->init_req(2, src_dest_conjoint);
7737
7738 record_for_igvn(region);
7739 return _gvn.transform(region);
7740 }
7741
7742 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7743 // Return node representing slow path of predicate check.
7744 // the pseudo code we want to emulate with this predicate is:
7745 // for encryption:
7746 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7747 // for decryption:
7748 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7749 // note cipher==plain is more conservative than the original java code but that's OK
7750 //
7751
7752 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7753 // The receiver was checked for null already.
7754 Node* objCTR = argument(0);
7755
7756 // Load embeddedCipher field of CipherBlockChaining object.
7757 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7758
7759 // get AESCrypt klass for instanceOf check
7760 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7761 // will have same classloader as CipherBlockChaining object
7762 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7763 assert(tinst != nullptr, "CTRobj is null");
7764 assert(tinst->is_loaded(), "CTRobj is not loaded");
7765
7766 // we want to do an instanceof comparison against the AESCrypt class
7767 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7768 if (!klass_AESCrypt->is_loaded()) {
7769 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7770 Node* ctrl = control();
7771 set_control(top()); // no regular fast path
7772 return ctrl;
7773 }
7774
7775 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7776 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7777 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7778 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7779 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7780
7781 return instof_false; // even if it is null
7782 }
7783
7784 //------------------------------inline_ghash_processBlocks
7785 bool LibraryCallKit::inline_ghash_processBlocks() {
7786 address stubAddr;
7787 const char *stubName;
7788 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7789
7790 stubAddr = StubRoutines::ghash_processBlocks();
7791 stubName = "ghash_processBlocks";
7792
7793 Node* data = argument(0);
7794 Node* offset = argument(1);
7795 Node* len = argument(2);
7796 Node* state = argument(3);
7797 Node* subkeyH = argument(4);
7798
7799 state = must_be_not_null(state, true);
7800 subkeyH = must_be_not_null(subkeyH, true);
7801 data = must_be_not_null(data, true);
7802
7803 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7804 assert(state_start, "state is null");
7805 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7806 assert(subkeyH_start, "subkeyH is null");
7807 Node* data_start = array_element_address(data, offset, T_BYTE);
7808 assert(data_start, "data is null");
7809
7810 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7811 OptoRuntime::ghash_processBlocks_Type(),
7812 stubAddr, stubName, TypePtr::BOTTOM,
7813 state_start, subkeyH_start, data_start, len);
7814 return true;
7815 }
7816
7817 //------------------------------inline_chacha20Block
7818 bool LibraryCallKit::inline_chacha20Block() {
7819 address stubAddr;
7820 const char *stubName;
7821 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7822
7823 stubAddr = StubRoutines::chacha20Block();
7824 stubName = "chacha20Block";
7825
7826 Node* state = argument(0);
7827 Node* result = argument(1);
7828
7829 state = must_be_not_null(state, true);
7830 result = must_be_not_null(result, true);
7831
7832 Node* state_start = array_element_address(state, intcon(0), T_INT);
7833 assert(state_start, "state is null");
7834 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7835 assert(result_start, "result is null");
7836
7837 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7838 OptoRuntime::chacha20Block_Type(),
7839 stubAddr, stubName, TypePtr::BOTTOM,
7840 state_start, result_start);
7841 // return key stream length (int)
7842 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7843 set_result(retvalue);
7844 return true;
7845 }
7846
7847 //------------------------------inline_kyberNtt
7848 bool LibraryCallKit::inline_kyberNtt() {
7849 address stubAddr;
7850 const char *stubName;
7851 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7852 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7853
7854 stubAddr = StubRoutines::kyberNtt();
7855 stubName = "kyberNtt";
7856 if (!stubAddr) return false;
7857
7858 Node* coeffs = argument(0);
7859 Node* ntt_zetas = argument(1);
7860
7861 coeffs = must_be_not_null(coeffs, true);
7862 ntt_zetas = must_be_not_null(ntt_zetas, true);
7863
7864 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7865 assert(coeffs_start, "coeffs is null");
7866 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7867 assert(ntt_zetas_start, "ntt_zetas is null");
7868 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7869 OptoRuntime::kyberNtt_Type(),
7870 stubAddr, stubName, TypePtr::BOTTOM,
7871 coeffs_start, ntt_zetas_start);
7872 // return an int
7873 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7874 set_result(retvalue);
7875 return true;
7876 }
7877
7878 //------------------------------inline_kyberInverseNtt
7879 bool LibraryCallKit::inline_kyberInverseNtt() {
7880 address stubAddr;
7881 const char *stubName;
7882 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7883 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7884
7885 stubAddr = StubRoutines::kyberInverseNtt();
7886 stubName = "kyberInverseNtt";
7887 if (!stubAddr) return false;
7888
7889 Node* coeffs = argument(0);
7890 Node* zetas = argument(1);
7891
7892 coeffs = must_be_not_null(coeffs, true);
7893 zetas = must_be_not_null(zetas, true);
7894
7895 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7896 assert(coeffs_start, "coeffs is null");
7897 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7898 assert(zetas_start, "inverseNtt_zetas is null");
7899 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7900 OptoRuntime::kyberInverseNtt_Type(),
7901 stubAddr, stubName, TypePtr::BOTTOM,
7902 coeffs_start, zetas_start);
7903
7904 // return an int
7905 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7906 set_result(retvalue);
7907 return true;
7908 }
7909
7910 //------------------------------inline_kyberNttMult
7911 bool LibraryCallKit::inline_kyberNttMult() {
7912 address stubAddr;
7913 const char *stubName;
7914 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7915 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7916
7917 stubAddr = StubRoutines::kyberNttMult();
7918 stubName = "kyberNttMult";
7919 if (!stubAddr) return false;
7920
7921 Node* result = argument(0);
7922 Node* ntta = argument(1);
7923 Node* nttb = argument(2);
7924 Node* zetas = argument(3);
7925
7926 result = must_be_not_null(result, true);
7927 ntta = must_be_not_null(ntta, true);
7928 nttb = must_be_not_null(nttb, true);
7929 zetas = must_be_not_null(zetas, true);
7930
7931 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7932 assert(result_start, "result is null");
7933 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7934 assert(ntta_start, "ntta is null");
7935 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7936 assert(nttb_start, "nttb is null");
7937 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7938 assert(zetas_start, "nttMult_zetas is null");
7939 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7940 OptoRuntime::kyberNttMult_Type(),
7941 stubAddr, stubName, TypePtr::BOTTOM,
7942 result_start, ntta_start, nttb_start,
7943 zetas_start);
7944
7945 // return an int
7946 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7947 set_result(retvalue);
7948
7949 return true;
7950 }
7951
7952 //------------------------------inline_kyberAddPoly_2
7953 bool LibraryCallKit::inline_kyberAddPoly_2() {
7954 address stubAddr;
7955 const char *stubName;
7956 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7957 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7958
7959 stubAddr = StubRoutines::kyberAddPoly_2();
7960 stubName = "kyberAddPoly_2";
7961 if (!stubAddr) return false;
7962
7963 Node* result = argument(0);
7964 Node* a = argument(1);
7965 Node* b = argument(2);
7966
7967 result = must_be_not_null(result, true);
7968 a = must_be_not_null(a, true);
7969 b = must_be_not_null(b, true);
7970
7971 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7972 assert(result_start, "result is null");
7973 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7974 assert(a_start, "a is null");
7975 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7976 assert(b_start, "b is null");
7977 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7978 OptoRuntime::kyberAddPoly_2_Type(),
7979 stubAddr, stubName, TypePtr::BOTTOM,
7980 result_start, a_start, b_start);
7981 // return an int
7982 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7983 set_result(retvalue);
7984 return true;
7985 }
7986
7987 //------------------------------inline_kyberAddPoly_3
7988 bool LibraryCallKit::inline_kyberAddPoly_3() {
7989 address stubAddr;
7990 const char *stubName;
7991 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7992 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7993
7994 stubAddr = StubRoutines::kyberAddPoly_3();
7995 stubName = "kyberAddPoly_3";
7996 if (!stubAddr) return false;
7997
7998 Node* result = argument(0);
7999 Node* a = argument(1);
8000 Node* b = argument(2);
8001 Node* c = argument(3);
8002
8003 result = must_be_not_null(result, true);
8004 a = must_be_not_null(a, true);
8005 b = must_be_not_null(b, true);
8006 c = must_be_not_null(c, true);
8007
8008 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8009 assert(result_start, "result is null");
8010 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8011 assert(a_start, "a is null");
8012 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8013 assert(b_start, "b is null");
8014 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8015 assert(c_start, "c is null");
8016 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8017 OptoRuntime::kyberAddPoly_3_Type(),
8018 stubAddr, stubName, TypePtr::BOTTOM,
8019 result_start, a_start, b_start, c_start);
8020 // return an int
8021 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8022 set_result(retvalue);
8023 return true;
8024 }
8025
8026 //------------------------------inline_kyber12To16
8027 bool LibraryCallKit::inline_kyber12To16() {
8028 address stubAddr;
8029 const char *stubName;
8030 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8031 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8032
8033 stubAddr = StubRoutines::kyber12To16();
8034 stubName = "kyber12To16";
8035 if (!stubAddr) return false;
8036
8037 Node* condensed = argument(0);
8038 Node* condensedOffs = argument(1);
8039 Node* parsed = argument(2);
8040 Node* parsedLength = argument(3);
8041
8042 condensed = must_be_not_null(condensed, true);
8043 parsed = must_be_not_null(parsed, true);
8044
8045 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8046 assert(condensed_start, "condensed is null");
8047 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8048 assert(parsed_start, "parsed is null");
8049 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8050 OptoRuntime::kyber12To16_Type(),
8051 stubAddr, stubName, TypePtr::BOTTOM,
8052 condensed_start, condensedOffs, parsed_start, parsedLength);
8053 // return an int
8054 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8055 set_result(retvalue);
8056 return true;
8057
8058 }
8059
8060 //------------------------------inline_kyberBarrettReduce
8061 bool LibraryCallKit::inline_kyberBarrettReduce() {
8062 address stubAddr;
8063 const char *stubName;
8064 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8065 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8066
8067 stubAddr = StubRoutines::kyberBarrettReduce();
8068 stubName = "kyberBarrettReduce";
8069 if (!stubAddr) return false;
8070
8071 Node* coeffs = argument(0);
8072
8073 coeffs = must_be_not_null(coeffs, true);
8074
8075 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8076 assert(coeffs_start, "coeffs is null");
8077 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8078 OptoRuntime::kyberBarrettReduce_Type(),
8079 stubAddr, stubName, TypePtr::BOTTOM,
8080 coeffs_start);
8081 // return an int
8082 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8083 set_result(retvalue);
8084 return true;
8085 }
8086
8087 //------------------------------inline_dilithiumAlmostNtt
8088 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8089 address stubAddr;
8090 const char *stubName;
8091 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8092 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8093
8094 stubAddr = StubRoutines::dilithiumAlmostNtt();
8095 stubName = "dilithiumAlmostNtt";
8096 if (!stubAddr) return false;
8097
8098 Node* coeffs = argument(0);
8099 Node* ntt_zetas = argument(1);
8100
8101 coeffs = must_be_not_null(coeffs, true);
8102 ntt_zetas = must_be_not_null(ntt_zetas, true);
8103
8104 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8105 assert(coeffs_start, "coeffs is null");
8106 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8107 assert(ntt_zetas_start, "ntt_zetas is null");
8108 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8109 OptoRuntime::dilithiumAlmostNtt_Type(),
8110 stubAddr, stubName, TypePtr::BOTTOM,
8111 coeffs_start, ntt_zetas_start);
8112 // return an int
8113 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8114 set_result(retvalue);
8115 return true;
8116 }
8117
8118 //------------------------------inline_dilithiumAlmostInverseNtt
8119 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8120 address stubAddr;
8121 const char *stubName;
8122 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8123 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8124
8125 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8126 stubName = "dilithiumAlmostInverseNtt";
8127 if (!stubAddr) return false;
8128
8129 Node* coeffs = argument(0);
8130 Node* zetas = argument(1);
8131
8132 coeffs = must_be_not_null(coeffs, true);
8133 zetas = must_be_not_null(zetas, true);
8134
8135 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8136 assert(coeffs_start, "coeffs is null");
8137 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8138 assert(zetas_start, "inverseNtt_zetas is null");
8139 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8140 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8141 stubAddr, stubName, TypePtr::BOTTOM,
8142 coeffs_start, zetas_start);
8143 // return an int
8144 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8145 set_result(retvalue);
8146 return true;
8147 }
8148
8149 //------------------------------inline_dilithiumNttMult
8150 bool LibraryCallKit::inline_dilithiumNttMult() {
8151 address stubAddr;
8152 const char *stubName;
8153 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8154 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8155
8156 stubAddr = StubRoutines::dilithiumNttMult();
8157 stubName = "dilithiumNttMult";
8158 if (!stubAddr) return false;
8159
8160 Node* result = argument(0);
8161 Node* ntta = argument(1);
8162 Node* nttb = argument(2);
8163 Node* zetas = argument(3);
8164
8165 result = must_be_not_null(result, true);
8166 ntta = must_be_not_null(ntta, true);
8167 nttb = must_be_not_null(nttb, true);
8168 zetas = must_be_not_null(zetas, true);
8169
8170 Node* result_start = array_element_address(result, intcon(0), T_INT);
8171 assert(result_start, "result is null");
8172 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8173 assert(ntta_start, "ntta is null");
8174 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8175 assert(nttb_start, "nttb is null");
8176 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8177 OptoRuntime::dilithiumNttMult_Type(),
8178 stubAddr, stubName, TypePtr::BOTTOM,
8179 result_start, ntta_start, nttb_start);
8180
8181 // return an int
8182 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8183 set_result(retvalue);
8184
8185 return true;
8186 }
8187
8188 //------------------------------inline_dilithiumMontMulByConstant
8189 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8190 address stubAddr;
8191 const char *stubName;
8192 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8193 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8194
8195 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8196 stubName = "dilithiumMontMulByConstant";
8197 if (!stubAddr) return false;
8198
8199 Node* coeffs = argument(0);
8200 Node* constant = argument(1);
8201
8202 coeffs = must_be_not_null(coeffs, true);
8203
8204 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8205 assert(coeffs_start, "coeffs is null");
8206 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8207 OptoRuntime::dilithiumMontMulByConstant_Type(),
8208 stubAddr, stubName, TypePtr::BOTTOM,
8209 coeffs_start, constant);
8210
8211 // return an int
8212 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8213 set_result(retvalue);
8214 return true;
8215 }
8216
8217
8218 //------------------------------inline_dilithiumDecomposePoly
8219 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8220 address stubAddr;
8221 const char *stubName;
8222 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8223 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8224
8225 stubAddr = StubRoutines::dilithiumDecomposePoly();
8226 stubName = "dilithiumDecomposePoly";
8227 if (!stubAddr) return false;
8228
8229 Node* input = argument(0);
8230 Node* lowPart = argument(1);
8231 Node* highPart = argument(2);
8232 Node* twoGamma2 = argument(3);
8233 Node* multiplier = argument(4);
8234
8235 input = must_be_not_null(input, true);
8236 lowPart = must_be_not_null(lowPart, true);
8237 highPart = must_be_not_null(highPart, true);
8238
8239 Node* input_start = array_element_address(input, intcon(0), T_INT);
8240 assert(input_start, "input is null");
8241 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8242 assert(lowPart_start, "lowPart is null");
8243 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8244 assert(highPart_start, "highPart is null");
8245
8246 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8247 OptoRuntime::dilithiumDecomposePoly_Type(),
8248 stubAddr, stubName, TypePtr::BOTTOM,
8249 input_start, lowPart_start, highPart_start,
8250 twoGamma2, multiplier);
8251
8252 // return an int
8253 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8254 set_result(retvalue);
8255 return true;
8256 }
8257
8258 bool LibraryCallKit::inline_base64_encodeBlock() {
8259 address stubAddr;
8260 const char *stubName;
8261 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8262 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8263 stubAddr = StubRoutines::base64_encodeBlock();
8264 stubName = "encodeBlock";
8265
8266 if (!stubAddr) return false;
8267 Node* base64obj = argument(0);
8268 Node* src = argument(1);
8269 Node* offset = argument(2);
8270 Node* len = argument(3);
8271 Node* dest = argument(4);
8272 Node* dp = argument(5);
8273 Node* isURL = argument(6);
8274
8275 src = must_be_not_null(src, true);
8276 dest = must_be_not_null(dest, true);
8277
8278 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8279 assert(src_start, "source array is null");
8280 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8281 assert(dest_start, "destination array is null");
8282
8283 Node* base64 = make_runtime_call(RC_LEAF,
8284 OptoRuntime::base64_encodeBlock_Type(),
8285 stubAddr, stubName, TypePtr::BOTTOM,
8286 src_start, offset, len, dest_start, dp, isURL);
8287 return true;
8288 }
8289
8290 bool LibraryCallKit::inline_base64_decodeBlock() {
8291 address stubAddr;
8292 const char *stubName;
8293 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8294 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8295 stubAddr = StubRoutines::base64_decodeBlock();
8296 stubName = "decodeBlock";
8297
8298 if (!stubAddr) return false;
8299 Node* base64obj = argument(0);
8300 Node* src = argument(1);
8301 Node* src_offset = argument(2);
8302 Node* len = argument(3);
8303 Node* dest = argument(4);
8304 Node* dest_offset = argument(5);
8305 Node* isURL = argument(6);
8306 Node* isMIME = argument(7);
8307
8308 src = must_be_not_null(src, true);
8309 dest = must_be_not_null(dest, true);
8310
8311 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8312 assert(src_start, "source array is null");
8313 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8314 assert(dest_start, "destination array is null");
8315
8316 Node* call = make_runtime_call(RC_LEAF,
8317 OptoRuntime::base64_decodeBlock_Type(),
8318 stubAddr, stubName, TypePtr::BOTTOM,
8319 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8320 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8321 set_result(result);
8322 return true;
8323 }
8324
8325 bool LibraryCallKit::inline_poly1305_processBlocks() {
8326 address stubAddr;
8327 const char *stubName;
8328 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8329 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8330 stubAddr = StubRoutines::poly1305_processBlocks();
8331 stubName = "poly1305_processBlocks";
8332
8333 if (!stubAddr) return false;
8334 null_check_receiver(); // null-check receiver
8335 if (stopped()) return true;
8336
8337 Node* input = argument(1);
8338 Node* input_offset = argument(2);
8339 Node* len = argument(3);
8340 Node* alimbs = argument(4);
8341 Node* rlimbs = argument(5);
8342
8343 input = must_be_not_null(input, true);
8344 alimbs = must_be_not_null(alimbs, true);
8345 rlimbs = must_be_not_null(rlimbs, true);
8346
8347 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8348 assert(input_start, "input array is null");
8349 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8350 assert(acc_start, "acc array is null");
8351 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8352 assert(r_start, "r array is null");
8353
8354 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8355 OptoRuntime::poly1305_processBlocks_Type(),
8356 stubAddr, stubName, TypePtr::BOTTOM,
8357 input_start, len, acc_start, r_start);
8358 return true;
8359 }
8360
8361 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8362 address stubAddr;
8363 const char *stubName;
8364 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8365 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8366 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8367 stubName = "intpoly_montgomeryMult_P256";
8368
8369 if (!stubAddr) return false;
8370 null_check_receiver(); // null-check receiver
8371 if (stopped()) return true;
8372
8373 Node* a = argument(1);
8374 Node* b = argument(2);
8375 Node* r = argument(3);
8376
8377 a = must_be_not_null(a, true);
8378 b = must_be_not_null(b, true);
8379 r = must_be_not_null(r, true);
8380
8381 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8382 assert(a_start, "a array is null");
8383 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8384 assert(b_start, "b array is null");
8385 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8386 assert(r_start, "r array is null");
8387
8388 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8389 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8390 stubAddr, stubName, TypePtr::BOTTOM,
8391 a_start, b_start, r_start);
8392 return true;
8393 }
8394
8395 bool LibraryCallKit::inline_intpoly_assign() {
8396 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8397 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8398 const char *stubName = "intpoly_assign";
8399 address stubAddr = StubRoutines::intpoly_assign();
8400 if (!stubAddr) return false;
8401
8402 Node* set = argument(0);
8403 Node* a = argument(1);
8404 Node* b = argument(2);
8405 Node* arr_length = load_array_length(a);
8406
8407 a = must_be_not_null(a, true);
8408 b = must_be_not_null(b, true);
8409
8410 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8411 assert(a_start, "a array is null");
8412 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8413 assert(b_start, "b array is null");
8414
8415 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8416 OptoRuntime::intpoly_assign_Type(),
8417 stubAddr, stubName, TypePtr::BOTTOM,
8418 set, a_start, b_start, arr_length);
8419 return true;
8420 }
8421
8422 //------------------------------inline_digestBase_implCompress-----------------------
8423 //
8424 // Calculate MD5 for single-block byte[] array.
8425 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8426 //
8427 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8428 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8429 //
8430 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8431 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8432 //
8433 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8434 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8435 //
8436 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8437 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8438 //
8439 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8440 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8441
8442 Node* digestBase_obj = argument(0);
8443 Node* src = argument(1); // type oop
8444 Node* ofs = argument(2); // type int
8445
8446 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8447 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8448 // failed array check
8449 return false;
8450 }
8451 // Figure out the size and type of the elements we will be copying.
8452 BasicType src_elem = src_type->elem()->array_element_basic_type();
8453 if (src_elem != T_BYTE) {
8454 return false;
8455 }
8456 // 'src_start' points to src array + offset
8457 src = must_be_not_null(src, true);
8458 Node* src_start = array_element_address(src, ofs, src_elem);
8459 Node* state = nullptr;
8460 Node* block_size = nullptr;
8461 address stubAddr;
8462 const char *stubName;
8463
8464 switch(id) {
8465 case vmIntrinsics::_md5_implCompress:
8466 assert(UseMD5Intrinsics, "need MD5 instruction support");
8467 state = get_state_from_digest_object(digestBase_obj, T_INT);
8468 stubAddr = StubRoutines::md5_implCompress();
8469 stubName = "md5_implCompress";
8470 break;
8471 case vmIntrinsics::_sha_implCompress:
8472 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8473 state = get_state_from_digest_object(digestBase_obj, T_INT);
8474 stubAddr = StubRoutines::sha1_implCompress();
8475 stubName = "sha1_implCompress";
8476 break;
8477 case vmIntrinsics::_sha2_implCompress:
8478 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8479 state = get_state_from_digest_object(digestBase_obj, T_INT);
8480 stubAddr = StubRoutines::sha256_implCompress();
8481 stubName = "sha256_implCompress";
8482 break;
8483 case vmIntrinsics::_sha5_implCompress:
8484 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8485 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8486 stubAddr = StubRoutines::sha512_implCompress();
8487 stubName = "sha512_implCompress";
8488 break;
8489 case vmIntrinsics::_sha3_implCompress:
8490 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8491 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8492 stubAddr = StubRoutines::sha3_implCompress();
8493 stubName = "sha3_implCompress";
8494 block_size = get_block_size_from_digest_object(digestBase_obj);
8495 if (block_size == nullptr) return false;
8496 break;
8497 default:
8498 fatal_unexpected_iid(id);
8499 return false;
8500 }
8501 if (state == nullptr) return false;
8502
8503 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8504 if (stubAddr == nullptr) return false;
8505
8506 // Call the stub.
8507 Node* call;
8508 if (block_size == nullptr) {
8509 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8510 stubAddr, stubName, TypePtr::BOTTOM,
8511 src_start, state);
8512 } else {
8513 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8514 stubAddr, stubName, TypePtr::BOTTOM,
8515 src_start, state, block_size);
8516 }
8517
8518 return true;
8519 }
8520
8521 //------------------------------inline_double_keccak
8522 bool LibraryCallKit::inline_double_keccak() {
8523 address stubAddr;
8524 const char *stubName;
8525 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8526 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8527
8528 stubAddr = StubRoutines::double_keccak();
8529 stubName = "double_keccak";
8530 if (!stubAddr) return false;
8531
8532 Node* status0 = argument(0);
8533 Node* status1 = argument(1);
8534
8535 status0 = must_be_not_null(status0, true);
8536 status1 = must_be_not_null(status1, true);
8537
8538 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8539 assert(status0_start, "status0 is null");
8540 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8541 assert(status1_start, "status1 is null");
8542 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8543 OptoRuntime::double_keccak_Type(),
8544 stubAddr, stubName, TypePtr::BOTTOM,
8545 status0_start, status1_start);
8546 // return an int
8547 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8548 set_result(retvalue);
8549 return true;
8550 }
8551
8552
8553 //------------------------------inline_digestBase_implCompressMB-----------------------
8554 //
8555 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8556 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8557 //
8558 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8559 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8560 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8561 assert((uint)predicate < 5, "sanity");
8562 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8563
8564 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8565 Node* src = argument(1); // byte[] array
8566 Node* ofs = argument(2); // type int
8567 Node* limit = argument(3); // type int
8568
8569 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8570 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8571 // failed array check
8572 return false;
8573 }
8574 // Figure out the size and type of the elements we will be copying.
8575 BasicType src_elem = src_type->elem()->array_element_basic_type();
8576 if (src_elem != T_BYTE) {
8577 return false;
8578 }
8579 // 'src_start' points to src array + offset
8580 src = must_be_not_null(src, false);
8581 Node* src_start = array_element_address(src, ofs, src_elem);
8582
8583 const char* klass_digestBase_name = nullptr;
8584 const char* stub_name = nullptr;
8585 address stub_addr = nullptr;
8586 BasicType elem_type = T_INT;
8587
8588 switch (predicate) {
8589 case 0:
8590 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8591 klass_digestBase_name = "sun/security/provider/MD5";
8592 stub_name = "md5_implCompressMB";
8593 stub_addr = StubRoutines::md5_implCompressMB();
8594 }
8595 break;
8596 case 1:
8597 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8598 klass_digestBase_name = "sun/security/provider/SHA";
8599 stub_name = "sha1_implCompressMB";
8600 stub_addr = StubRoutines::sha1_implCompressMB();
8601 }
8602 break;
8603 case 2:
8604 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8605 klass_digestBase_name = "sun/security/provider/SHA2";
8606 stub_name = "sha256_implCompressMB";
8607 stub_addr = StubRoutines::sha256_implCompressMB();
8608 }
8609 break;
8610 case 3:
8611 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8612 klass_digestBase_name = "sun/security/provider/SHA5";
8613 stub_name = "sha512_implCompressMB";
8614 stub_addr = StubRoutines::sha512_implCompressMB();
8615 elem_type = T_LONG;
8616 }
8617 break;
8618 case 4:
8619 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8620 klass_digestBase_name = "sun/security/provider/SHA3";
8621 stub_name = "sha3_implCompressMB";
8622 stub_addr = StubRoutines::sha3_implCompressMB();
8623 elem_type = T_LONG;
8624 }
8625 break;
8626 default:
8627 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8628 }
8629 if (klass_digestBase_name != nullptr) {
8630 assert(stub_addr != nullptr, "Stub is generated");
8631 if (stub_addr == nullptr) return false;
8632
8633 // get DigestBase klass to lookup for SHA klass
8634 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8635 assert(tinst != nullptr, "digestBase_obj is not instance???");
8636 assert(tinst->is_loaded(), "DigestBase is not loaded");
8637
8638 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8639 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8640 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8641 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8642 }
8643 return false;
8644 }
8645
8646 //------------------------------inline_digestBase_implCompressMB-----------------------
8647 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8648 BasicType elem_type, address stubAddr, const char *stubName,
8649 Node* src_start, Node* ofs, Node* limit) {
8650 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8651 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8652 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8653 digest_obj = _gvn.transform(digest_obj);
8654
8655 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8656 if (state == nullptr) return false;
8657
8658 Node* block_size = nullptr;
8659 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8660 block_size = get_block_size_from_digest_object(digest_obj);
8661 if (block_size == nullptr) return false;
8662 }
8663
8664 // Call the stub.
8665 Node* call;
8666 if (block_size == nullptr) {
8667 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8668 OptoRuntime::digestBase_implCompressMB_Type(false),
8669 stubAddr, stubName, TypePtr::BOTTOM,
8670 src_start, state, ofs, limit);
8671 } else {
8672 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8673 OptoRuntime::digestBase_implCompressMB_Type(true),
8674 stubAddr, stubName, TypePtr::BOTTOM,
8675 src_start, state, block_size, ofs, limit);
8676 }
8677
8678 // return ofs (int)
8679 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8680 set_result(result);
8681
8682 return true;
8683 }
8684
8685 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8686 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8687 assert(UseAES, "need AES instruction support");
8688 address stubAddr = nullptr;
8689 const char *stubName = nullptr;
8690 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8691 stubName = "galoisCounterMode_AESCrypt";
8692
8693 if (stubAddr == nullptr) return false;
8694
8695 Node* in = argument(0);
8696 Node* inOfs = argument(1);
8697 Node* len = argument(2);
8698 Node* ct = argument(3);
8699 Node* ctOfs = argument(4);
8700 Node* out = argument(5);
8701 Node* outOfs = argument(6);
8702 Node* gctr_object = argument(7);
8703 Node* ghash_object = argument(8);
8704
8705 // (1) in, ct and out are arrays.
8706 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8707 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8708 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8709 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8710 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8711 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8712
8713 // checks are the responsibility of the caller
8714 Node* in_start = in;
8715 Node* ct_start = ct;
8716 Node* out_start = out;
8717 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8718 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8719 in_start = array_element_address(in, inOfs, T_BYTE);
8720 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8721 out_start = array_element_address(out, outOfs, T_BYTE);
8722 }
8723
8724 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8725 // (because of the predicated logic executed earlier).
8726 // so we cast it here safely.
8727 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8728 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8729 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8730 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8731 Node* state = load_field_from_object(ghash_object, "state", "[J");
8732
8733 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8734 return false;
8735 }
8736 // cast it to what we know it will be at runtime
8737 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8738 assert(tinst != nullptr, "GCTR obj is null");
8739 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8740 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8741 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8742 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8743 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8744 const TypeOopPtr* xtype = aklass->as_instance_type();
8745 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8746 aescrypt_object = _gvn.transform(aescrypt_object);
8747 // we need to get the start of the aescrypt_object's expanded key array
8748 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8749 if (k_start == nullptr) return false;
8750 // similarly, get the start address of the r vector
8751 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8752 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8753 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8754
8755
8756 // Call the stub, passing params
8757 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8758 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8759 stubAddr, stubName, TypePtr::BOTTOM,
8760 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8761
8762 // return cipher length (int)
8763 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8764 set_result(retvalue);
8765
8766 return true;
8767 }
8768
8769 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8770 // Return node representing slow path of predicate check.
8771 // the pseudo code we want to emulate with this predicate is:
8772 // for encryption:
8773 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8774 // for decryption:
8775 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8776 // note cipher==plain is more conservative than the original java code but that's OK
8777 //
8778
8779 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8780 // The receiver was checked for null already.
8781 Node* objGCTR = argument(7);
8782 // Load embeddedCipher field of GCTR object.
8783 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8784 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8785
8786 // get AESCrypt klass for instanceOf check
8787 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8788 // will have same classloader as CipherBlockChaining object
8789 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8790 assert(tinst != nullptr, "GCTR obj is null");
8791 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8792
8793 // we want to do an instanceof comparison against the AESCrypt class
8794 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8795 if (!klass_AESCrypt->is_loaded()) {
8796 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8797 Node* ctrl = control();
8798 set_control(top()); // no regular fast path
8799 return ctrl;
8800 }
8801
8802 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8803 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8804 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8805 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8806 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8807
8808 return instof_false; // even if it is null
8809 }
8810
8811 //------------------------------get_state_from_digest_object-----------------------
8812 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8813 const char* state_type;
8814 switch (elem_type) {
8815 case T_BYTE: state_type = "[B"; break;
8816 case T_INT: state_type = "[I"; break;
8817 case T_LONG: state_type = "[J"; break;
8818 default: ShouldNotReachHere();
8819 }
8820 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8821 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8822 if (digest_state == nullptr) return (Node *) nullptr;
8823
8824 // now have the array, need to get the start address of the state array
8825 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8826 return state;
8827 }
8828
8829 //------------------------------get_block_size_from_sha3_object----------------------------------
8830 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8831 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8832 assert (block_size != nullptr, "sanity");
8833 return block_size;
8834 }
8835
8836 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8837 // Return node representing slow path of predicate check.
8838 // the pseudo code we want to emulate with this predicate is:
8839 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8840 //
8841 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8842 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8843 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8844 assert((uint)predicate < 5, "sanity");
8845
8846 // The receiver was checked for null already.
8847 Node* digestBaseObj = argument(0);
8848
8849 // get DigestBase klass for instanceOf check
8850 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8851 assert(tinst != nullptr, "digestBaseObj is null");
8852 assert(tinst->is_loaded(), "DigestBase is not loaded");
8853
8854 const char* klass_name = nullptr;
8855 switch (predicate) {
8856 case 0:
8857 if (UseMD5Intrinsics) {
8858 // we want to do an instanceof comparison against the MD5 class
8859 klass_name = "sun/security/provider/MD5";
8860 }
8861 break;
8862 case 1:
8863 if (UseSHA1Intrinsics) {
8864 // we want to do an instanceof comparison against the SHA class
8865 klass_name = "sun/security/provider/SHA";
8866 }
8867 break;
8868 case 2:
8869 if (UseSHA256Intrinsics) {
8870 // we want to do an instanceof comparison against the SHA2 class
8871 klass_name = "sun/security/provider/SHA2";
8872 }
8873 break;
8874 case 3:
8875 if (UseSHA512Intrinsics) {
8876 // we want to do an instanceof comparison against the SHA5 class
8877 klass_name = "sun/security/provider/SHA5";
8878 }
8879 break;
8880 case 4:
8881 if (UseSHA3Intrinsics) {
8882 // we want to do an instanceof comparison against the SHA3 class
8883 klass_name = "sun/security/provider/SHA3";
8884 }
8885 break;
8886 default:
8887 fatal("unknown SHA intrinsic predicate: %d", predicate);
8888 }
8889
8890 ciKlass* klass = nullptr;
8891 if (klass_name != nullptr) {
8892 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8893 }
8894 if ((klass == nullptr) || !klass->is_loaded()) {
8895 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8896 Node* ctrl = control();
8897 set_control(top()); // no intrinsic path
8898 return ctrl;
8899 }
8900 ciInstanceKlass* instklass = klass->as_instance_klass();
8901
8902 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8903 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8904 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8905 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8906
8907 return instof_false; // even if it is null
8908 }
8909
8910 //-------------inline_fma-----------------------------------
8911 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8912 Node *a = nullptr;
8913 Node *b = nullptr;
8914 Node *c = nullptr;
8915 Node* result = nullptr;
8916 switch (id) {
8917 case vmIntrinsics::_fmaD:
8918 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8919 // no receiver since it is static method
8920 a = argument(0);
8921 b = argument(2);
8922 c = argument(4);
8923 result = _gvn.transform(new FmaDNode(a, b, c));
8924 break;
8925 case vmIntrinsics::_fmaF:
8926 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8927 a = argument(0);
8928 b = argument(1);
8929 c = argument(2);
8930 result = _gvn.transform(new FmaFNode(a, b, c));
8931 break;
8932 default:
8933 fatal_unexpected_iid(id); break;
8934 }
8935 set_result(result);
8936 return true;
8937 }
8938
8939 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8940 // argument(0) is receiver
8941 Node* codePoint = argument(1);
8942 Node* n = nullptr;
8943
8944 switch (id) {
8945 case vmIntrinsics::_isDigit :
8946 n = new DigitNode(control(), codePoint);
8947 break;
8948 case vmIntrinsics::_isLowerCase :
8949 n = new LowerCaseNode(control(), codePoint);
8950 break;
8951 case vmIntrinsics::_isUpperCase :
8952 n = new UpperCaseNode(control(), codePoint);
8953 break;
8954 case vmIntrinsics::_isWhitespace :
8955 n = new WhitespaceNode(control(), codePoint);
8956 break;
8957 default:
8958 fatal_unexpected_iid(id);
8959 }
8960
8961 set_result(_gvn.transform(n));
8962 return true;
8963 }
8964
8965 bool LibraryCallKit::inline_profileBoolean() {
8966 Node* counts = argument(1);
8967 const TypeAryPtr* ary = nullptr;
8968 ciArray* aobj = nullptr;
8969 if (counts->is_Con()
8970 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8971 && (aobj = ary->const_oop()->as_array()) != nullptr
8972 && (aobj->length() == 2)) {
8973 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8974 jint false_cnt = aobj->element_value(0).as_int();
8975 jint true_cnt = aobj->element_value(1).as_int();
8976
8977 if (C->log() != nullptr) {
8978 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8979 false_cnt, true_cnt);
8980 }
8981
8982 if (false_cnt + true_cnt == 0) {
8983 // According to profile, never executed.
8984 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8985 Deoptimization::Action_reinterpret);
8986 return true;
8987 }
8988
8989 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8990 // is a number of each value occurrences.
8991 Node* result = argument(0);
8992 if (false_cnt == 0 || true_cnt == 0) {
8993 // According to profile, one value has been never seen.
8994 int expected_val = (false_cnt == 0) ? 1 : 0;
8995
8996 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8997 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8998
8999 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9000 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9001 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9002
9003 { // Slow path: uncommon trap for never seen value and then reexecute
9004 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9005 // the value has been seen at least once.
9006 PreserveJVMState pjvms(this);
9007 PreserveReexecuteState preexecs(this);
9008 jvms()->set_should_reexecute(true);
9009
9010 set_control(slow_path);
9011 set_i_o(i_o());
9012
9013 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9014 Deoptimization::Action_reinterpret);
9015 }
9016 // The guard for never seen value enables sharpening of the result and
9017 // returning a constant. It allows to eliminate branches on the same value
9018 // later on.
9019 set_control(fast_path);
9020 result = intcon(expected_val);
9021 }
9022 // Stop profiling.
9023 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9024 // By replacing method body with profile data (represented as ProfileBooleanNode
9025 // on IR level) we effectively disable profiling.
9026 // It enables full speed execution once optimized code is generated.
9027 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9028 C->record_for_igvn(profile);
9029 set_result(profile);
9030 return true;
9031 } else {
9032 // Continue profiling.
9033 // Profile data isn't available at the moment. So, execute method's bytecode version.
9034 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9035 // is compiled and counters aren't available since corresponding MethodHandle
9036 // isn't a compile-time constant.
9037 return false;
9038 }
9039 }
9040
9041 bool LibraryCallKit::inline_isCompileConstant() {
9042 Node* n = argument(0);
9043 set_result(n->is_Con() ? intcon(1) : intcon(0));
9044 return true;
9045 }
9046
9047 //------------------------------- inline_getObjectSize --------------------------------------
9048 //
9049 // Calculate the runtime size of the object/array.
9050 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9051 //
9052 bool LibraryCallKit::inline_getObjectSize() {
9053 Node* obj = argument(3);
9054 Node* klass_node = load_object_klass(obj);
9055
9056 jint layout_con = Klass::_lh_neutral_value;
9057 Node* layout_val = get_layout_helper(klass_node, layout_con);
9058 int layout_is_con = (layout_val == nullptr);
9059
9060 if (layout_is_con) {
9061 // Layout helper is constant, can figure out things at compile time.
9062
9063 if (Klass::layout_helper_is_instance(layout_con)) {
9064 // Instance case: layout_con contains the size itself.
9065 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9066 set_result(size);
9067 } else {
9068 // Array case: size is round(header + element_size*arraylength).
9069 // Since arraylength is different for every array instance, we have to
9070 // compute the whole thing at runtime.
9071
9072 Node* arr_length = load_array_length(obj);
9073
9074 int round_mask = MinObjAlignmentInBytes - 1;
9075 int hsize = Klass::layout_helper_header_size(layout_con);
9076 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9077
9078 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9079 round_mask = 0; // strength-reduce it if it goes away completely
9080 }
9081 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9082 Node* header_size = intcon(hsize + round_mask);
9083
9084 Node* lengthx = ConvI2X(arr_length);
9085 Node* headerx = ConvI2X(header_size);
9086
9087 Node* abody = lengthx;
9088 if (eshift != 0) {
9089 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9090 }
9091 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9092 if (round_mask != 0) {
9093 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9094 }
9095 size = ConvX2L(size);
9096 set_result(size);
9097 }
9098 } else {
9099 // Layout helper is not constant, need to test for array-ness at runtime.
9100
9101 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9102 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9103 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9104 record_for_igvn(result_reg);
9105
9106 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9107 if (array_ctl != nullptr) {
9108 // Array case: size is round(header + element_size*arraylength).
9109 // Since arraylength is different for every array instance, we have to
9110 // compute the whole thing at runtime.
9111
9112 PreserveJVMState pjvms(this);
9113 set_control(array_ctl);
9114 Node* arr_length = load_array_length(obj);
9115
9116 int round_mask = MinObjAlignmentInBytes - 1;
9117 Node* mask = intcon(round_mask);
9118
9119 Node* hss = intcon(Klass::_lh_header_size_shift);
9120 Node* hsm = intcon(Klass::_lh_header_size_mask);
9121 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9122 header_size = _gvn.transform(new AndINode(header_size, hsm));
9123 header_size = _gvn.transform(new AddINode(header_size, mask));
9124
9125 // There is no need to mask or shift this value.
9126 // The semantics of LShiftINode include an implicit mask to 0x1F.
9127 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9128 Node* elem_shift = layout_val;
9129
9130 Node* lengthx = ConvI2X(arr_length);
9131 Node* headerx = ConvI2X(header_size);
9132
9133 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9134 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9135 if (round_mask != 0) {
9136 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9137 }
9138 size = ConvX2L(size);
9139
9140 result_reg->init_req(_array_path, control());
9141 result_val->init_req(_array_path, size);
9142 }
9143
9144 if (!stopped()) {
9145 // Instance case: the layout helper gives us instance size almost directly,
9146 // but we need to mask out the _lh_instance_slow_path_bit.
9147 Node* size = ConvI2X(layout_val);
9148 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9149 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9150 size = _gvn.transform(new AndXNode(size, mask));
9151 size = ConvX2L(size);
9152
9153 result_reg->init_req(_instance_path, control());
9154 result_val->init_req(_instance_path, size);
9155 }
9156
9157 set_result(result_reg, result_val);
9158 }
9159
9160 return true;
9161 }
9162
9163 //------------------------------- inline_blackhole --------------------------------------
9164 //
9165 // Make sure all arguments to this node are alive.
9166 // This matches methods that were requested to be blackholed through compile commands.
9167 //
9168 bool LibraryCallKit::inline_blackhole() {
9169 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9170 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9171 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9172
9173 // Blackhole node pinches only the control, not memory. This allows
9174 // the blackhole to be pinned in the loop that computes blackholed
9175 // values, but have no other side effects, like breaking the optimizations
9176 // across the blackhole.
9177
9178 Node* bh = _gvn.transform(new BlackholeNode(control()));
9179 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9180
9181 // Bind call arguments as blackhole arguments to keep them alive
9182 uint nargs = callee()->arg_size();
9183 for (uint i = 0; i < nargs; i++) {
9184 bh->add_req(argument(i));
9185 }
9186
9187 return true;
9188 }
9189
9190 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9191 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9192 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9193 return nullptr; // box klass is not Float16
9194 }
9195
9196 // Null check; get notnull casted pointer
9197 Node* null_ctl = top();
9198 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9199 // If not_null_box is dead, only null-path is taken
9200 if (stopped()) {
9201 set_control(null_ctl);
9202 return nullptr;
9203 }
9204 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9205 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9206 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9207 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9208 }
9209
9210 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9211 PreserveReexecuteState preexecs(this);
9212 jvms()->set_should_reexecute(true);
9213
9214 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9215 Node* klass_node = makecon(klass_type);
9216 Node* box = new_instance(klass_node);
9217
9218 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9219 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9220
9221 Node* field_store = _gvn.transform(access_store_at(box,
9222 value_field,
9223 value_adr_type,
9224 value,
9225 TypeInt::SHORT,
9226 T_SHORT,
9227 IN_HEAP));
9228 set_memory(field_store, value_adr_type);
9229 return box;
9230 }
9231
9232 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9233 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9234 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9235 return false;
9236 }
9237
9238 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9239 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9240 return false;
9241 }
9242
9243 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9244 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9245 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9246 ciSymbols::short_signature(),
9247 false);
9248 assert(field != nullptr, "");
9249
9250 // Transformed nodes
9251 Node* fld1 = nullptr;
9252 Node* fld2 = nullptr;
9253 Node* fld3 = nullptr;
9254 switch(num_args) {
9255 case 3:
9256 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9257 if (fld3 == nullptr) {
9258 return false;
9259 }
9260 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9261 // fall-through
9262 case 2:
9263 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9264 if (fld2 == nullptr) {
9265 return false;
9266 }
9267 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9268 // fall-through
9269 case 1:
9270 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9271 if (fld1 == nullptr) {
9272 return false;
9273 }
9274 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9275 break;
9276 default: fatal("Unsupported number of arguments %d", num_args);
9277 }
9278
9279 Node* result = nullptr;
9280 switch (id) {
9281 // Unary operations
9282 case vmIntrinsics::_sqrt_float16:
9283 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9284 break;
9285 // Ternary operations
9286 case vmIntrinsics::_fma_float16:
9287 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9288 break;
9289 default:
9290 fatal_unexpected_iid(id);
9291 break;
9292 }
9293 result = _gvn.transform(new ReinterpretHF2SNode(result));
9294 set_result(box_fp16_value(float16_box_type, field, result));
9295 return true;
9296 }
9297