1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciUtilities.inline.hpp"
27 #include "ci/ciSymbols.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/runtime.hpp"
51 #include "opto/rootnode.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*)CompilerThread::current()->compiler();
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 if (jvms->has_method()) {
139 // Not a root compile.
140 const char* msg;
141 if (callee->intrinsic_candidate()) {
142 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
143 } else {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
145 : "failed to inline (intrinsic), method not annotated";
146 }
147 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
148 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
149 } else {
150 // Root compile
151 ResourceMark rm;
152 stringStream msg_stream;
153 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
154 vmIntrinsics::name_at(intrinsic_id()),
155 is_virtual() ? " (virtual)" : "", bci);
156 const char *msg = msg_stream.freeze();
157 log_debug(jit, inlining)("%s", msg);
158 if (C->print_intrinsics() || C->print_inlining()) {
159 tty->print("%s", msg);
160 }
161 }
162 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
163
164 return nullptr;
165 }
166
167 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
168 LibraryCallKit kit(jvms, this);
169 Compile* C = kit.C;
170 int nodes = C->unique();
171 _last_predicate = predicate;
172 #ifndef PRODUCT
173 assert(is_predicated() && predicate < predicates_count(), "sanity");
174 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
175 char buf[1000];
176 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
177 tty->print_cr("Predicate for intrinsic %s", str);
178 }
179 #endif
180 ciMethod* callee = kit.callee();
181 const int bci = kit.bci();
182
183 Node* slow_ctl = kit.try_to_predicate(predicate);
184 if (!kit.failing()) {
185 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
186 : "(intrinsic, predicate)";
187 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
188 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
189
190 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
191 if (C->log()) {
192 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
193 vmIntrinsics::name_at(intrinsic_id()),
194 (is_virtual() ? " virtual='1'" : ""),
195 C->unique() - nodes);
196 }
197 return slow_ctl; // Could be null if the check folds.
198 }
199
200 // The intrinsic bailed out
201 if (jvms->has_method()) {
202 // Not a root compile.
203 const char* msg = "failed to generate predicate for intrinsic";
204 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
205 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
206 } else {
207 // Root compile
208 ResourceMark rm;
209 stringStream msg_stream;
210 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
211 vmIntrinsics::name_at(intrinsic_id()),
212 is_virtual() ? " (virtual)" : "", bci);
213 const char *msg = msg_stream.freeze();
214 log_debug(jit, inlining)("%s", msg);
215 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
216 }
217 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
218 return nullptr;
219 }
220
221 bool LibraryCallKit::try_to_inline(int predicate) {
222 // Handle symbolic names for otherwise undistinguished boolean switches:
223 const bool is_store = true;
224 const bool is_compress = true;
225 const bool is_static = true;
226 const bool is_volatile = true;
227
228 if (!jvms()->has_method()) {
229 // Root JVMState has a null method.
230 assert(map()->memory()->Opcode() == Op_Parm, "");
231 // Insert the memory aliasing node
232 set_all_memory(reset_memory());
233 }
234 assert(merged_memory(), "");
235
236 switch (intrinsic_id()) {
237 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
238 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
239 case vmIntrinsics::_getClass: return inline_native_getClass();
240
241 case vmIntrinsics::_ceil:
242 case vmIntrinsics::_floor:
243 case vmIntrinsics::_rint:
244 case vmIntrinsics::_dsin:
245 case vmIntrinsics::_dcos:
246 case vmIntrinsics::_dtan:
247 case vmIntrinsics::_dtanh:
248 case vmIntrinsics::_dabs:
249 case vmIntrinsics::_fabs:
250 case vmIntrinsics::_iabs:
251 case vmIntrinsics::_labs:
252 case vmIntrinsics::_datan2:
253 case vmIntrinsics::_dsqrt:
254 case vmIntrinsics::_dsqrt_strict:
255 case vmIntrinsics::_dexp:
256 case vmIntrinsics::_dlog:
257 case vmIntrinsics::_dlog10:
258 case vmIntrinsics::_dpow:
259 case vmIntrinsics::_dcopySign:
260 case vmIntrinsics::_fcopySign:
261 case vmIntrinsics::_dsignum:
262 case vmIntrinsics::_roundF:
263 case vmIntrinsics::_roundD:
264 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
265
266 case vmIntrinsics::_notify:
267 case vmIntrinsics::_notifyAll:
268 return inline_notify(intrinsic_id());
269
270 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
271 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
272 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
273 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
274 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
275 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
276 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
277 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
278 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
279 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
280 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
281 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
282 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
283 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
284
285 case vmIntrinsics::_arraycopy: return inline_arraycopy();
286
287 case vmIntrinsics::_arraySort: return inline_array_sort();
288 case vmIntrinsics::_arrayPartition: return inline_array_partition();
289
290 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
291 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
292 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
293 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
294
295 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
296 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
297 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
298 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
302 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
303
304 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
305
306 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
307
308 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
309 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
310 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
311 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
312
313 case vmIntrinsics::_compressStringC:
314 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
315 case vmIntrinsics::_inflateStringC:
316 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
317
318 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
319 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
320 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
321 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
322 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
323 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
324 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
325 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
326 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
327
328 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
339 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
340 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
341 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
342 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
343 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
344 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
345 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
346 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
347
348 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
359 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
360 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
361 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
362
363 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
369 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
370 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
371 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
372 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
373 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
374 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
375 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
376 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
377
378 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
379 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
380 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
381 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
382 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
383 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
384 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
385 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
386 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
387
388 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
389 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
390 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
391 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
392 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
393 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
394 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
395 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
396 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
397
398 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
409 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
413
414 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
415 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
416 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
417 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
418 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
434
435 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
436 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
437 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
438 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
439 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
440 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
441 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
450
451 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
452 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
455
456 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
457 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
461
462 case vmIntrinsics::_loadFence:
463 case vmIntrinsics::_storeFence:
464 case vmIntrinsics::_storeStoreFence:
465 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
466
467 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
468
469 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
470 case vmIntrinsics::_currentThread: return inline_native_currentThread();
471 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
472
473 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
474 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
475
476 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
477 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
478
479 #if INCLUDE_JVMTI
480 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
481 "notifyJvmtiStart", true, false);
482 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
483 "notifyJvmtiEnd", false, true);
484 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
485 "notifyJvmtiMount", false, false);
486 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
487 "notifyJvmtiUnmount", false, false);
488 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
489 #endif
490
491 #ifdef JFR_HAVE_INTRINSICS
492 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
493 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
494 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
495 #endif
496 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
497 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
498 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
499 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
500 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
501 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
502 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
503 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
504 case vmIntrinsics::_getLength: return inline_native_getLength();
505 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
506 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
507 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
508 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
509 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
510 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
511 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
512
513 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
514 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
515
516 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
517
518 case vmIntrinsics::_isInstance:
519 case vmIntrinsics::_isHidden:
520 case vmIntrinsics::_getSuperclass:
521 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
522
523 case vmIntrinsics::_floatToRawIntBits:
524 case vmIntrinsics::_floatToIntBits:
525 case vmIntrinsics::_intBitsToFloat:
526 case vmIntrinsics::_doubleToRawLongBits:
527 case vmIntrinsics::_doubleToLongBits:
528 case vmIntrinsics::_longBitsToDouble:
529 case vmIntrinsics::_floatToFloat16:
530 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
531 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
532 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
533 case vmIntrinsics::_floatIsFinite:
534 case vmIntrinsics::_floatIsInfinite:
535 case vmIntrinsics::_doubleIsFinite:
536 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
537
538 case vmIntrinsics::_numberOfLeadingZeros_i:
539 case vmIntrinsics::_numberOfLeadingZeros_l:
540 case vmIntrinsics::_numberOfTrailingZeros_i:
541 case vmIntrinsics::_numberOfTrailingZeros_l:
542 case vmIntrinsics::_bitCount_i:
543 case vmIntrinsics::_bitCount_l:
544 case vmIntrinsics::_reverse_i:
545 case vmIntrinsics::_reverse_l:
546 case vmIntrinsics::_reverseBytes_i:
547 case vmIntrinsics::_reverseBytes_l:
548 case vmIntrinsics::_reverseBytes_s:
549 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
550
551 case vmIntrinsics::_compress_i:
552 case vmIntrinsics::_compress_l:
553 case vmIntrinsics::_expand_i:
554 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
555
556 case vmIntrinsics::_compareUnsigned_i:
557 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
558
559 case vmIntrinsics::_divideUnsigned_i:
560 case vmIntrinsics::_divideUnsigned_l:
561 case vmIntrinsics::_remainderUnsigned_i:
562 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
563
564 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
565
566 case vmIntrinsics::_Reference_get: return inline_reference_get();
567 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
568 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
569 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
570 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
571
572 case vmIntrinsics::_Class_cast: return inline_Class_cast();
573
574 case vmIntrinsics::_aescrypt_encryptBlock:
575 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
576
577 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
578 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
579 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
580
581 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
582 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
583 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_counterMode_AESCrypt:
586 return inline_counterMode_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_galoisCounterMode_AESCrypt:
589 return inline_galoisCounterMode_AESCrypt();
590
591 case vmIntrinsics::_md5_implCompress:
592 case vmIntrinsics::_sha_implCompress:
593 case vmIntrinsics::_sha2_implCompress:
594 case vmIntrinsics::_sha5_implCompress:
595 case vmIntrinsics::_sha3_implCompress:
596 return inline_digestBase_implCompress(intrinsic_id());
597 case vmIntrinsics::_double_keccak:
598 return inline_double_keccak();
599
600 case vmIntrinsics::_digestBase_implCompressMB:
601 return inline_digestBase_implCompressMB(predicate);
602
603 case vmIntrinsics::_multiplyToLen:
604 return inline_multiplyToLen();
605
606 case vmIntrinsics::_squareToLen:
607 return inline_squareToLen();
608
609 case vmIntrinsics::_mulAdd:
610 return inline_mulAdd();
611
612 case vmIntrinsics::_montgomeryMultiply:
613 return inline_montgomeryMultiply();
614 case vmIntrinsics::_montgomerySquare:
615 return inline_montgomerySquare();
616
617 case vmIntrinsics::_bigIntegerRightShiftWorker:
618 return inline_bigIntegerShift(true);
619 case vmIntrinsics::_bigIntegerLeftShiftWorker:
620 return inline_bigIntegerShift(false);
621
622 case vmIntrinsics::_vectorizedMismatch:
623 return inline_vectorizedMismatch();
624
625 case vmIntrinsics::_ghash_processBlocks:
626 return inline_ghash_processBlocks();
627 case vmIntrinsics::_chacha20Block:
628 return inline_chacha20Block();
629 case vmIntrinsics::_dilithiumAlmostNtt:
630 return inline_dilithiumAlmostNtt();
631 case vmIntrinsics::_dilithiumAlmostInverseNtt:
632 return inline_dilithiumAlmostInverseNtt();
633 case vmIntrinsics::_dilithiumNttMult:
634 return inline_dilithiumNttMult();
635 case vmIntrinsics::_dilithiumMontMulByConstant:
636 return inline_dilithiumMontMulByConstant();
637 case vmIntrinsics::_dilithiumDecomposePoly:
638 return inline_dilithiumDecomposePoly();
639 case vmIntrinsics::_base64_encodeBlock:
640 return inline_base64_encodeBlock();
641 case vmIntrinsics::_base64_decodeBlock:
642 return inline_base64_decodeBlock();
643 case vmIntrinsics::_poly1305_processBlocks:
644 return inline_poly1305_processBlocks();
645 case vmIntrinsics::_intpoly_montgomeryMult_P256:
646 return inline_intpoly_montgomeryMult_P256();
647 case vmIntrinsics::_intpoly_assign:
648 return inline_intpoly_assign();
649 case vmIntrinsics::_encodeISOArray:
650 case vmIntrinsics::_encodeByteISOArray:
651 return inline_encodeISOArray(false);
652 case vmIntrinsics::_encodeAsciiArray:
653 return inline_encodeISOArray(true);
654
655 case vmIntrinsics::_updateCRC32:
656 return inline_updateCRC32();
657 case vmIntrinsics::_updateBytesCRC32:
658 return inline_updateBytesCRC32();
659 case vmIntrinsics::_updateByteBufferCRC32:
660 return inline_updateByteBufferCRC32();
661
662 case vmIntrinsics::_updateBytesCRC32C:
663 return inline_updateBytesCRC32C();
664 case vmIntrinsics::_updateDirectByteBufferCRC32C:
665 return inline_updateDirectByteBufferCRC32C();
666
667 case vmIntrinsics::_updateBytesAdler32:
668 return inline_updateBytesAdler32();
669 case vmIntrinsics::_updateByteBufferAdler32:
670 return inline_updateByteBufferAdler32();
671
672 case vmIntrinsics::_profileBoolean:
673 return inline_profileBoolean();
674 case vmIntrinsics::_isCompileConstant:
675 return inline_isCompileConstant();
676
677 case vmIntrinsics::_countPositives:
678 return inline_countPositives();
679
680 case vmIntrinsics::_fmaD:
681 case vmIntrinsics::_fmaF:
682 return inline_fma(intrinsic_id());
683
684 case vmIntrinsics::_isDigit:
685 case vmIntrinsics::_isLowerCase:
686 case vmIntrinsics::_isUpperCase:
687 case vmIntrinsics::_isWhitespace:
688 return inline_character_compare(intrinsic_id());
689
690 case vmIntrinsics::_min:
691 case vmIntrinsics::_max:
692 case vmIntrinsics::_min_strict:
693 case vmIntrinsics::_max_strict:
694 case vmIntrinsics::_minL:
695 case vmIntrinsics::_maxL:
696 case vmIntrinsics::_minF:
697 case vmIntrinsics::_maxF:
698 case vmIntrinsics::_minD:
699 case vmIntrinsics::_maxD:
700 case vmIntrinsics::_minF_strict:
701 case vmIntrinsics::_maxF_strict:
702 case vmIntrinsics::_minD_strict:
703 case vmIntrinsics::_maxD_strict:
704 return inline_min_max(intrinsic_id());
705
706 case vmIntrinsics::_VectorUnaryOp:
707 return inline_vector_nary_operation(1);
708 case vmIntrinsics::_VectorBinaryOp:
709 return inline_vector_nary_operation(2);
710 case vmIntrinsics::_VectorTernaryOp:
711 return inline_vector_nary_operation(3);
712 case vmIntrinsics::_VectorFromBitsCoerced:
713 return inline_vector_frombits_coerced();
714 case vmIntrinsics::_VectorMaskOp:
715 return inline_vector_mask_operation();
716 case vmIntrinsics::_VectorLoadOp:
717 return inline_vector_mem_operation(/*is_store=*/false);
718 case vmIntrinsics::_VectorLoadMaskedOp:
719 return inline_vector_mem_masked_operation(/*is_store*/false);
720 case vmIntrinsics::_VectorStoreOp:
721 return inline_vector_mem_operation(/*is_store=*/true);
722 case vmIntrinsics::_VectorStoreMaskedOp:
723 return inline_vector_mem_masked_operation(/*is_store=*/true);
724 case vmIntrinsics::_VectorGatherOp:
725 return inline_vector_gather_scatter(/*is_scatter*/ false);
726 case vmIntrinsics::_VectorScatterOp:
727 return inline_vector_gather_scatter(/*is_scatter*/ true);
728 case vmIntrinsics::_VectorReductionCoerced:
729 return inline_vector_reduction();
730 case vmIntrinsics::_VectorTest:
731 return inline_vector_test();
732 case vmIntrinsics::_VectorBlend:
733 return inline_vector_blend();
734 case vmIntrinsics::_VectorRearrange:
735 return inline_vector_rearrange();
736 case vmIntrinsics::_VectorSelectFrom:
737 return inline_vector_select_from();
738 case vmIntrinsics::_VectorCompare:
739 return inline_vector_compare();
740 case vmIntrinsics::_VectorBroadcastInt:
741 return inline_vector_broadcast_int();
742 case vmIntrinsics::_VectorConvert:
743 return inline_vector_convert();
744 case vmIntrinsics::_VectorInsert:
745 return inline_vector_insert();
746 case vmIntrinsics::_VectorExtract:
747 return inline_vector_extract();
748 case vmIntrinsics::_VectorCompressExpand:
749 return inline_vector_compress_expand();
750 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
751 return inline_vector_select_from_two_vectors();
752 case vmIntrinsics::_IndexVector:
753 return inline_index_vector();
754 case vmIntrinsics::_IndexPartiallyInUpperRange:
755 return inline_index_partially_in_upper_range();
756
757 case vmIntrinsics::_getObjectSize:
758 return inline_getObjectSize();
759
760 case vmIntrinsics::_blackhole:
761 return inline_blackhole();
762
763 default:
764 // If you get here, it may be that someone has added a new intrinsic
765 // to the list in vmIntrinsics.hpp without implementing it here.
766 #ifndef PRODUCT
767 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
768 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
769 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
770 }
771 #endif
772 return false;
773 }
774 }
775
776 Node* LibraryCallKit::try_to_predicate(int predicate) {
777 if (!jvms()->has_method()) {
778 // Root JVMState has a null method.
779 assert(map()->memory()->Opcode() == Op_Parm, "");
780 // Insert the memory aliasing node
781 set_all_memory(reset_memory());
782 }
783 assert(merged_memory(), "");
784
785 switch (intrinsic_id()) {
786 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
787 return inline_cipherBlockChaining_AESCrypt_predicate(false);
788 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
789 return inline_cipherBlockChaining_AESCrypt_predicate(true);
790 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
791 return inline_electronicCodeBook_AESCrypt_predicate(false);
792 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
793 return inline_electronicCodeBook_AESCrypt_predicate(true);
794 case vmIntrinsics::_counterMode_AESCrypt:
795 return inline_counterMode_AESCrypt_predicate();
796 case vmIntrinsics::_digestBase_implCompressMB:
797 return inline_digestBase_implCompressMB_predicate(predicate);
798 case vmIntrinsics::_galoisCounterMode_AESCrypt:
799 return inline_galoisCounterMode_AESCrypt_predicate();
800
801 default:
802 // If you get here, it may be that someone has added a new intrinsic
803 // to the list in vmIntrinsics.hpp without implementing it here.
804 #ifndef PRODUCT
805 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
806 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
807 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
808 }
809 #endif
810 Node* slow_ctl = control();
811 set_control(top()); // No fast path intrinsic
812 return slow_ctl;
813 }
814 }
815
816 //------------------------------set_result-------------------------------
817 // Helper function for finishing intrinsics.
818 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
819 record_for_igvn(region);
820 set_control(_gvn.transform(region));
821 set_result( _gvn.transform(value));
822 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
823 }
824
825 //------------------------------generate_guard---------------------------
826 // Helper function for generating guarded fast-slow graph structures.
827 // The given 'test', if true, guards a slow path. If the test fails
828 // then a fast path can be taken. (We generally hope it fails.)
829 // In all cases, GraphKit::control() is updated to the fast path.
830 // The returned value represents the control for the slow path.
831 // The return value is never 'top'; it is either a valid control
832 // or null if it is obvious that the slow path can never be taken.
833 // Also, if region and the slow control are not null, the slow edge
834 // is appended to the region.
835 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
836 if (stopped()) {
837 // Already short circuited.
838 return nullptr;
839 }
840
841 // Build an if node and its projections.
842 // If test is true we take the slow path, which we assume is uncommon.
843 if (_gvn.type(test) == TypeInt::ZERO) {
844 // The slow branch is never taken. No need to build this guard.
845 return nullptr;
846 }
847
848 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
849
850 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
851 if (if_slow == top()) {
852 // The slow branch is never taken. No need to build this guard.
853 return nullptr;
854 }
855
856 if (region != nullptr)
857 region->add_req(if_slow);
858
859 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
860 set_control(if_fast);
861
862 return if_slow;
863 }
864
865 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
866 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
867 }
868 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
869 return generate_guard(test, region, PROB_FAIR);
870 }
871
872 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
873 Node* *pos_index) {
874 if (stopped())
875 return nullptr; // already stopped
876 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
877 return nullptr; // index is already adequately typed
878 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
879 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
880 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
881 if (is_neg != nullptr && pos_index != nullptr) {
882 // Emulate effect of Parse::adjust_map_after_if.
883 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
884 (*pos_index) = _gvn.transform(ccast);
885 }
886 return is_neg;
887 }
888
889 // Make sure that 'position' is a valid limit index, in [0..length].
890 // There are two equivalent plans for checking this:
891 // A. (offset + copyLength) unsigned<= arrayLength
892 // B. offset <= (arrayLength - copyLength)
893 // We require that all of the values above, except for the sum and
894 // difference, are already known to be non-negative.
895 // Plan A is robust in the face of overflow, if offset and copyLength
896 // are both hugely positive.
897 //
898 // Plan B is less direct and intuitive, but it does not overflow at
899 // all, since the difference of two non-negatives is always
900 // representable. Whenever Java methods must perform the equivalent
901 // check they generally use Plan B instead of Plan A.
902 // For the moment we use Plan A.
903 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
904 Node* subseq_length,
905 Node* array_length,
906 RegionNode* region) {
907 if (stopped())
908 return nullptr; // already stopped
909 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
910 if (zero_offset && subseq_length->eqv_uncast(array_length))
911 return nullptr; // common case of whole-array copy
912 Node* last = subseq_length;
913 if (!zero_offset) // last += offset
914 last = _gvn.transform(new AddINode(last, offset));
915 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
916 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
917 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
918 return is_over;
919 }
920
921 // Emit range checks for the given String.value byte array
922 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
923 if (stopped()) {
924 return; // already stopped
925 }
926 RegionNode* bailout = new RegionNode(1);
927 record_for_igvn(bailout);
928 if (char_count) {
929 // Convert char count to byte count
930 count = _gvn.transform(new LShiftINode(count, intcon(1)));
931 }
932
933 // Offset and count must not be negative
934 generate_negative_guard(offset, bailout);
935 generate_negative_guard(count, bailout);
936 // Offset + count must not exceed length of array
937 generate_limit_guard(offset, count, load_array_length(array), bailout);
938
939 if (bailout->req() > 1) {
940 PreserveJVMState pjvms(this);
941 set_control(_gvn.transform(bailout));
942 uncommon_trap(Deoptimization::Reason_intrinsic,
943 Deoptimization::Action_maybe_recompile);
944 }
945 }
946
947 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
948 bool is_immutable) {
949 ciKlass* thread_klass = env()->Thread_klass();
950 const Type* thread_type
951 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
952
953 Node* thread = _gvn.transform(new ThreadLocalNode());
954 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
955 tls_output = thread;
956
957 Node* thread_obj_handle
958 = (is_immutable
959 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
960 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
961 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
962 thread_obj_handle = _gvn.transform(thread_obj_handle);
963
964 DecoratorSet decorators = IN_NATIVE;
965 if (is_immutable) {
966 decorators |= C2_IMMUTABLE_MEMORY;
967 }
968 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
969 }
970
971 //--------------------------generate_current_thread--------------------
972 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
973 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
974 /*is_immutable*/false);
975 }
976
977 //--------------------------generate_virtual_thread--------------------
978 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
979 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
980 !C->method()->changes_current_thread());
981 }
982
983 //------------------------------make_string_method_node------------------------
984 // Helper method for String intrinsic functions. This version is called with
985 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
986 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
987 // containing the lengths of str1 and str2.
988 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
989 Node* result = nullptr;
990 switch (opcode) {
991 case Op_StrIndexOf:
992 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
993 str1_start, cnt1, str2_start, cnt2, ae);
994 break;
995 case Op_StrComp:
996 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
997 str1_start, cnt1, str2_start, cnt2, ae);
998 break;
999 case Op_StrEquals:
1000 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1001 // Use the constant length if there is one because optimized match rule may exist.
1002 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1003 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1004 break;
1005 default:
1006 ShouldNotReachHere();
1007 return nullptr;
1008 }
1009
1010 // All these intrinsics have checks.
1011 C->set_has_split_ifs(true); // Has chance for split-if optimization
1012 clear_upper_avx();
1013
1014 return _gvn.transform(result);
1015 }
1016
1017 //------------------------------inline_string_compareTo------------------------
1018 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1019 Node* arg1 = argument(0);
1020 Node* arg2 = argument(1);
1021
1022 arg1 = must_be_not_null(arg1, true);
1023 arg2 = must_be_not_null(arg2, true);
1024
1025 // Get start addr and length of first argument
1026 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1027 Node* arg1_cnt = load_array_length(arg1);
1028
1029 // Get start addr and length of second argument
1030 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1031 Node* arg2_cnt = load_array_length(arg2);
1032
1033 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1034 set_result(result);
1035 return true;
1036 }
1037
1038 //------------------------------inline_string_equals------------------------
1039 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1040 Node* arg1 = argument(0);
1041 Node* arg2 = argument(1);
1042
1043 // paths (plus control) merge
1044 RegionNode* region = new RegionNode(3);
1045 Node* phi = new PhiNode(region, TypeInt::BOOL);
1046
1047 if (!stopped()) {
1048
1049 arg1 = must_be_not_null(arg1, true);
1050 arg2 = must_be_not_null(arg2, true);
1051
1052 // Get start addr and length of first argument
1053 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1054 Node* arg1_cnt = load_array_length(arg1);
1055
1056 // Get start addr and length of second argument
1057 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1058 Node* arg2_cnt = load_array_length(arg2);
1059
1060 // Check for arg1_cnt != arg2_cnt
1061 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1062 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1063 Node* if_ne = generate_slow_guard(bol, nullptr);
1064 if (if_ne != nullptr) {
1065 phi->init_req(2, intcon(0));
1066 region->init_req(2, if_ne);
1067 }
1068
1069 // Check for count == 0 is done by assembler code for StrEquals.
1070
1071 if (!stopped()) {
1072 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1073 phi->init_req(1, equals);
1074 region->init_req(1, control());
1075 }
1076 }
1077
1078 // post merge
1079 set_control(_gvn.transform(region));
1080 record_for_igvn(region);
1081
1082 set_result(_gvn.transform(phi));
1083 return true;
1084 }
1085
1086 //------------------------------inline_array_equals----------------------------
1087 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1088 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1089 Node* arg1 = argument(0);
1090 Node* arg2 = argument(1);
1091
1092 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1093 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1094 clear_upper_avx();
1095
1096 return true;
1097 }
1098
1099
1100 //------------------------------inline_countPositives------------------------------
1101 bool LibraryCallKit::inline_countPositives() {
1102 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1103 return false;
1104 }
1105
1106 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1107 // no receiver since it is static method
1108 Node* ba = argument(0);
1109 Node* offset = argument(1);
1110 Node* len = argument(2);
1111
1112 ba = must_be_not_null(ba, true);
1113
1114 // Range checks
1115 generate_string_range_check(ba, offset, len, false);
1116 if (stopped()) {
1117 return true;
1118 }
1119 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1120 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1121 set_result(_gvn.transform(result));
1122 clear_upper_avx();
1123 return true;
1124 }
1125
1126 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1127 Node* index = argument(0);
1128 Node* length = bt == T_INT ? argument(1) : argument(2);
1129 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1130 return false;
1131 }
1132
1133 // check that length is positive
1134 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1135 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1136
1137 {
1138 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1139 uncommon_trap(Deoptimization::Reason_intrinsic,
1140 Deoptimization::Action_make_not_entrant);
1141 }
1142
1143 if (stopped()) {
1144 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1145 return true;
1146 }
1147
1148 // length is now known positive, add a cast node to make this explicit
1149 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1150 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1151 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1152 ConstraintCastNode::RegularDependency, bt);
1153 casted_length = _gvn.transform(casted_length);
1154 replace_in_map(length, casted_length);
1155 length = casted_length;
1156
1157 // Use an unsigned comparison for the range check itself
1158 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1159 BoolTest::mask btest = BoolTest::lt;
1160 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1161 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1162 _gvn.set_type(rc, rc->Value(&_gvn));
1163 if (!rc_bool->is_Con()) {
1164 record_for_igvn(rc);
1165 }
1166 set_control(_gvn.transform(new IfTrueNode(rc)));
1167 {
1168 PreserveJVMState pjvms(this);
1169 set_control(_gvn.transform(new IfFalseNode(rc)));
1170 uncommon_trap(Deoptimization::Reason_range_check,
1171 Deoptimization::Action_make_not_entrant);
1172 }
1173
1174 if (stopped()) {
1175 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1176 return true;
1177 }
1178
1179 // index is now known to be >= 0 and < length, cast it
1180 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1181 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1182 ConstraintCastNode::RegularDependency, bt);
1183 result = _gvn.transform(result);
1184 set_result(result);
1185 replace_in_map(index, result);
1186 return true;
1187 }
1188
1189 //------------------------------inline_string_indexOf------------------------
1190 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1191 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1192 return false;
1193 }
1194 Node* src = argument(0);
1195 Node* tgt = argument(1);
1196
1197 // Make the merge point
1198 RegionNode* result_rgn = new RegionNode(4);
1199 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1200
1201 src = must_be_not_null(src, true);
1202 tgt = must_be_not_null(tgt, true);
1203
1204 // Get start addr and length of source string
1205 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1206 Node* src_count = load_array_length(src);
1207
1208 // Get start addr and length of substring
1209 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1210 Node* tgt_count = load_array_length(tgt);
1211
1212 Node* result = nullptr;
1213 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1214
1215 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1216 // Divide src size by 2 if String is UTF16 encoded
1217 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1218 }
1219 if (ae == StrIntrinsicNode::UU) {
1220 // Divide substring size by 2 if String is UTF16 encoded
1221 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1222 }
1223
1224 if (call_opt_stub) {
1225 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1226 StubRoutines::_string_indexof_array[ae],
1227 "stringIndexOf", TypePtr::BOTTOM, src_start,
1228 src_count, tgt_start, tgt_count);
1229 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1230 } else {
1231 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1232 result_rgn, result_phi, ae);
1233 }
1234 if (result != nullptr) {
1235 result_phi->init_req(3, result);
1236 result_rgn->init_req(3, control());
1237 }
1238 set_control(_gvn.transform(result_rgn));
1239 record_for_igvn(result_rgn);
1240 set_result(_gvn.transform(result_phi));
1241
1242 return true;
1243 }
1244
1245 //-----------------------------inline_string_indexOfI-----------------------
1246 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1247 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1248 return false;
1249 }
1250 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1251 return false;
1252 }
1253
1254 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1255 Node* src = argument(0); // byte[]
1256 Node* src_count = argument(1); // char count
1257 Node* tgt = argument(2); // byte[]
1258 Node* tgt_count = argument(3); // char count
1259 Node* from_index = argument(4); // char index
1260
1261 src = must_be_not_null(src, true);
1262 tgt = must_be_not_null(tgt, true);
1263
1264 // Multiply byte array index by 2 if String is UTF16 encoded
1265 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1266 src_count = _gvn.transform(new SubINode(src_count, from_index));
1267 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1268 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1269
1270 // Range checks
1271 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1272 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1273 if (stopped()) {
1274 return true;
1275 }
1276
1277 RegionNode* region = new RegionNode(5);
1278 Node* phi = new PhiNode(region, TypeInt::INT);
1279 Node* result = nullptr;
1280
1281 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1282
1283 if (call_opt_stub) {
1284 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1285 StubRoutines::_string_indexof_array[ae],
1286 "stringIndexOf", TypePtr::BOTTOM, src_start,
1287 src_count, tgt_start, tgt_count);
1288 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1289 } else {
1290 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1291 region, phi, ae);
1292 }
1293 if (result != nullptr) {
1294 // The result is index relative to from_index if substring was found, -1 otherwise.
1295 // Generate code which will fold into cmove.
1296 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1297 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1298
1299 Node* if_lt = generate_slow_guard(bol, nullptr);
1300 if (if_lt != nullptr) {
1301 // result == -1
1302 phi->init_req(3, result);
1303 region->init_req(3, if_lt);
1304 }
1305 if (!stopped()) {
1306 result = _gvn.transform(new AddINode(result, from_index));
1307 phi->init_req(4, result);
1308 region->init_req(4, control());
1309 }
1310 }
1311
1312 set_control(_gvn.transform(region));
1313 record_for_igvn(region);
1314 set_result(_gvn.transform(phi));
1315 clear_upper_avx();
1316
1317 return true;
1318 }
1319
1320 // Create StrIndexOfNode with fast path checks
1321 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1322 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1323 // Check for substr count > string count
1324 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1325 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1326 Node* if_gt = generate_slow_guard(bol, nullptr);
1327 if (if_gt != nullptr) {
1328 phi->init_req(1, intcon(-1));
1329 region->init_req(1, if_gt);
1330 }
1331 if (!stopped()) {
1332 // Check for substr count == 0
1333 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1334 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1335 Node* if_zero = generate_slow_guard(bol, nullptr);
1336 if (if_zero != nullptr) {
1337 phi->init_req(2, intcon(0));
1338 region->init_req(2, if_zero);
1339 }
1340 }
1341 if (!stopped()) {
1342 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1343 }
1344 return nullptr;
1345 }
1346
1347 //-----------------------------inline_string_indexOfChar-----------------------
1348 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1349 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1350 return false;
1351 }
1352 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1353 return false;
1354 }
1355 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1356 Node* src = argument(0); // byte[]
1357 Node* int_ch = argument(1);
1358 Node* from_index = argument(2);
1359 Node* max = argument(3);
1360
1361 src = must_be_not_null(src, true);
1362
1363 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1364 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1365 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1366
1367 // Range checks
1368 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1369
1370 // Check for int_ch >= 0
1371 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1372 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1373 {
1374 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1375 uncommon_trap(Deoptimization::Reason_intrinsic,
1376 Deoptimization::Action_maybe_recompile);
1377 }
1378 if (stopped()) {
1379 return true;
1380 }
1381
1382 RegionNode* region = new RegionNode(3);
1383 Node* phi = new PhiNode(region, TypeInt::INT);
1384
1385 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1386 C->set_has_split_ifs(true); // Has chance for split-if optimization
1387 _gvn.transform(result);
1388
1389 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1390 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1391
1392 Node* if_lt = generate_slow_guard(bol, nullptr);
1393 if (if_lt != nullptr) {
1394 // result == -1
1395 phi->init_req(2, result);
1396 region->init_req(2, if_lt);
1397 }
1398 if (!stopped()) {
1399 result = _gvn.transform(new AddINode(result, from_index));
1400 phi->init_req(1, result);
1401 region->init_req(1, control());
1402 }
1403 set_control(_gvn.transform(region));
1404 record_for_igvn(region);
1405 set_result(_gvn.transform(phi));
1406 clear_upper_avx();
1407
1408 return true;
1409 }
1410 //---------------------------inline_string_copy---------------------
1411 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1412 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1413 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1414 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1415 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1416 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1417 bool LibraryCallKit::inline_string_copy(bool compress) {
1418 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1419 return false;
1420 }
1421 int nargs = 5; // 2 oops, 3 ints
1422 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1423
1424 Node* src = argument(0);
1425 Node* src_offset = argument(1);
1426 Node* dst = argument(2);
1427 Node* dst_offset = argument(3);
1428 Node* length = argument(4);
1429
1430 // Check for allocation before we add nodes that would confuse
1431 // tightly_coupled_allocation()
1432 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1433
1434 // Figure out the size and type of the elements we will be copying.
1435 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1436 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1437 if (src_type == nullptr || dst_type == nullptr) {
1438 return false;
1439 }
1440 BasicType src_elem = src_type->elem()->array_element_basic_type();
1441 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1442 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1443 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1444 "Unsupported array types for inline_string_copy");
1445
1446 src = must_be_not_null(src, true);
1447 dst = must_be_not_null(dst, true);
1448
1449 // Convert char[] offsets to byte[] offsets
1450 bool convert_src = (compress && src_elem == T_BYTE);
1451 bool convert_dst = (!compress && dst_elem == T_BYTE);
1452 if (convert_src) {
1453 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1454 } else if (convert_dst) {
1455 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1456 }
1457
1458 // Range checks
1459 generate_string_range_check(src, src_offset, length, convert_src);
1460 generate_string_range_check(dst, dst_offset, length, convert_dst);
1461 if (stopped()) {
1462 return true;
1463 }
1464
1465 Node* src_start = array_element_address(src, src_offset, src_elem);
1466 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1467 // 'src_start' points to src array + scaled offset
1468 // 'dst_start' points to dst array + scaled offset
1469 Node* count = nullptr;
1470 if (compress) {
1471 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1472 } else {
1473 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1474 }
1475
1476 if (alloc != nullptr) {
1477 if (alloc->maybe_set_complete(&_gvn)) {
1478 // "You break it, you buy it."
1479 InitializeNode* init = alloc->initialization();
1480 assert(init->is_complete(), "we just did this");
1481 init->set_complete_with_arraycopy();
1482 assert(dst->is_CheckCastPP(), "sanity");
1483 assert(dst->in(0)->in(0) == init, "dest pinned");
1484 }
1485 // Do not let stores that initialize this object be reordered with
1486 // a subsequent store that would make this object accessible by
1487 // other threads.
1488 // Record what AllocateNode this StoreStore protects so that
1489 // escape analysis can go from the MemBarStoreStoreNode to the
1490 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1491 // based on the escape status of the AllocateNode.
1492 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1493 }
1494 if (compress) {
1495 set_result(_gvn.transform(count));
1496 }
1497 clear_upper_avx();
1498
1499 return true;
1500 }
1501
1502 #ifdef _LP64
1503 #define XTOP ,top() /*additional argument*/
1504 #else //_LP64
1505 #define XTOP /*no additional argument*/
1506 #endif //_LP64
1507
1508 //------------------------inline_string_toBytesU--------------------------
1509 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1510 bool LibraryCallKit::inline_string_toBytesU() {
1511 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1512 return false;
1513 }
1514 // Get the arguments.
1515 Node* value = argument(0);
1516 Node* offset = argument(1);
1517 Node* length = argument(2);
1518
1519 Node* newcopy = nullptr;
1520
1521 // Set the original stack and the reexecute bit for the interpreter to reexecute
1522 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1523 { PreserveReexecuteState preexecs(this);
1524 jvms()->set_should_reexecute(true);
1525
1526 // Check if a null path was taken unconditionally.
1527 value = null_check(value);
1528
1529 RegionNode* bailout = new RegionNode(1);
1530 record_for_igvn(bailout);
1531
1532 // Range checks
1533 generate_negative_guard(offset, bailout);
1534 generate_negative_guard(length, bailout);
1535 generate_limit_guard(offset, length, load_array_length(value), bailout);
1536 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1537 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1538
1539 if (bailout->req() > 1) {
1540 PreserveJVMState pjvms(this);
1541 set_control(_gvn.transform(bailout));
1542 uncommon_trap(Deoptimization::Reason_intrinsic,
1543 Deoptimization::Action_maybe_recompile);
1544 }
1545 if (stopped()) {
1546 return true;
1547 }
1548
1549 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1550 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1551 newcopy = new_array(klass_node, size, 0); // no arguments to push
1552 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1553 guarantee(alloc != nullptr, "created above");
1554
1555 // Calculate starting addresses.
1556 Node* src_start = array_element_address(value, offset, T_CHAR);
1557 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1558
1559 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1560 const TypeInt* toffset = gvn().type(offset)->is_int();
1561 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1562
1563 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1564 const char* copyfunc_name = "arraycopy";
1565 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1566 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1567 OptoRuntime::fast_arraycopy_Type(),
1568 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1569 src_start, dst_start, ConvI2X(length) XTOP);
1570 // Do not let reads from the cloned object float above the arraycopy.
1571 if (alloc->maybe_set_complete(&_gvn)) {
1572 // "You break it, you buy it."
1573 InitializeNode* init = alloc->initialization();
1574 assert(init->is_complete(), "we just did this");
1575 init->set_complete_with_arraycopy();
1576 assert(newcopy->is_CheckCastPP(), "sanity");
1577 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1578 }
1579 // Do not let stores that initialize this object be reordered with
1580 // a subsequent store that would make this object accessible by
1581 // other threads.
1582 // Record what AllocateNode this StoreStore protects so that
1583 // escape analysis can go from the MemBarStoreStoreNode to the
1584 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1585 // based on the escape status of the AllocateNode.
1586 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1587 } // original reexecute is set back here
1588
1589 C->set_has_split_ifs(true); // Has chance for split-if optimization
1590 if (!stopped()) {
1591 set_result(newcopy);
1592 }
1593 clear_upper_avx();
1594
1595 return true;
1596 }
1597
1598 //------------------------inline_string_getCharsU--------------------------
1599 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1600 bool LibraryCallKit::inline_string_getCharsU() {
1601 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1602 return false;
1603 }
1604
1605 // Get the arguments.
1606 Node* src = argument(0);
1607 Node* src_begin = argument(1);
1608 Node* src_end = argument(2); // exclusive offset (i < src_end)
1609 Node* dst = argument(3);
1610 Node* dst_begin = argument(4);
1611
1612 // Check for allocation before we add nodes that would confuse
1613 // tightly_coupled_allocation()
1614 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1615
1616 // Check if a null path was taken unconditionally.
1617 src = null_check(src);
1618 dst = null_check(dst);
1619 if (stopped()) {
1620 return true;
1621 }
1622
1623 // Get length and convert char[] offset to byte[] offset
1624 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1625 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1626
1627 // Range checks
1628 generate_string_range_check(src, src_begin, length, true);
1629 generate_string_range_check(dst, dst_begin, length, false);
1630 if (stopped()) {
1631 return true;
1632 }
1633
1634 if (!stopped()) {
1635 // Calculate starting addresses.
1636 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1637 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1638
1639 // Check if array addresses are aligned to HeapWordSize
1640 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1641 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1642 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1643 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1644
1645 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1646 const char* copyfunc_name = "arraycopy";
1647 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1648 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1649 OptoRuntime::fast_arraycopy_Type(),
1650 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1651 src_start, dst_start, ConvI2X(length) XTOP);
1652 // Do not let reads from the cloned object float above the arraycopy.
1653 if (alloc != nullptr) {
1654 if (alloc->maybe_set_complete(&_gvn)) {
1655 // "You break it, you buy it."
1656 InitializeNode* init = alloc->initialization();
1657 assert(init->is_complete(), "we just did this");
1658 init->set_complete_with_arraycopy();
1659 assert(dst->is_CheckCastPP(), "sanity");
1660 assert(dst->in(0)->in(0) == init, "dest pinned");
1661 }
1662 // Do not let stores that initialize this object be reordered with
1663 // a subsequent store that would make this object accessible by
1664 // other threads.
1665 // Record what AllocateNode this StoreStore protects so that
1666 // escape analysis can go from the MemBarStoreStoreNode to the
1667 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1668 // based on the escape status of the AllocateNode.
1669 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1670 } else {
1671 insert_mem_bar(Op_MemBarCPUOrder);
1672 }
1673 }
1674
1675 C->set_has_split_ifs(true); // Has chance for split-if optimization
1676 return true;
1677 }
1678
1679 //----------------------inline_string_char_access----------------------------
1680 // Store/Load char to/from byte[] array.
1681 // static void StringUTF16.putChar(byte[] val, int index, int c)
1682 // static char StringUTF16.getChar(byte[] val, int index)
1683 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1684 Node* value = argument(0);
1685 Node* index = argument(1);
1686 Node* ch = is_store ? argument(2) : nullptr;
1687
1688 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1689 // correctly requires matched array shapes.
1690 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1691 "sanity: byte[] and char[] bases agree");
1692 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1693 "sanity: byte[] and char[] scales agree");
1694
1695 // Bail when getChar over constants is requested: constant folding would
1696 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1697 // Java method would constant fold nicely instead.
1698 if (!is_store && value->is_Con() && index->is_Con()) {
1699 return false;
1700 }
1701
1702 // Save state and restore on bailout
1703 uint old_sp = sp();
1704 SafePointNode* old_map = clone_map();
1705
1706 value = must_be_not_null(value, true);
1707
1708 Node* adr = array_element_address(value, index, T_CHAR);
1709 if (adr->is_top()) {
1710 set_map(old_map);
1711 set_sp(old_sp);
1712 return false;
1713 }
1714 destruct_map_clone(old_map);
1715 if (is_store) {
1716 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1717 } else {
1718 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);
1719 set_result(ch);
1720 }
1721 return true;
1722 }
1723
1724 //--------------------------round_double_node--------------------------------
1725 // Round a double node if necessary.
1726 Node* LibraryCallKit::round_double_node(Node* n) {
1727 if (Matcher::strict_fp_requires_explicit_rounding) {
1728 #ifdef IA32
1729 if (UseSSE < 2) {
1730 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1731 }
1732 #else
1733 Unimplemented();
1734 #endif // IA32
1735 }
1736 return n;
1737 }
1738
1739 //------------------------------inline_math-----------------------------------
1740 // public static double Math.abs(double)
1741 // public static double Math.sqrt(double)
1742 // public static double Math.log(double)
1743 // public static double Math.log10(double)
1744 // public static double Math.round(double)
1745 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1746 Node* arg = round_double_node(argument(0));
1747 Node* n = nullptr;
1748 switch (id) {
1749 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1750 case vmIntrinsics::_dsqrt:
1751 case vmIntrinsics::_dsqrt_strict:
1752 n = new SqrtDNode(C, control(), arg); break;
1753 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1754 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1755 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1756 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1757 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1758 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1759 default: fatal_unexpected_iid(id); break;
1760 }
1761 set_result(_gvn.transform(n));
1762 return true;
1763 }
1764
1765 //------------------------------inline_math-----------------------------------
1766 // public static float Math.abs(float)
1767 // public static int Math.abs(int)
1768 // public static long Math.abs(long)
1769 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1770 Node* arg = argument(0);
1771 Node* n = nullptr;
1772 switch (id) {
1773 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1774 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1775 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1776 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1777 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1778 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1779 default: fatal_unexpected_iid(id); break;
1780 }
1781 set_result(_gvn.transform(n));
1782 return true;
1783 }
1784
1785 //------------------------------runtime_math-----------------------------
1786 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1787 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1788 "must be (DD)D or (D)D type");
1789
1790 // Inputs
1791 Node* a = round_double_node(argument(0));
1792 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1793
1794 const TypePtr* no_memory_effects = nullptr;
1795 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1796 no_memory_effects,
1797 a, top(), b, b ? top() : nullptr);
1798 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1799 #ifdef ASSERT
1800 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1801 assert(value_top == top(), "second value must be top");
1802 #endif
1803
1804 set_result(value);
1805 return true;
1806 }
1807
1808 //------------------------------inline_math_pow-----------------------------
1809 bool LibraryCallKit::inline_math_pow() {
1810 Node* exp = round_double_node(argument(2));
1811 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1812 if (d != nullptr) {
1813 if (d->getd() == 2.0) {
1814 // Special case: pow(x, 2.0) => x * x
1815 Node* base = round_double_node(argument(0));
1816 set_result(_gvn.transform(new MulDNode(base, base)));
1817 return true;
1818 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1819 // Special case: pow(x, 0.5) => sqrt(x)
1820 Node* base = round_double_node(argument(0));
1821 Node* zero = _gvn.zerocon(T_DOUBLE);
1822
1823 RegionNode* region = new RegionNode(3);
1824 Node* phi = new PhiNode(region, Type::DOUBLE);
1825
1826 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1827 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1828 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1829 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1830 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1831
1832 Node* if_pow = generate_slow_guard(test, nullptr);
1833 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1834 phi->init_req(1, value_sqrt);
1835 region->init_req(1, control());
1836
1837 if (if_pow != nullptr) {
1838 set_control(if_pow);
1839 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1840 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1841 const TypePtr* no_memory_effects = nullptr;
1842 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1843 no_memory_effects, base, top(), exp, top());
1844 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1845 #ifdef ASSERT
1846 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1847 assert(value_top == top(), "second value must be top");
1848 #endif
1849 phi->init_req(2, value_pow);
1850 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1851 }
1852
1853 C->set_has_split_ifs(true); // Has chance for split-if optimization
1854 set_control(_gvn.transform(region));
1855 record_for_igvn(region);
1856 set_result(_gvn.transform(phi));
1857
1858 return true;
1859 }
1860 }
1861
1862 return StubRoutines::dpow() != nullptr ?
1863 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1864 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1865 }
1866
1867 //------------------------------inline_math_native-----------------------------
1868 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1869 switch (id) {
1870 case vmIntrinsics::_dsin:
1871 return StubRoutines::dsin() != nullptr ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1874 case vmIntrinsics::_dcos:
1875 return StubRoutines::dcos() != nullptr ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1878 case vmIntrinsics::_dtan:
1879 return StubRoutines::dtan() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1882 case vmIntrinsics::_dtanh:
1883 return StubRoutines::dtanh() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1885 case vmIntrinsics::_dexp:
1886 return StubRoutines::dexp() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1889 case vmIntrinsics::_dlog:
1890 return StubRoutines::dlog() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1893 case vmIntrinsics::_dlog10:
1894 return StubRoutines::dlog10() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1897
1898 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1899 case vmIntrinsics::_ceil:
1900 case vmIntrinsics::_floor:
1901 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1902
1903 case vmIntrinsics::_dsqrt:
1904 case vmIntrinsics::_dsqrt_strict:
1905 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1908 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1909 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1910
1911 case vmIntrinsics::_dpow: return inline_math_pow();
1912 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1913 case vmIntrinsics::_fcopySign: return inline_math(id);
1914 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1916 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1917
1918 // These intrinsics are not yet correctly implemented
1919 case vmIntrinsics::_datan2:
1920 return false;
1921
1922 default:
1923 fatal_unexpected_iid(id);
1924 return false;
1925 }
1926 }
1927
1928 //----------------------------inline_notify-----------------------------------*
1929 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1930 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1931 address func;
1932 if (id == vmIntrinsics::_notify) {
1933 func = OptoRuntime::monitor_notify_Java();
1934 } else {
1935 func = OptoRuntime::monitor_notifyAll_Java();
1936 }
1937 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1938 make_slow_call_ex(call, env()->Throwable_klass(), false);
1939 return true;
1940 }
1941
1942
1943 //----------------------------inline_min_max-----------------------------------
1944 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1945 Node* a = nullptr;
1946 Node* b = nullptr;
1947 Node* n = nullptr;
1948 switch (id) {
1949 case vmIntrinsics::_min:
1950 case vmIntrinsics::_max:
1951 case vmIntrinsics::_minF:
1952 case vmIntrinsics::_maxF:
1953 case vmIntrinsics::_minF_strict:
1954 case vmIntrinsics::_maxF_strict:
1955 case vmIntrinsics::_min_strict:
1956 case vmIntrinsics::_max_strict:
1957 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1958 a = argument(0);
1959 b = argument(1);
1960 break;
1961 case vmIntrinsics::_minD:
1962 case vmIntrinsics::_maxD:
1963 case vmIntrinsics::_minD_strict:
1964 case vmIntrinsics::_maxD_strict:
1965 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1966 a = round_double_node(argument(0));
1967 b = round_double_node(argument(2));
1968 break;
1969 case vmIntrinsics::_minL:
1970 case vmIntrinsics::_maxL:
1971 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1972 a = argument(0);
1973 b = argument(2);
1974 break;
1975 default:
1976 fatal_unexpected_iid(id);
1977 break;
1978 }
1979
1980 switch (id) {
1981 case vmIntrinsics::_min:
1982 case vmIntrinsics::_min_strict:
1983 n = new MinINode(a, b);
1984 break;
1985 case vmIntrinsics::_max:
1986 case vmIntrinsics::_max_strict:
1987 n = new MaxINode(a, b);
1988 break;
1989 case vmIntrinsics::_minF:
1990 case vmIntrinsics::_minF_strict:
1991 n = new MinFNode(a, b);
1992 break;
1993 case vmIntrinsics::_maxF:
1994 case vmIntrinsics::_maxF_strict:
1995 n = new MaxFNode(a, b);
1996 break;
1997 case vmIntrinsics::_minD:
1998 case vmIntrinsics::_minD_strict:
1999 n = new MinDNode(a, b);
2000 break;
2001 case vmIntrinsics::_maxD:
2002 case vmIntrinsics::_maxD_strict:
2003 n = new MaxDNode(a, b);
2004 break;
2005 case vmIntrinsics::_minL:
2006 n = new MinLNode(_gvn.C, a, b);
2007 break;
2008 case vmIntrinsics::_maxL:
2009 n = new MaxLNode(_gvn.C, a, b);
2010 break;
2011 default:
2012 fatal_unexpected_iid(id);
2013 break;
2014 }
2015
2016 set_result(_gvn.transform(n));
2017 return true;
2018 }
2019
2020 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
2021 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2022 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2023 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2024 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2025
2026 {
2027 PreserveJVMState pjvms(this);
2028 PreserveReexecuteState preexecs(this);
2029 jvms()->set_should_reexecute(true);
2030
2031 set_control(slow_path);
2032 set_i_o(i_o());
2033
2034 uncommon_trap(Deoptimization::Reason_intrinsic,
2035 Deoptimization::Action_none);
2036 }
2037
2038 set_control(fast_path);
2039 set_result(math);
2040 }
2041
2042 template <typename OverflowOp>
2043 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2044 typedef typename OverflowOp::MathOp MathOp;
2045
2046 MathOp* mathOp = new MathOp(arg1, arg2);
2047 Node* operation = _gvn.transform( mathOp );
2048 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2049 inline_math_mathExact(operation, ofcheck);
2050 return true;
2051 }
2052
2053 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2054 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2055 }
2056
2057 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2058 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2059 }
2060
2061 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2062 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2063 }
2064
2065 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2066 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2067 }
2068
2069 bool LibraryCallKit::inline_math_negateExactI() {
2070 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2071 }
2072
2073 bool LibraryCallKit::inline_math_negateExactL() {
2074 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2075 }
2076
2077 bool LibraryCallKit::inline_math_multiplyExactI() {
2078 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2079 }
2080
2081 bool LibraryCallKit::inline_math_multiplyExactL() {
2082 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2083 }
2084
2085 bool LibraryCallKit::inline_math_multiplyHigh() {
2086 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2087 return true;
2088 }
2089
2090 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2091 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2092 return true;
2093 }
2094
2095 inline int
2096 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2097 const TypePtr* base_type = TypePtr::NULL_PTR;
2098 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2099 if (base_type == nullptr) {
2100 // Unknown type.
2101 return Type::AnyPtr;
2102 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2103 // Since this is a null+long form, we have to switch to a rawptr.
2104 base = _gvn.transform(new CastX2PNode(offset));
2105 offset = MakeConX(0);
2106 return Type::RawPtr;
2107 } else if (base_type->base() == Type::RawPtr) {
2108 return Type::RawPtr;
2109 } else if (base_type->isa_oopptr()) {
2110 // Base is never null => always a heap address.
2111 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2112 return Type::OopPtr;
2113 }
2114 // Offset is small => always a heap address.
2115 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2116 if (offset_type != nullptr &&
2117 base_type->offset() == 0 && // (should always be?)
2118 offset_type->_lo >= 0 &&
2119 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2120 return Type::OopPtr;
2121 } else if (type == T_OBJECT) {
2122 // off heap access to an oop doesn't make any sense. Has to be on
2123 // heap.
2124 return Type::OopPtr;
2125 }
2126 // Otherwise, it might either be oop+off or null+addr.
2127 return Type::AnyPtr;
2128 } else {
2129 // No information:
2130 return Type::AnyPtr;
2131 }
2132 }
2133
2134 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2135 Node* uncasted_base = base;
2136 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2137 if (kind == Type::RawPtr) {
2138 return basic_plus_adr(top(), uncasted_base, offset);
2139 } else if (kind == Type::AnyPtr) {
2140 assert(base == uncasted_base, "unexpected base change");
2141 if (can_cast) {
2142 if (!_gvn.type(base)->speculative_maybe_null() &&
2143 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2144 // According to profiling, this access is always on
2145 // heap. Casting the base to not null and thus avoiding membars
2146 // around the access should allow better optimizations
2147 Node* null_ctl = top();
2148 base = null_check_oop(base, &null_ctl, true, true, true);
2149 assert(null_ctl->is_top(), "no null control here");
2150 return basic_plus_adr(base, offset);
2151 } else if (_gvn.type(base)->speculative_always_null() &&
2152 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2153 // According to profiling, this access is always off
2154 // heap.
2155 base = null_assert(base);
2156 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2157 offset = MakeConX(0);
2158 return basic_plus_adr(top(), raw_base, offset);
2159 }
2160 }
2161 // We don't know if it's an on heap or off heap access. Fall back
2162 // to raw memory access.
2163 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2164 return basic_plus_adr(top(), raw, offset);
2165 } else {
2166 assert(base == uncasted_base, "unexpected base change");
2167 // We know it's an on heap access so base can't be null
2168 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2169 base = must_be_not_null(base, true);
2170 }
2171 return basic_plus_adr(base, offset);
2172 }
2173 }
2174
2175 //--------------------------inline_number_methods-----------------------------
2176 // inline int Integer.numberOfLeadingZeros(int)
2177 // inline int Long.numberOfLeadingZeros(long)
2178 //
2179 // inline int Integer.numberOfTrailingZeros(int)
2180 // inline int Long.numberOfTrailingZeros(long)
2181 //
2182 // inline int Integer.bitCount(int)
2183 // inline int Long.bitCount(long)
2184 //
2185 // inline char Character.reverseBytes(char)
2186 // inline short Short.reverseBytes(short)
2187 // inline int Integer.reverseBytes(int)
2188 // inline long Long.reverseBytes(long)
2189 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2190 Node* arg = argument(0);
2191 Node* n = nullptr;
2192 switch (id) {
2193 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2194 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2195 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2196 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2197 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2198 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2199 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2200 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2201 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2202 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2203 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2204 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2205 default: fatal_unexpected_iid(id); break;
2206 }
2207 set_result(_gvn.transform(n));
2208 return true;
2209 }
2210
2211 //--------------------------inline_bitshuffle_methods-----------------------------
2212 // inline int Integer.compress(int, int)
2213 // inline int Integer.expand(int, int)
2214 // inline long Long.compress(long, long)
2215 // inline long Long.expand(long, long)
2216 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2217 Node* n = nullptr;
2218 switch (id) {
2219 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2220 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2221 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2222 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2223 default: fatal_unexpected_iid(id); break;
2224 }
2225 set_result(_gvn.transform(n));
2226 return true;
2227 }
2228
2229 //--------------------------inline_number_methods-----------------------------
2230 // inline int Integer.compareUnsigned(int, int)
2231 // inline int Long.compareUnsigned(long, long)
2232 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2233 Node* arg1 = argument(0);
2234 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2235 Node* n = nullptr;
2236 switch (id) {
2237 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2238 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2239 default: fatal_unexpected_iid(id); break;
2240 }
2241 set_result(_gvn.transform(n));
2242 return true;
2243 }
2244
2245 //--------------------------inline_unsigned_divmod_methods-----------------------------
2246 // inline int Integer.divideUnsigned(int, int)
2247 // inline int Integer.remainderUnsigned(int, int)
2248 // inline long Long.divideUnsigned(long, long)
2249 // inline long Long.remainderUnsigned(long, long)
2250 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2251 Node* n = nullptr;
2252 switch (id) {
2253 case vmIntrinsics::_divideUnsigned_i: {
2254 zero_check_int(argument(1));
2255 // Compile-time detect of null-exception
2256 if (stopped()) {
2257 return true; // keep the graph constructed so far
2258 }
2259 n = new UDivINode(control(), argument(0), argument(1));
2260 break;
2261 }
2262 case vmIntrinsics::_divideUnsigned_l: {
2263 zero_check_long(argument(2));
2264 // Compile-time detect of null-exception
2265 if (stopped()) {
2266 return true; // keep the graph constructed so far
2267 }
2268 n = new UDivLNode(control(), argument(0), argument(2));
2269 break;
2270 }
2271 case vmIntrinsics::_remainderUnsigned_i: {
2272 zero_check_int(argument(1));
2273 // Compile-time detect of null-exception
2274 if (stopped()) {
2275 return true; // keep the graph constructed so far
2276 }
2277 n = new UModINode(control(), argument(0), argument(1));
2278 break;
2279 }
2280 case vmIntrinsics::_remainderUnsigned_l: {
2281 zero_check_long(argument(2));
2282 // Compile-time detect of null-exception
2283 if (stopped()) {
2284 return true; // keep the graph constructed so far
2285 }
2286 n = new UModLNode(control(), argument(0), argument(2));
2287 break;
2288 }
2289 default: fatal_unexpected_iid(id); break;
2290 }
2291 set_result(_gvn.transform(n));
2292 return true;
2293 }
2294
2295 //----------------------------inline_unsafe_access----------------------------
2296
2297 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2298 // Attempt to infer a sharper value type from the offset and base type.
2299 ciKlass* sharpened_klass = nullptr;
2300
2301 // See if it is an instance field, with an object type.
2302 if (alias_type->field() != nullptr) {
2303 if (alias_type->field()->type()->is_klass()) {
2304 sharpened_klass = alias_type->field()->type()->as_klass();
2305 }
2306 }
2307
2308 const TypeOopPtr* result = nullptr;
2309 // See if it is a narrow oop array.
2310 if (adr_type->isa_aryptr()) {
2311 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2312 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2313 if (elem_type != nullptr && elem_type->is_loaded()) {
2314 // Sharpen the value type.
2315 result = elem_type;
2316 }
2317 }
2318 }
2319
2320 // The sharpened class might be unloaded if there is no class loader
2321 // contraint in place.
2322 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2323 // Sharpen the value type.
2324 result = TypeOopPtr::make_from_klass(sharpened_klass);
2325 }
2326 if (result != nullptr) {
2327 #ifndef PRODUCT
2328 if (C->print_intrinsics() || C->print_inlining()) {
2329 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2330 tty->print(" sharpened value: "); result->dump(); tty->cr();
2331 }
2332 #endif
2333 }
2334 return result;
2335 }
2336
2337 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2338 switch (kind) {
2339 case Relaxed:
2340 return MO_UNORDERED;
2341 case Opaque:
2342 return MO_RELAXED;
2343 case Acquire:
2344 return MO_ACQUIRE;
2345 case Release:
2346 return MO_RELEASE;
2347 case Volatile:
2348 return MO_SEQ_CST;
2349 default:
2350 ShouldNotReachHere();
2351 return 0;
2352 }
2353 }
2354
2355 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2356 if (callee()->is_static()) return false; // caller must have the capability!
2357 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2358 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2359 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2360 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2361
2362 if (is_reference_type(type)) {
2363 decorators |= ON_UNKNOWN_OOP_REF;
2364 }
2365
2366 if (unaligned) {
2367 decorators |= C2_UNALIGNED;
2368 }
2369
2370 #ifndef PRODUCT
2371 {
2372 ResourceMark rm;
2373 // Check the signatures.
2374 ciSignature* sig = callee()->signature();
2375 #ifdef ASSERT
2376 if (!is_store) {
2377 // Object getReference(Object base, int/long offset), etc.
2378 BasicType rtype = sig->return_type()->basic_type();
2379 assert(rtype == type, "getter must return the expected value");
2380 assert(sig->count() == 2, "oop getter has 2 arguments");
2381 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2382 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2383 } else {
2384 // void putReference(Object base, int/long offset, Object x), etc.
2385 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2386 assert(sig->count() == 3, "oop putter has 3 arguments");
2387 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2388 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2389 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2390 assert(vtype == type, "putter must accept the expected value");
2391 }
2392 #endif // ASSERT
2393 }
2394 #endif //PRODUCT
2395
2396 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2397
2398 Node* receiver = argument(0); // type: oop
2399
2400 // Build address expression.
2401 Node* heap_base_oop = top();
2402
2403 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2404 Node* base = argument(1); // type: oop
2405 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2406 Node* offset = argument(2); // type: long
2407 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2408 // to be plain byte offsets, which are also the same as those accepted
2409 // by oopDesc::field_addr.
2410 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2411 "fieldOffset must be byte-scaled");
2412 // 32-bit machines ignore the high half!
2413 offset = ConvL2X(offset);
2414
2415 // Save state and restore on bailout
2416 uint old_sp = sp();
2417 SafePointNode* old_map = clone_map();
2418
2419 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2420 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2421
2422 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2423 if (type != T_OBJECT) {
2424 decorators |= IN_NATIVE; // off-heap primitive access
2425 } else {
2426 set_map(old_map);
2427 set_sp(old_sp);
2428 return false; // off-heap oop accesses are not supported
2429 }
2430 } else {
2431 heap_base_oop = base; // on-heap or mixed access
2432 }
2433
2434 // Can base be null? Otherwise, always on-heap access.
2435 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2436
2437 if (!can_access_non_heap) {
2438 decorators |= IN_HEAP;
2439 }
2440
2441 Node* val = is_store ? argument(4) : nullptr;
2442
2443 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2444 if (adr_type == TypePtr::NULL_PTR) {
2445 set_map(old_map);
2446 set_sp(old_sp);
2447 return false; // off-heap access with zero address
2448 }
2449
2450 // Try to categorize the address.
2451 Compile::AliasType* alias_type = C->alias_type(adr_type);
2452 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2453
2454 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2455 alias_type->adr_type() == TypeAryPtr::RANGE) {
2456 set_map(old_map);
2457 set_sp(old_sp);
2458 return false; // not supported
2459 }
2460
2461 bool mismatched = false;
2462 BasicType bt = alias_type->basic_type();
2463 if (bt != T_ILLEGAL) {
2464 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2465 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2466 // Alias type doesn't differentiate between byte[] and boolean[]).
2467 // Use address type to get the element type.
2468 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2469 }
2470 if (is_reference_type(bt, true)) {
2471 // accessing an array field with getReference is not a mismatch
2472 bt = T_OBJECT;
2473 }
2474 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2475 // Don't intrinsify mismatched object accesses
2476 set_map(old_map);
2477 set_sp(old_sp);
2478 return false;
2479 }
2480 mismatched = (bt != type);
2481 } else if (alias_type->adr_type()->isa_oopptr()) {
2482 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2483 }
2484
2485 destruct_map_clone(old_map);
2486 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2487
2488 if (mismatched) {
2489 decorators |= C2_MISMATCHED;
2490 }
2491
2492 // First guess at the value type.
2493 const Type *value_type = Type::get_const_basic_type(type);
2494
2495 // Figure out the memory ordering.
2496 decorators |= mo_decorator_for_access_kind(kind);
2497
2498 if (!is_store && type == T_OBJECT) {
2499 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2500 if (tjp != nullptr) {
2501 value_type = tjp;
2502 }
2503 }
2504
2505 receiver = null_check(receiver);
2506 if (stopped()) {
2507 return true;
2508 }
2509 // Heap pointers get a null-check from the interpreter,
2510 // as a courtesy. However, this is not guaranteed by Unsafe,
2511 // and it is not possible to fully distinguish unintended nulls
2512 // from intended ones in this API.
2513
2514 if (!is_store) {
2515 Node* p = nullptr;
2516 // Try to constant fold a load from a constant field
2517 ciField* field = alias_type->field();
2518 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2519 // final or stable field
2520 p = make_constant_from_field(field, heap_base_oop);
2521 }
2522
2523 if (p == nullptr) { // Could not constant fold the load
2524 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2525 // Normalize the value returned by getBoolean in the following cases
2526 if (type == T_BOOLEAN &&
2527 (mismatched ||
2528 heap_base_oop == top() || // - heap_base_oop is null or
2529 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2530 // and the unsafe access is made to large offset
2531 // (i.e., larger than the maximum offset necessary for any
2532 // field access)
2533 ) {
2534 IdealKit ideal = IdealKit(this);
2535 #define __ ideal.
2536 IdealVariable normalized_result(ideal);
2537 __ declarations_done();
2538 __ set(normalized_result, p);
2539 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2540 __ set(normalized_result, ideal.ConI(1));
2541 ideal.end_if();
2542 final_sync(ideal);
2543 p = __ value(normalized_result);
2544 #undef __
2545 }
2546 }
2547 if (type == T_ADDRESS) {
2548 p = gvn().transform(new CastP2XNode(nullptr, p));
2549 p = ConvX2UL(p);
2550 }
2551 // The load node has the control of the preceding MemBarCPUOrder. All
2552 // following nodes will have the control of the MemBarCPUOrder inserted at
2553 // the end of this method. So, pushing the load onto the stack at a later
2554 // point is fine.
2555 set_result(p);
2556 } else {
2557 if (bt == T_ADDRESS) {
2558 // Repackage the long as a pointer.
2559 val = ConvL2X(val);
2560 val = gvn().transform(new CastX2PNode(val));
2561 }
2562 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2563 }
2564
2565 return true;
2566 }
2567
2568 //----------------------------inline_unsafe_load_store----------------------------
2569 // This method serves a couple of different customers (depending on LoadStoreKind):
2570 //
2571 // LS_cmp_swap:
2572 //
2573 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2574 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2575 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2576 //
2577 // LS_cmp_swap_weak:
2578 //
2579 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2580 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2581 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2582 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2583 //
2584 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2585 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2586 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2587 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2588 //
2589 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2590 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2591 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2592 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2593 //
2594 // LS_cmp_exchange:
2595 //
2596 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2597 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2598 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2599 //
2600 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2601 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2602 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2603 //
2604 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2605 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2606 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2607 //
2608 // LS_get_add:
2609 //
2610 // int getAndAddInt( Object o, long offset, int delta)
2611 // long getAndAddLong(Object o, long offset, long delta)
2612 //
2613 // LS_get_set:
2614 //
2615 // int getAndSet(Object o, long offset, int newValue)
2616 // long getAndSet(Object o, long offset, long newValue)
2617 // Object getAndSet(Object o, long offset, Object newValue)
2618 //
2619 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2620 // This basic scheme here is the same as inline_unsafe_access, but
2621 // differs in enough details that combining them would make the code
2622 // overly confusing. (This is a true fact! I originally combined
2623 // them, but even I was confused by it!) As much code/comments as
2624 // possible are retained from inline_unsafe_access though to make
2625 // the correspondences clearer. - dl
2626
2627 if (callee()->is_static()) return false; // caller must have the capability!
2628
2629 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2630 decorators |= mo_decorator_for_access_kind(access_kind);
2631
2632 #ifndef PRODUCT
2633 BasicType rtype;
2634 {
2635 ResourceMark rm;
2636 // Check the signatures.
2637 ciSignature* sig = callee()->signature();
2638 rtype = sig->return_type()->basic_type();
2639 switch(kind) {
2640 case LS_get_add:
2641 case LS_get_set: {
2642 // Check the signatures.
2643 #ifdef ASSERT
2644 assert(rtype == type, "get and set must return the expected type");
2645 assert(sig->count() == 3, "get and set has 3 arguments");
2646 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2647 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2648 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2649 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2650 #endif // ASSERT
2651 break;
2652 }
2653 case LS_cmp_swap:
2654 case LS_cmp_swap_weak: {
2655 // Check the signatures.
2656 #ifdef ASSERT
2657 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2658 assert(sig->count() == 4, "CAS has 4 arguments");
2659 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2660 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2661 #endif // ASSERT
2662 break;
2663 }
2664 case LS_cmp_exchange: {
2665 // Check the signatures.
2666 #ifdef ASSERT
2667 assert(rtype == type, "CAS must return the expected type");
2668 assert(sig->count() == 4, "CAS has 4 arguments");
2669 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2670 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2671 #endif // ASSERT
2672 break;
2673 }
2674 default:
2675 ShouldNotReachHere();
2676 }
2677 }
2678 #endif //PRODUCT
2679
2680 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2681
2682 // Get arguments:
2683 Node* receiver = nullptr;
2684 Node* base = nullptr;
2685 Node* offset = nullptr;
2686 Node* oldval = nullptr;
2687 Node* newval = nullptr;
2688 switch(kind) {
2689 case LS_cmp_swap:
2690 case LS_cmp_swap_weak:
2691 case LS_cmp_exchange: {
2692 const bool two_slot_type = type2size[type] == 2;
2693 receiver = argument(0); // type: oop
2694 base = argument(1); // type: oop
2695 offset = argument(2); // type: long
2696 oldval = argument(4); // type: oop, int, or long
2697 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2698 break;
2699 }
2700 case LS_get_add:
2701 case LS_get_set: {
2702 receiver = argument(0); // type: oop
2703 base = argument(1); // type: oop
2704 offset = argument(2); // type: long
2705 oldval = nullptr;
2706 newval = argument(4); // type: oop, int, or long
2707 break;
2708 }
2709 default:
2710 ShouldNotReachHere();
2711 }
2712
2713 // Build field offset expression.
2714 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2715 // to be plain byte offsets, which are also the same as those accepted
2716 // by oopDesc::field_addr.
2717 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2718 // 32-bit machines ignore the high half of long offsets
2719 offset = ConvL2X(offset);
2720 // Save state and restore on bailout
2721 uint old_sp = sp();
2722 SafePointNode* old_map = clone_map();
2723 Node* adr = make_unsafe_address(base, offset,type, false);
2724 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2725
2726 Compile::AliasType* alias_type = C->alias_type(adr_type);
2727 BasicType bt = alias_type->basic_type();
2728 if (bt != T_ILLEGAL &&
2729 (is_reference_type(bt) != (type == T_OBJECT))) {
2730 // Don't intrinsify mismatched object accesses.
2731 set_map(old_map);
2732 set_sp(old_sp);
2733 return false;
2734 }
2735
2736 destruct_map_clone(old_map);
2737
2738 // For CAS, unlike inline_unsafe_access, there seems no point in
2739 // trying to refine types. Just use the coarse types here.
2740 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2741 const Type *value_type = Type::get_const_basic_type(type);
2742
2743 switch (kind) {
2744 case LS_get_set:
2745 case LS_cmp_exchange: {
2746 if (type == T_OBJECT) {
2747 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2748 if (tjp != nullptr) {
2749 value_type = tjp;
2750 }
2751 }
2752 break;
2753 }
2754 case LS_cmp_swap:
2755 case LS_cmp_swap_weak:
2756 case LS_get_add:
2757 break;
2758 default:
2759 ShouldNotReachHere();
2760 }
2761
2762 // Null check receiver.
2763 receiver = null_check(receiver);
2764 if (stopped()) {
2765 return true;
2766 }
2767
2768 int alias_idx = C->get_alias_index(adr_type);
2769
2770 if (is_reference_type(type)) {
2771 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2772
2773 // Transformation of a value which could be null pointer (CastPP #null)
2774 // could be delayed during Parse (for example, in adjust_map_after_if()).
2775 // Execute transformation here to avoid barrier generation in such case.
2776 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2777 newval = _gvn.makecon(TypePtr::NULL_PTR);
2778
2779 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2780 // Refine the value to a null constant, when it is known to be null
2781 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2782 }
2783 }
2784
2785 Node* result = nullptr;
2786 switch (kind) {
2787 case LS_cmp_exchange: {
2788 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2789 oldval, newval, value_type, type, decorators);
2790 break;
2791 }
2792 case LS_cmp_swap_weak:
2793 decorators |= C2_WEAK_CMPXCHG;
2794 case LS_cmp_swap: {
2795 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2796 oldval, newval, value_type, type, decorators);
2797 break;
2798 }
2799 case LS_get_set: {
2800 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2801 newval, value_type, type, decorators);
2802 break;
2803 }
2804 case LS_get_add: {
2805 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2806 newval, value_type, type, decorators);
2807 break;
2808 }
2809 default:
2810 ShouldNotReachHere();
2811 }
2812
2813 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2814 set_result(result);
2815 return true;
2816 }
2817
2818 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2819 // Regardless of form, don't allow previous ld/st to move down,
2820 // then issue acquire, release, or volatile mem_bar.
2821 insert_mem_bar(Op_MemBarCPUOrder);
2822 switch(id) {
2823 case vmIntrinsics::_loadFence:
2824 insert_mem_bar(Op_LoadFence);
2825 return true;
2826 case vmIntrinsics::_storeFence:
2827 insert_mem_bar(Op_StoreFence);
2828 return true;
2829 case vmIntrinsics::_storeStoreFence:
2830 insert_mem_bar(Op_StoreStoreFence);
2831 return true;
2832 case vmIntrinsics::_fullFence:
2833 insert_mem_bar(Op_MemBarVolatile);
2834 return true;
2835 default:
2836 fatal_unexpected_iid(id);
2837 return false;
2838 }
2839 }
2840
2841 bool LibraryCallKit::inline_onspinwait() {
2842 insert_mem_bar(Op_OnSpinWait);
2843 return true;
2844 }
2845
2846 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2847 if (!kls->is_Con()) {
2848 return true;
2849 }
2850 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2851 if (klsptr == nullptr) {
2852 return true;
2853 }
2854 ciInstanceKlass* ik = klsptr->instance_klass();
2855 // don't need a guard for a klass that is already initialized
2856 return !ik->is_initialized();
2857 }
2858
2859 //----------------------------inline_unsafe_writeback0-------------------------
2860 // public native void Unsafe.writeback0(long address)
2861 bool LibraryCallKit::inline_unsafe_writeback0() {
2862 if (!Matcher::has_match_rule(Op_CacheWB)) {
2863 return false;
2864 }
2865 #ifndef PRODUCT
2866 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2867 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2868 ciSignature* sig = callee()->signature();
2869 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2870 #endif
2871 null_check_receiver(); // null-check, then ignore
2872 Node *addr = argument(1);
2873 addr = new CastX2PNode(addr);
2874 addr = _gvn.transform(addr);
2875 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2876 flush = _gvn.transform(flush);
2877 set_memory(flush, TypeRawPtr::BOTTOM);
2878 return true;
2879 }
2880
2881 //----------------------------inline_unsafe_writeback0-------------------------
2882 // public native void Unsafe.writeback0(long address)
2883 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2884 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2885 return false;
2886 }
2887 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2888 return false;
2889 }
2890 #ifndef PRODUCT
2891 assert(Matcher::has_match_rule(Op_CacheWB),
2892 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2893 : "found match rule for CacheWBPostSync but not CacheWB"));
2894
2895 #endif
2896 null_check_receiver(); // null-check, then ignore
2897 Node *sync;
2898 if (is_pre) {
2899 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2900 } else {
2901 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2902 }
2903 sync = _gvn.transform(sync);
2904 set_memory(sync, TypeRawPtr::BOTTOM);
2905 return true;
2906 }
2907
2908 //----------------------------inline_unsafe_allocate---------------------------
2909 // public native Object Unsafe.allocateInstance(Class<?> cls);
2910 bool LibraryCallKit::inline_unsafe_allocate() {
2911
2912 #if INCLUDE_JVMTI
2913 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2914 return false;
2915 }
2916 #endif //INCLUDE_JVMTI
2917
2918 if (callee()->is_static()) return false; // caller must have the capability!
2919
2920 null_check_receiver(); // null-check, then ignore
2921 Node* cls = null_check(argument(1));
2922 if (stopped()) return true;
2923
2924 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2925 kls = null_check(kls);
2926 if (stopped()) return true; // argument was like int.class
2927
2928 #if INCLUDE_JVMTI
2929 // Don't try to access new allocated obj in the intrinsic.
2930 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2931 // Deoptimize and allocate in interpreter instead.
2932 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2933 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2934 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2935 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2936 {
2937 BuildCutout unless(this, tst, PROB_MAX);
2938 uncommon_trap(Deoptimization::Reason_intrinsic,
2939 Deoptimization::Action_make_not_entrant);
2940 }
2941 if (stopped()) {
2942 return true;
2943 }
2944 #endif //INCLUDE_JVMTI
2945
2946 Node* test = nullptr;
2947 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2948 // Note: The argument might still be an illegal value like
2949 // Serializable.class or Object[].class. The runtime will handle it.
2950 // But we must make an explicit check for initialization.
2951 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2952 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2953 // can generate code to load it as unsigned byte.
2954 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2955 Node* bits = intcon(InstanceKlass::fully_initialized);
2956 test = _gvn.transform(new SubINode(inst, bits));
2957 // The 'test' is non-zero if we need to take a slow path.
2958 }
2959
2960 Node* obj = new_instance(kls, test);
2961 set_result(obj);
2962 return true;
2963 }
2964
2965 //------------------------inline_native_time_funcs--------------
2966 // inline code for System.currentTimeMillis() and System.nanoTime()
2967 // these have the same type and signature
2968 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2969 const TypeFunc* tf = OptoRuntime::void_long_Type();
2970 const TypePtr* no_memory_effects = nullptr;
2971 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2972 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2973 #ifdef ASSERT
2974 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2975 assert(value_top == top(), "second value must be top");
2976 #endif
2977 set_result(value);
2978 return true;
2979 }
2980
2981
2982 #if INCLUDE_JVMTI
2983
2984 // When notifications are disabled then just update the VTMS transition bit and return.
2985 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2986 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2987 if (!DoJVMTIVirtualThreadTransitions) {
2988 return true;
2989 }
2990 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2991 IdealKit ideal(this);
2992
2993 Node* ONE = ideal.ConI(1);
2994 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2995 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2996 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2997
2998 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2999 sync_kit(ideal);
3000 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3001 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3002 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3003 ideal.sync_kit(this);
3004 } ideal.else_(); {
3005 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3006 Node* thread = ideal.thread();
3007 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3008 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3009
3010 sync_kit(ideal);
3011 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3012 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3013
3014 ideal.sync_kit(this);
3015 } ideal.end_if();
3016 final_sync(ideal);
3017
3018 return true;
3019 }
3020
3021 // Always update the is_disable_suspend bit.
3022 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3023 if (!DoJVMTIVirtualThreadTransitions) {
3024 return true;
3025 }
3026 IdealKit ideal(this);
3027
3028 {
3029 // unconditionally update the is_disable_suspend bit in current JavaThread
3030 Node* thread = ideal.thread();
3031 Node* arg = _gvn.transform(argument(0)); // argument for notification
3032 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3033 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3034
3035 sync_kit(ideal);
3036 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3037 ideal.sync_kit(this);
3038 }
3039 final_sync(ideal);
3040
3041 return true;
3042 }
3043
3044 #endif // INCLUDE_JVMTI
3045
3046 #ifdef JFR_HAVE_INTRINSICS
3047
3048 /**
3049 * if oop->klass != null
3050 * // normal class
3051 * epoch = _epoch_state ? 2 : 1
3052 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3053 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3054 * }
3055 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3056 * else
3057 * // primitive class
3058 * if oop->array_klass != null
3059 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3060 * else
3061 * id = LAST_TYPE_ID + 1 // void class path
3062 * if (!signaled)
3063 * signaled = true
3064 */
3065 bool LibraryCallKit::inline_native_classID() {
3066 Node* cls = argument(0);
3067
3068 IdealKit ideal(this);
3069 #define __ ideal.
3070 IdealVariable result(ideal); __ declarations_done();
3071 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3072 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3073 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3074
3075
3076 __ if_then(kls, BoolTest::ne, null()); {
3077 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3078 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3079
3080 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3081 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3082 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3083 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3084 mask = _gvn.transform(new OrLNode(mask, epoch));
3085 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3086
3087 float unlikely = PROB_UNLIKELY(0.999);
3088 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3089 sync_kit(ideal);
3090 make_runtime_call(RC_LEAF,
3091 OptoRuntime::class_id_load_barrier_Type(),
3092 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3093 "class id load barrier",
3094 TypePtr::BOTTOM,
3095 kls);
3096 ideal.sync_kit(this);
3097 } __ end_if();
3098
3099 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3100 } __ else_(); {
3101 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3102 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3103 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3104 __ if_then(array_kls, BoolTest::ne, null()); {
3105 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3106 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3107 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3108 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3109 } __ else_(); {
3110 // void class case
3111 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3112 } __ end_if();
3113
3114 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3115 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3116 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3117 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3118 } __ end_if();
3119 } __ end_if();
3120
3121 final_sync(ideal);
3122 set_result(ideal.value(result));
3123 #undef __
3124 return true;
3125 }
3126
3127 //------------------------inline_native_jvm_commit------------------
3128 bool LibraryCallKit::inline_native_jvm_commit() {
3129 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3130
3131 // Save input memory and i_o state.
3132 Node* input_memory_state = reset_memory();
3133 set_all_memory(input_memory_state);
3134 Node* input_io_state = i_o();
3135
3136 // TLS.
3137 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3138 // Jfr java buffer.
3139 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3140 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3141 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3142
3143 // Load the current value of the notified field in the JfrThreadLocal.
3144 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3145 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3146
3147 // Test for notification.
3148 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3149 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3150 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3151
3152 // True branch, is notified.
3153 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3154 set_control(is_notified);
3155
3156 // Reset notified state.
3157 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3158
3159 // 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.
3160 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3161 // Convert the machine-word to a long.
3162 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3163
3164 // False branch, not notified.
3165 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3166 set_control(not_notified);
3167 set_all_memory(input_memory_state);
3168
3169 // Arg is the next position as a long.
3170 Node* arg = argument(0);
3171 // Convert long to machine-word.
3172 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3173
3174 // Store the next_position to the underlying jfr java buffer.
3175 Node* commit_memory;
3176 #ifdef _LP64
3177 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, MemNode::release);
3178 #else
3179 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, MemNode::release);
3180 #endif
3181
3182 // 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.
3183 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3184 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3185 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3186
3187 // And flags with lease constant.
3188 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3189
3190 // Branch on lease to conditionalize returning the leased java buffer.
3191 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3192 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3193 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3194
3195 // False branch, not a lease.
3196 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3197
3198 // True branch, is lease.
3199 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3200 set_control(is_lease);
3201
3202 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3203 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3204 OptoRuntime::void_void_Type(),
3205 SharedRuntime::jfr_return_lease(),
3206 "return_lease", TypePtr::BOTTOM);
3207 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3208
3209 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3210 record_for_igvn(lease_compare_rgn);
3211 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3212 record_for_igvn(lease_compare_mem);
3213 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3214 record_for_igvn(lease_compare_io);
3215 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3216 record_for_igvn(lease_result_value);
3217
3218 // Update control and phi nodes.
3219 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3220 lease_compare_rgn->init_req(_false_path, not_lease);
3221
3222 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3223 lease_compare_mem->init_req(_false_path, commit_memory);
3224
3225 lease_compare_io->init_req(_true_path, i_o());
3226 lease_compare_io->init_req(_false_path, input_io_state);
3227
3228 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3229 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3230
3231 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3232 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3233 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3234 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3235
3236 // Update control and phi nodes.
3237 result_rgn->init_req(_true_path, is_notified);
3238 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3239
3240 result_mem->init_req(_true_path, notified_reset_memory);
3241 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3242
3243 result_io->init_req(_true_path, input_io_state);
3244 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3245
3246 result_value->init_req(_true_path, current_pos);
3247 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3248
3249 // Set output state.
3250 set_control(_gvn.transform(result_rgn));
3251 set_all_memory(_gvn.transform(result_mem));
3252 set_i_o(_gvn.transform(result_io));
3253 set_result(result_rgn, result_value);
3254 return true;
3255 }
3256
3257 /*
3258 * The intrinsic is a model of this pseudo-code:
3259 *
3260 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3261 * jobject h_event_writer = tl->java_event_writer();
3262 * if (h_event_writer == nullptr) {
3263 * return nullptr;
3264 * }
3265 * oop threadObj = Thread::threadObj();
3266 * oop vthread = java_lang_Thread::vthread(threadObj);
3267 * traceid tid;
3268 * bool pinVirtualThread;
3269 * bool excluded;
3270 * if (vthread != threadObj) { // i.e. current thread is virtual
3271 * tid = java_lang_Thread::tid(vthread);
3272 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3273 * pinVirtualThread = VMContinuations;
3274 * excluded = vthread_epoch_raw & excluded_mask;
3275 * if (!excluded) {
3276 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3277 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3278 * if (vthread_epoch != current_epoch) {
3279 * write_checkpoint();
3280 * }
3281 * }
3282 * } else {
3283 * tid = java_lang_Thread::tid(threadObj);
3284 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3285 * pinVirtualThread = false;
3286 * excluded = thread_epoch_raw & excluded_mask;
3287 * }
3288 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3289 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3290 * if (tid_in_event_writer != tid) {
3291 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3292 * setField(event_writer, "excluded", excluded);
3293 * setField(event_writer, "threadID", tid);
3294 * }
3295 * return event_writer
3296 */
3297 bool LibraryCallKit::inline_native_getEventWriter() {
3298 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3299
3300 // Save input memory and i_o state.
3301 Node* input_memory_state = reset_memory();
3302 set_all_memory(input_memory_state);
3303 Node* input_io_state = i_o();
3304
3305 // The most significant bit of the u2 is used to denote thread exclusion
3306 Node* excluded_shift = _gvn.intcon(15);
3307 Node* excluded_mask = _gvn.intcon(1 << 15);
3308 // The epoch generation is the range [1-32767]
3309 Node* epoch_mask = _gvn.intcon(32767);
3310
3311 // TLS
3312 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3313
3314 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3315 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3316
3317 // Load the eventwriter jobject handle.
3318 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3319
3320 // Null check the jobject handle.
3321 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3322 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3323 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3324
3325 // False path, jobj is null.
3326 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3327
3328 // True path, jobj is not null.
3329 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3330
3331 set_control(jobj_is_not_null);
3332
3333 // Load the threadObj for the CarrierThread.
3334 Node* threadObj = generate_current_thread(tls_ptr);
3335
3336 // Load the vthread.
3337 Node* vthread = generate_virtual_thread(tls_ptr);
3338
3339 // If vthread != threadObj, this is a virtual thread.
3340 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3341 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3342 IfNode* iff_vthread_not_equal_threadObj =
3343 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3344
3345 // False branch, fallback to threadObj.
3346 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3347 set_control(vthread_equal_threadObj);
3348
3349 // Load the tid field from the vthread object.
3350 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3351
3352 // Load the raw epoch value from the threadObj.
3353 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3354 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3355 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3356 TypeInt::CHAR, T_CHAR,
3357 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3358
3359 // Mask off the excluded information from the epoch.
3360 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3361
3362 // True branch, this is a virtual thread.
3363 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3364 set_control(vthread_not_equal_threadObj);
3365
3366 // Load the tid field from the vthread object.
3367 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3368
3369 // Continuation support determines if a virtual thread should be pinned.
3370 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3371 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3372
3373 // Load the raw epoch value from the vthread.
3374 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3375 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3376 TypeInt::CHAR, T_CHAR,
3377 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3378
3379 // Mask off the excluded information from the epoch.
3380 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3381
3382 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3383 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3384 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3385 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3386
3387 // False branch, vthread is excluded, no need to write epoch info.
3388 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3389
3390 // True branch, vthread is included, update epoch info.
3391 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3392 set_control(included);
3393
3394 // Get epoch value.
3395 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3396
3397 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3398 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3399 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3400
3401 // Compare the epoch in the vthread to the current epoch generation.
3402 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3403 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3404 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3405
3406 // False path, epoch is equal, checkpoint information is valid.
3407 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3408
3409 // True path, epoch is not equal, write a checkpoint for the vthread.
3410 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3411
3412 set_control(epoch_is_not_equal);
3413
3414 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3415 // The call also updates the native thread local thread id and the vthread with the current epoch.
3416 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3417 OptoRuntime::jfr_write_checkpoint_Type(),
3418 SharedRuntime::jfr_write_checkpoint(),
3419 "write_checkpoint", TypePtr::BOTTOM);
3420 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3421
3422 // vthread epoch != current epoch
3423 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3424 record_for_igvn(epoch_compare_rgn);
3425 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3426 record_for_igvn(epoch_compare_mem);
3427 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3428 record_for_igvn(epoch_compare_io);
3429
3430 // Update control and phi nodes.
3431 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3432 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3433 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3434 epoch_compare_mem->init_req(_false_path, input_memory_state);
3435 epoch_compare_io->init_req(_true_path, i_o());
3436 epoch_compare_io->init_req(_false_path, input_io_state);
3437
3438 // excluded != true
3439 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3440 record_for_igvn(exclude_compare_rgn);
3441 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3442 record_for_igvn(exclude_compare_mem);
3443 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3444 record_for_igvn(exclude_compare_io);
3445
3446 // Update control and phi nodes.
3447 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3448 exclude_compare_rgn->init_req(_false_path, excluded);
3449 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3450 exclude_compare_mem->init_req(_false_path, input_memory_state);
3451 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3452 exclude_compare_io->init_req(_false_path, input_io_state);
3453
3454 // vthread != threadObj
3455 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3456 record_for_igvn(vthread_compare_rgn);
3457 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3458 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3459 record_for_igvn(vthread_compare_io);
3460 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3461 record_for_igvn(tid);
3462 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3463 record_for_igvn(exclusion);
3464 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3465 record_for_igvn(pinVirtualThread);
3466
3467 // Update control and phi nodes.
3468 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3469 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3470 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3471 vthread_compare_mem->init_req(_false_path, input_memory_state);
3472 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3473 vthread_compare_io->init_req(_false_path, input_io_state);
3474 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3475 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3476 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3477 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3478 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3479 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3480
3481 // Update branch state.
3482 set_control(_gvn.transform(vthread_compare_rgn));
3483 set_all_memory(_gvn.transform(vthread_compare_mem));
3484 set_i_o(_gvn.transform(vthread_compare_io));
3485
3486 // Load the event writer oop by dereferencing the jobject handle.
3487 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3488 assert(klass_EventWriter->is_loaded(), "invariant");
3489 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3490 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3491 const TypeOopPtr* const xtype = aklass->as_instance_type();
3492 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3493 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3494
3495 // Load the current thread id from the event writer object.
3496 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3497 // Get the field offset to, conditionally, store an updated tid value later.
3498 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3499 // Get the field offset to, conditionally, store an updated exclusion value later.
3500 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3501 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3502 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3503
3504 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3505 record_for_igvn(event_writer_tid_compare_rgn);
3506 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3507 record_for_igvn(event_writer_tid_compare_mem);
3508 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3509 record_for_igvn(event_writer_tid_compare_io);
3510
3511 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3512 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3513 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3514 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3515
3516 // False path, tids are the same.
3517 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3518
3519 // True path, tid is not equal, need to update the tid in the event writer.
3520 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3521 record_for_igvn(tid_is_not_equal);
3522
3523 // Store the pin state to the event writer.
3524 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3525
3526 // Store the exclusion state to the event writer.
3527 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3528 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3529
3530 // Store the tid to the event writer.
3531 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3532
3533 // Update control and phi nodes.
3534 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3535 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3536 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3537 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3538 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3539 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3540
3541 // Result of top level CFG, Memory, IO and Value.
3542 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3543 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3544 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3545 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3546
3547 // Result control.
3548 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3549 result_rgn->init_req(_false_path, jobj_is_null);
3550
3551 // Result memory.
3552 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3553 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3554
3555 // Result IO.
3556 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3557 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3558
3559 // Result value.
3560 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3561 result_value->init_req(_false_path, null()); // return null
3562
3563 // Set output state.
3564 set_control(_gvn.transform(result_rgn));
3565 set_all_memory(_gvn.transform(result_mem));
3566 set_i_o(_gvn.transform(result_io));
3567 set_result(result_rgn, result_value);
3568 return true;
3569 }
3570
3571 /*
3572 * The intrinsic is a model of this pseudo-code:
3573 *
3574 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3575 * if (carrierThread != thread) { // is virtual thread
3576 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3577 * bool excluded = vthread_epoch_raw & excluded_mask;
3578 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3579 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3580 * if (!excluded) {
3581 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3582 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3583 * }
3584 * Atomic::release_store(&tl->_vthread, true);
3585 * return;
3586 * }
3587 * Atomic::release_store(&tl->_vthread, false);
3588 */
3589 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3590 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3591
3592 Node* input_memory_state = reset_memory();
3593 set_all_memory(input_memory_state);
3594
3595 // The most significant bit of the u2 is used to denote thread exclusion
3596 Node* excluded_mask = _gvn.intcon(1 << 15);
3597 // The epoch generation is the range [1-32767]
3598 Node* epoch_mask = _gvn.intcon(32767);
3599
3600 Node* const carrierThread = generate_current_thread(jt);
3601 // If thread != carrierThread, this is a virtual thread.
3602 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3603 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3604 IfNode* iff_thread_not_equal_carrierThread =
3605 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3606
3607 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3608
3609 // False branch, is carrierThread.
3610 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3611 // Store release
3612 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3613
3614 set_all_memory(input_memory_state);
3615
3616 // True branch, is virtual thread.
3617 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3618 set_control(thread_not_equal_carrierThread);
3619
3620 // Load the raw epoch value from the vthread.
3621 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3622 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3623 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3624
3625 // Mask off the excluded information from the epoch.
3626 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3627
3628 // Load the tid field from the thread.
3629 Node* tid = load_field_from_object(thread, "tid", "J");
3630
3631 // Store the vthread tid to the jfr thread local.
3632 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3633 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3634
3635 // Branch is_excluded to conditionalize updating the epoch .
3636 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3637 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3638 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3639
3640 // True branch, vthread is excluded, no need to write epoch info.
3641 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3642 set_control(excluded);
3643 Node* vthread_is_excluded = _gvn.intcon(1);
3644
3645 // False branch, vthread is included, update epoch info.
3646 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3647 set_control(included);
3648 Node* vthread_is_included = _gvn.intcon(0);
3649
3650 // Get epoch value.
3651 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3652
3653 // Store the vthread epoch to the jfr thread local.
3654 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3655 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3656
3657 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3658 record_for_igvn(excluded_rgn);
3659 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3660 record_for_igvn(excluded_mem);
3661 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3662 record_for_igvn(exclusion);
3663
3664 // Merge the excluded control and memory.
3665 excluded_rgn->init_req(_true_path, excluded);
3666 excluded_rgn->init_req(_false_path, included);
3667 excluded_mem->init_req(_true_path, tid_memory);
3668 excluded_mem->init_req(_false_path, included_memory);
3669 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3670 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3671
3672 // Set intermediate state.
3673 set_control(_gvn.transform(excluded_rgn));
3674 set_all_memory(excluded_mem);
3675
3676 // Store the vthread exclusion state to the jfr thread local.
3677 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3678 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3679
3680 // Store release
3681 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3682
3683 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3684 record_for_igvn(thread_compare_rgn);
3685 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3686 record_for_igvn(thread_compare_mem);
3687 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3688 record_for_igvn(vthread);
3689
3690 // Merge the thread_compare control and memory.
3691 thread_compare_rgn->init_req(_true_path, control());
3692 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3693 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3694 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3695
3696 // Set output state.
3697 set_control(_gvn.transform(thread_compare_rgn));
3698 set_all_memory(_gvn.transform(thread_compare_mem));
3699 }
3700
3701 #endif // JFR_HAVE_INTRINSICS
3702
3703 //------------------------inline_native_currentCarrierThread------------------
3704 bool LibraryCallKit::inline_native_currentCarrierThread() {
3705 Node* junk = nullptr;
3706 set_result(generate_current_thread(junk));
3707 return true;
3708 }
3709
3710 //------------------------inline_native_currentThread------------------
3711 bool LibraryCallKit::inline_native_currentThread() {
3712 Node* junk = nullptr;
3713 set_result(generate_virtual_thread(junk));
3714 return true;
3715 }
3716
3717 //------------------------inline_native_setVthread------------------
3718 bool LibraryCallKit::inline_native_setCurrentThread() {
3719 assert(C->method()->changes_current_thread(),
3720 "method changes current Thread but is not annotated ChangesCurrentThread");
3721 Node* arr = argument(1);
3722 Node* thread = _gvn.transform(new ThreadLocalNode());
3723 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3724 Node* thread_obj_handle
3725 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3726 thread_obj_handle = _gvn.transform(thread_obj_handle);
3727 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3728 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3729
3730 // Change the _monitor_owner_id of the JavaThread
3731 Node* tid = load_field_from_object(arr, "tid", "J");
3732 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3733 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3734
3735 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3736 return true;
3737 }
3738
3739 const Type* LibraryCallKit::scopedValueCache_type() {
3740 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3741 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3742 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3743
3744 // Because we create the scopedValue cache lazily we have to make the
3745 // type of the result BotPTR.
3746 bool xk = etype->klass_is_exact();
3747 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3748 return objects_type;
3749 }
3750
3751 Node* LibraryCallKit::scopedValueCache_helper() {
3752 Node* thread = _gvn.transform(new ThreadLocalNode());
3753 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3754 // We cannot use immutable_memory() because we might flip onto a
3755 // different carrier thread, at which point we'll need to use that
3756 // carrier thread's cache.
3757 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3758 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3759 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3760 }
3761
3762 //------------------------inline_native_scopedValueCache------------------
3763 bool LibraryCallKit::inline_native_scopedValueCache() {
3764 Node* cache_obj_handle = scopedValueCache_helper();
3765 const Type* objects_type = scopedValueCache_type();
3766 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3767
3768 return true;
3769 }
3770
3771 //------------------------inline_native_setScopedValueCache------------------
3772 bool LibraryCallKit::inline_native_setScopedValueCache() {
3773 Node* arr = argument(0);
3774 Node* cache_obj_handle = scopedValueCache_helper();
3775 const Type* objects_type = scopedValueCache_type();
3776
3777 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3778 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3779
3780 return true;
3781 }
3782
3783 //------------------------inline_native_Continuation_pin and unpin-----------
3784
3785 // Shared implementation routine for both pin and unpin.
3786 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3787 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3788
3789 // Save input memory.
3790 Node* input_memory_state = reset_memory();
3791 set_all_memory(input_memory_state);
3792
3793 // TLS
3794 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3795 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3796 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3797
3798 // Null check the last continuation object.
3799 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3800 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3801 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3802
3803 // False path, last continuation is null.
3804 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3805
3806 // True path, last continuation is not null.
3807 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3808
3809 set_control(continuation_is_not_null);
3810
3811 // Load the pin count from the last continuation.
3812 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3813 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3814
3815 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3816 Node* pin_count_rhs;
3817 if (unpin) {
3818 pin_count_rhs = _gvn.intcon(0);
3819 } else {
3820 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3821 }
3822 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3823 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3824 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3825
3826 // True branch, pin count over/underflow.
3827 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3828 {
3829 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3830 // which will throw IllegalStateException for pin count over/underflow.
3831 // No memory changed so far - we can use memory create by reset_memory()
3832 // at the beginning of this intrinsic. No need to call reset_memory() again.
3833 PreserveJVMState pjvms(this);
3834 set_control(pin_count_over_underflow);
3835 uncommon_trap(Deoptimization::Reason_intrinsic,
3836 Deoptimization::Action_none);
3837 assert(stopped(), "invariant");
3838 }
3839
3840 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3841 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3842 set_control(valid_pin_count);
3843
3844 Node* next_pin_count;
3845 if (unpin) {
3846 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3847 } else {
3848 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3849 }
3850
3851 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3852
3853 // Result of top level CFG and Memory.
3854 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3855 record_for_igvn(result_rgn);
3856 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3857 record_for_igvn(result_mem);
3858
3859 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3860 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3861 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3862 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3863
3864 // Set output state.
3865 set_control(_gvn.transform(result_rgn));
3866 set_all_memory(_gvn.transform(result_mem));
3867
3868 return true;
3869 }
3870
3871 //---------------------------load_mirror_from_klass----------------------------
3872 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3873 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3874 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3875 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3876 // mirror = ((OopHandle)mirror)->resolve();
3877 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3878 }
3879
3880 //-----------------------load_klass_from_mirror_common-------------------------
3881 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3882 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3883 // and branch to the given path on the region.
3884 // If never_see_null, take an uncommon trap on null, so we can optimistically
3885 // compile for the non-null case.
3886 // If the region is null, force never_see_null = true.
3887 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3888 bool never_see_null,
3889 RegionNode* region,
3890 int null_path,
3891 int offset) {
3892 if (region == nullptr) never_see_null = true;
3893 Node* p = basic_plus_adr(mirror, offset);
3894 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3895 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3896 Node* null_ctl = top();
3897 kls = null_check_oop(kls, &null_ctl, never_see_null);
3898 if (region != nullptr) {
3899 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3900 region->init_req(null_path, null_ctl);
3901 } else {
3902 assert(null_ctl == top(), "no loose ends");
3903 }
3904 return kls;
3905 }
3906
3907 //--------------------(inline_native_Class_query helpers)---------------------
3908 // Use this for JVM_ACC_INTERFACE.
3909 // Fall through if (mods & mask) == bits, take the guard otherwise.
3910 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3911 ByteSize offset, const Type* type, BasicType bt) {
3912 // Branch around if the given klass has the given modifier bit set.
3913 // Like generate_guard, adds a new path onto the region.
3914 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3915 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3916 Node* mask = intcon(modifier_mask);
3917 Node* bits = intcon(modifier_bits);
3918 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3919 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3920 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3921 return generate_fair_guard(bol, region);
3922 }
3923 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3924 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3925 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3926 }
3927
3928 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3929 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3930 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3931 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3932 }
3933
3934 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3935 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3936 }
3937
3938 //-------------------------inline_native_Class_query-------------------
3939 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3940 const Type* return_type = TypeInt::BOOL;
3941 Node* prim_return_value = top(); // what happens if it's a primitive class?
3942 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3943 bool expect_prim = false; // most of these guys expect to work on refs
3944
3945 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3946
3947 Node* mirror = argument(0);
3948 Node* obj = top();
3949
3950 switch (id) {
3951 case vmIntrinsics::_isInstance:
3952 // nothing is an instance of a primitive type
3953 prim_return_value = intcon(0);
3954 obj = argument(1);
3955 break;
3956 case vmIntrinsics::_isHidden:
3957 prim_return_value = intcon(0);
3958 break;
3959 case vmIntrinsics::_getSuperclass:
3960 prim_return_value = null();
3961 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3962 break;
3963 case vmIntrinsics::_getClassAccessFlags:
3964 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3965 return_type = TypeInt::CHAR;
3966 break;
3967 default:
3968 fatal_unexpected_iid(id);
3969 break;
3970 }
3971
3972 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3973 if (mirror_con == nullptr) return false; // cannot happen?
3974
3975 #ifndef PRODUCT
3976 if (C->print_intrinsics() || C->print_inlining()) {
3977 ciType* k = mirror_con->java_mirror_type();
3978 if (k) {
3979 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3980 k->print_name();
3981 tty->cr();
3982 }
3983 }
3984 #endif
3985
3986 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3987 RegionNode* region = new RegionNode(PATH_LIMIT);
3988 record_for_igvn(region);
3989 PhiNode* phi = new PhiNode(region, return_type);
3990
3991 // The mirror will never be null of Reflection.getClassAccessFlags, however
3992 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3993 // if it is. See bug 4774291.
3994
3995 // For Reflection.getClassAccessFlags(), the null check occurs in
3996 // the wrong place; see inline_unsafe_access(), above, for a similar
3997 // situation.
3998 mirror = null_check(mirror);
3999 // If mirror or obj is dead, only null-path is taken.
4000 if (stopped()) return true;
4001
4002 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4003
4004 // Now load the mirror's klass metaobject, and null-check it.
4005 // Side-effects region with the control path if the klass is null.
4006 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4007 // If kls is null, we have a primitive mirror.
4008 phi->init_req(_prim_path, prim_return_value);
4009 if (stopped()) { set_result(region, phi); return true; }
4010 bool safe_for_replace = (region->in(_prim_path) == top());
4011
4012 Node* p; // handy temp
4013 Node* null_ctl;
4014
4015 // Now that we have the non-null klass, we can perform the real query.
4016 // For constant classes, the query will constant-fold in LoadNode::Value.
4017 Node* query_value = top();
4018 switch (id) {
4019 case vmIntrinsics::_isInstance:
4020 // nothing is an instance of a primitive type
4021 query_value = gen_instanceof(obj, kls, safe_for_replace);
4022 break;
4023
4024 case vmIntrinsics::_isHidden:
4025 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4026 if (generate_hidden_class_guard(kls, region) != nullptr)
4027 // A guard was added. If the guard is taken, it was an hidden class.
4028 phi->add_req(intcon(1));
4029 // If we fall through, it's a plain class.
4030 query_value = intcon(0);
4031 break;
4032
4033
4034 case vmIntrinsics::_getSuperclass:
4035 // The rules here are somewhat unfortunate, but we can still do better
4036 // with random logic than with a JNI call.
4037 // Interfaces store null or Object as _super, but must report null.
4038 // Arrays store an intermediate super as _super, but must report Object.
4039 // Other types can report the actual _super.
4040 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4041 if (generate_interface_guard(kls, region) != nullptr)
4042 // A guard was added. If the guard is taken, it was an interface.
4043 phi->add_req(null());
4044 if (generate_array_guard(kls, region) != nullptr)
4045 // A guard was added. If the guard is taken, it was an array.
4046 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4047 // If we fall through, it's a plain class. Get its _super.
4048 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4049 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4050 null_ctl = top();
4051 kls = null_check_oop(kls, &null_ctl);
4052 if (null_ctl != top()) {
4053 // If the guard is taken, Object.superClass is null (both klass and mirror).
4054 region->add_req(null_ctl);
4055 phi ->add_req(null());
4056 }
4057 if (!stopped()) {
4058 query_value = load_mirror_from_klass(kls);
4059 }
4060 break;
4061
4062 case vmIntrinsics::_getClassAccessFlags:
4063 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4064 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4065 break;
4066
4067 default:
4068 fatal_unexpected_iid(id);
4069 break;
4070 }
4071
4072 // Fall-through is the normal case of a query to a real class.
4073 phi->init_req(1, query_value);
4074 region->init_req(1, control());
4075
4076 C->set_has_split_ifs(true); // Has chance for split-if optimization
4077 set_result(region, phi);
4078 return true;
4079 }
4080
4081 //-------------------------inline_Class_cast-------------------
4082 bool LibraryCallKit::inline_Class_cast() {
4083 Node* mirror = argument(0); // Class
4084 Node* obj = argument(1);
4085 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4086 if (mirror_con == nullptr) {
4087 return false; // dead path (mirror->is_top()).
4088 }
4089 if (obj == nullptr || obj->is_top()) {
4090 return false; // dead path
4091 }
4092 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4093
4094 // First, see if Class.cast() can be folded statically.
4095 // java_mirror_type() returns non-null for compile-time Class constants.
4096 ciType* tm = mirror_con->java_mirror_type();
4097 if (tm != nullptr && tm->is_klass() &&
4098 tp != nullptr) {
4099 if (!tp->is_loaded()) {
4100 // Don't use intrinsic when class is not loaded.
4101 return false;
4102 } else {
4103 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4104 if (static_res == Compile::SSC_always_true) {
4105 // isInstance() is true - fold the code.
4106 set_result(obj);
4107 return true;
4108 } else if (static_res == Compile::SSC_always_false) {
4109 // Don't use intrinsic, have to throw ClassCastException.
4110 // If the reference is null, the non-intrinsic bytecode will
4111 // be optimized appropriately.
4112 return false;
4113 }
4114 }
4115 }
4116
4117 // Bailout intrinsic and do normal inlining if exception path is frequent.
4118 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4119 return false;
4120 }
4121
4122 // Generate dynamic checks.
4123 // Class.cast() is java implementation of _checkcast bytecode.
4124 // Do checkcast (Parse::do_checkcast()) optimizations here.
4125
4126 mirror = null_check(mirror);
4127 // If mirror is dead, only null-path is taken.
4128 if (stopped()) {
4129 return true;
4130 }
4131
4132 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4133 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4134 RegionNode* region = new RegionNode(PATH_LIMIT);
4135 record_for_igvn(region);
4136
4137 // Now load the mirror's klass metaobject, and null-check it.
4138 // If kls is null, we have a primitive mirror and
4139 // nothing is an instance of a primitive type.
4140 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4141
4142 Node* res = top();
4143 if (!stopped()) {
4144 Node* bad_type_ctrl = top();
4145 // Do checkcast optimizations.
4146 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4147 region->init_req(_bad_type_path, bad_type_ctrl);
4148 }
4149 if (region->in(_prim_path) != top() ||
4150 region->in(_bad_type_path) != top()) {
4151 // Let Interpreter throw ClassCastException.
4152 PreserveJVMState pjvms(this);
4153 set_control(_gvn.transform(region));
4154 uncommon_trap(Deoptimization::Reason_intrinsic,
4155 Deoptimization::Action_maybe_recompile);
4156 }
4157 if (!stopped()) {
4158 set_result(res);
4159 }
4160 return true;
4161 }
4162
4163
4164 //--------------------------inline_native_subtype_check------------------------
4165 // This intrinsic takes the JNI calls out of the heart of
4166 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4167 bool LibraryCallKit::inline_native_subtype_check() {
4168 // Pull both arguments off the stack.
4169 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4170 args[0] = argument(0);
4171 args[1] = argument(1);
4172 Node* klasses[2]; // corresponding Klasses: superk, subk
4173 klasses[0] = klasses[1] = top();
4174
4175 enum {
4176 // A full decision tree on {superc is prim, subc is prim}:
4177 _prim_0_path = 1, // {P,N} => false
4178 // {P,P} & superc!=subc => false
4179 _prim_same_path, // {P,P} & superc==subc => true
4180 _prim_1_path, // {N,P} => false
4181 _ref_subtype_path, // {N,N} & subtype check wins => true
4182 _both_ref_path, // {N,N} & subtype check loses => false
4183 PATH_LIMIT
4184 };
4185
4186 RegionNode* region = new RegionNode(PATH_LIMIT);
4187 Node* phi = new PhiNode(region, TypeInt::BOOL);
4188 record_for_igvn(region);
4189
4190 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4191 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4192 int class_klass_offset = java_lang_Class::klass_offset();
4193
4194 // First null-check both mirrors and load each mirror's klass metaobject.
4195 int which_arg;
4196 for (which_arg = 0; which_arg <= 1; which_arg++) {
4197 Node* arg = args[which_arg];
4198 arg = null_check(arg);
4199 if (stopped()) break;
4200 args[which_arg] = arg;
4201
4202 Node* p = basic_plus_adr(arg, class_klass_offset);
4203 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4204 klasses[which_arg] = _gvn.transform(kls);
4205 }
4206
4207 // Having loaded both klasses, test each for null.
4208 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4209 for (which_arg = 0; which_arg <= 1; which_arg++) {
4210 Node* kls = klasses[which_arg];
4211 Node* null_ctl = top();
4212 kls = null_check_oop(kls, &null_ctl, never_see_null);
4213 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4214 region->init_req(prim_path, null_ctl);
4215 if (stopped()) break;
4216 klasses[which_arg] = kls;
4217 }
4218
4219 if (!stopped()) {
4220 // now we have two reference types, in klasses[0..1]
4221 Node* subk = klasses[1]; // the argument to isAssignableFrom
4222 Node* superk = klasses[0]; // the receiver
4223 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4224 // now we have a successful reference subtype check
4225 region->set_req(_ref_subtype_path, control());
4226 }
4227
4228 // If both operands are primitive (both klasses null), then
4229 // we must return true when they are identical primitives.
4230 // It is convenient to test this after the first null klass check.
4231 set_control(region->in(_prim_0_path)); // go back to first null check
4232 if (!stopped()) {
4233 // Since superc is primitive, make a guard for the superc==subc case.
4234 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4235 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4236 generate_guard(bol_eq, region, PROB_FAIR);
4237 if (region->req() == PATH_LIMIT+1) {
4238 // A guard was added. If the added guard is taken, superc==subc.
4239 region->swap_edges(PATH_LIMIT, _prim_same_path);
4240 region->del_req(PATH_LIMIT);
4241 }
4242 region->set_req(_prim_0_path, control()); // Not equal after all.
4243 }
4244
4245 // these are the only paths that produce 'true':
4246 phi->set_req(_prim_same_path, intcon(1));
4247 phi->set_req(_ref_subtype_path, intcon(1));
4248
4249 // pull together the cases:
4250 assert(region->req() == PATH_LIMIT, "sane region");
4251 for (uint i = 1; i < region->req(); i++) {
4252 Node* ctl = region->in(i);
4253 if (ctl == nullptr || ctl == top()) {
4254 region->set_req(i, top());
4255 phi ->set_req(i, top());
4256 } else if (phi->in(i) == nullptr) {
4257 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4258 }
4259 }
4260
4261 set_control(_gvn.transform(region));
4262 set_result(_gvn.transform(phi));
4263 return true;
4264 }
4265
4266 //---------------------generate_array_guard_common------------------------
4267 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4268 bool obj_array, bool not_array, Node** obj) {
4269
4270 if (stopped()) {
4271 return nullptr;
4272 }
4273
4274 // If obj_array/non_array==false/false:
4275 // Branch around if the given klass is in fact an array (either obj or prim).
4276 // If obj_array/non_array==false/true:
4277 // Branch around if the given klass is not an array klass of any kind.
4278 // If obj_array/non_array==true/true:
4279 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4280 // If obj_array/non_array==true/false:
4281 // Branch around if the kls is an oop array (Object[] or subtype)
4282 //
4283 // Like generate_guard, adds a new path onto the region.
4284 jint layout_con = 0;
4285 Node* layout_val = get_layout_helper(kls, layout_con);
4286 if (layout_val == nullptr) {
4287 bool query = (obj_array
4288 ? Klass::layout_helper_is_objArray(layout_con)
4289 : Klass::layout_helper_is_array(layout_con));
4290 if (query == not_array) {
4291 return nullptr; // never a branch
4292 } else { // always a branch
4293 Node* always_branch = control();
4294 if (region != nullptr)
4295 region->add_req(always_branch);
4296 set_control(top());
4297 return always_branch;
4298 }
4299 }
4300 // Now test the correct condition.
4301 jint nval = (obj_array
4302 ? (jint)(Klass::_lh_array_tag_type_value
4303 << Klass::_lh_array_tag_shift)
4304 : Klass::_lh_neutral_value);
4305 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4306 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4307 // invert the test if we are looking for a non-array
4308 if (not_array) btest = BoolTest(btest).negate();
4309 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4310 Node* ctrl = generate_fair_guard(bol, region);
4311 Node* is_array_ctrl = not_array ? control() : ctrl;
4312 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4313 // Keep track of the fact that 'obj' is an array to prevent
4314 // array specific accesses from floating above the guard.
4315 Node* cast = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4316 // Check for top because in rare cases, the type system can determine that
4317 // the object can't be an array but the layout helper check is not folded.
4318 if (!cast->is_top()) {
4319 *obj = cast;
4320 }
4321 }
4322 return ctrl;
4323 }
4324
4325
4326 //-----------------------inline_native_newArray--------------------------
4327 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4328 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4329 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4330 Node* mirror;
4331 Node* count_val;
4332 if (uninitialized) {
4333 null_check_receiver();
4334 mirror = argument(1);
4335 count_val = argument(2);
4336 } else {
4337 mirror = argument(0);
4338 count_val = argument(1);
4339 }
4340
4341 mirror = null_check(mirror);
4342 // If mirror or obj is dead, only null-path is taken.
4343 if (stopped()) return true;
4344
4345 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4346 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4347 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4348 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4349 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4350
4351 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4352 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4353 result_reg, _slow_path);
4354 Node* normal_ctl = control();
4355 Node* no_array_ctl = result_reg->in(_slow_path);
4356
4357 // Generate code for the slow case. We make a call to newArray().
4358 set_control(no_array_ctl);
4359 if (!stopped()) {
4360 // Either the input type is void.class, or else the
4361 // array klass has not yet been cached. Either the
4362 // ensuing call will throw an exception, or else it
4363 // will cache the array klass for next time.
4364 PreserveJVMState pjvms(this);
4365 CallJavaNode* slow_call = nullptr;
4366 if (uninitialized) {
4367 // Generate optimized virtual call (holder class 'Unsafe' is final)
4368 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4369 } else {
4370 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4371 }
4372 Node* slow_result = set_results_for_java_call(slow_call);
4373 // this->control() comes from set_results_for_java_call
4374 result_reg->set_req(_slow_path, control());
4375 result_val->set_req(_slow_path, slow_result);
4376 result_io ->set_req(_slow_path, i_o());
4377 result_mem->set_req(_slow_path, reset_memory());
4378 }
4379
4380 set_control(normal_ctl);
4381 if (!stopped()) {
4382 // Normal case: The array type has been cached in the java.lang.Class.
4383 // The following call works fine even if the array type is polymorphic.
4384 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4385 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4386 result_reg->init_req(_normal_path, control());
4387 result_val->init_req(_normal_path, obj);
4388 result_io ->init_req(_normal_path, i_o());
4389 result_mem->init_req(_normal_path, reset_memory());
4390
4391 if (uninitialized) {
4392 // Mark the allocation so that zeroing is skipped
4393 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4394 alloc->maybe_set_complete(&_gvn);
4395 }
4396 }
4397
4398 // Return the combined state.
4399 set_i_o( _gvn.transform(result_io) );
4400 set_all_memory( _gvn.transform(result_mem));
4401
4402 C->set_has_split_ifs(true); // Has chance for split-if optimization
4403 set_result(result_reg, result_val);
4404 return true;
4405 }
4406
4407 //----------------------inline_native_getLength--------------------------
4408 // public static native int java.lang.reflect.Array.getLength(Object array);
4409 bool LibraryCallKit::inline_native_getLength() {
4410 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4411
4412 Node* array = null_check(argument(0));
4413 // If array is dead, only null-path is taken.
4414 if (stopped()) return true;
4415
4416 // Deoptimize if it is a non-array.
4417 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4418
4419 if (non_array != nullptr) {
4420 PreserveJVMState pjvms(this);
4421 set_control(non_array);
4422 uncommon_trap(Deoptimization::Reason_intrinsic,
4423 Deoptimization::Action_maybe_recompile);
4424 }
4425
4426 // If control is dead, only non-array-path is taken.
4427 if (stopped()) return true;
4428
4429 // The works fine even if the array type is polymorphic.
4430 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4431 Node* result = load_array_length(array);
4432
4433 C->set_has_split_ifs(true); // Has chance for split-if optimization
4434 set_result(result);
4435 return true;
4436 }
4437
4438 //------------------------inline_array_copyOf----------------------------
4439 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4440 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4441 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4442 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4443
4444 // Get the arguments.
4445 Node* original = argument(0);
4446 Node* start = is_copyOfRange? argument(1): intcon(0);
4447 Node* end = is_copyOfRange? argument(2): argument(1);
4448 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4449
4450 Node* newcopy = nullptr;
4451
4452 // Set the original stack and the reexecute bit for the interpreter to reexecute
4453 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4454 { PreserveReexecuteState preexecs(this);
4455 jvms()->set_should_reexecute(true);
4456
4457 array_type_mirror = null_check(array_type_mirror);
4458 original = null_check(original);
4459
4460 // Check if a null path was taken unconditionally.
4461 if (stopped()) return true;
4462
4463 Node* orig_length = load_array_length(original);
4464
4465 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4466 klass_node = null_check(klass_node);
4467
4468 RegionNode* bailout = new RegionNode(1);
4469 record_for_igvn(bailout);
4470
4471 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4472 // Bail out if that is so.
4473 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4474 if (not_objArray != nullptr) {
4475 // Improve the klass node's type from the new optimistic assumption:
4476 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4477 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4478 Node* cast = new CastPPNode(control(), klass_node, akls);
4479 klass_node = _gvn.transform(cast);
4480 }
4481
4482 // Bail out if either start or end is negative.
4483 generate_negative_guard(start, bailout, &start);
4484 generate_negative_guard(end, bailout, &end);
4485
4486 Node* length = end;
4487 if (_gvn.type(start) != TypeInt::ZERO) {
4488 length = _gvn.transform(new SubINode(end, start));
4489 }
4490
4491 // Bail out if length is negative (i.e., if start > end).
4492 // Without this the new_array would throw
4493 // NegativeArraySizeException but IllegalArgumentException is what
4494 // should be thrown
4495 generate_negative_guard(length, bailout, &length);
4496
4497 // Bail out if start is larger than the original length
4498 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4499 generate_negative_guard(orig_tail, bailout, &orig_tail);
4500
4501 if (bailout->req() > 1) {
4502 PreserveJVMState pjvms(this);
4503 set_control(_gvn.transform(bailout));
4504 uncommon_trap(Deoptimization::Reason_intrinsic,
4505 Deoptimization::Action_maybe_recompile);
4506 }
4507
4508 if (!stopped()) {
4509 // How many elements will we copy from the original?
4510 // The answer is MinI(orig_tail, length).
4511 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4512
4513 // Generate a direct call to the right arraycopy function(s).
4514 // We know the copy is disjoint but we might not know if the
4515 // oop stores need checking.
4516 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4517 // This will fail a store-check if x contains any non-nulls.
4518
4519 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4520 // loads/stores but it is legal only if we're sure the
4521 // Arrays.copyOf would succeed. So we need all input arguments
4522 // to the copyOf to be validated, including that the copy to the
4523 // new array won't trigger an ArrayStoreException. That subtype
4524 // check can be optimized if we know something on the type of
4525 // the input array from type speculation.
4526 if (_gvn.type(klass_node)->singleton()) {
4527 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4528 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4529
4530 int test = C->static_subtype_check(superk, subk);
4531 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4532 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4533 if (t_original->speculative_type() != nullptr) {
4534 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4535 }
4536 }
4537 }
4538
4539 bool validated = false;
4540 // Reason_class_check rather than Reason_intrinsic because we
4541 // want to intrinsify even if this traps.
4542 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4543 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4544
4545 if (not_subtype_ctrl != top()) {
4546 PreserveJVMState pjvms(this);
4547 set_control(not_subtype_ctrl);
4548 uncommon_trap(Deoptimization::Reason_class_check,
4549 Deoptimization::Action_make_not_entrant);
4550 assert(stopped(), "Should be stopped");
4551 }
4552 validated = true;
4553 }
4554
4555 if (!stopped()) {
4556 newcopy = new_array(klass_node, length, 0); // no arguments to push
4557
4558 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4559 load_object_klass(original), klass_node);
4560 if (!is_copyOfRange) {
4561 ac->set_copyof(validated);
4562 } else {
4563 ac->set_copyofrange(validated);
4564 }
4565 Node* n = _gvn.transform(ac);
4566 if (n == ac) {
4567 ac->connect_outputs(this);
4568 } else {
4569 assert(validated, "shouldn't transform if all arguments not validated");
4570 set_all_memory(n);
4571 }
4572 }
4573 }
4574 } // original reexecute is set back here
4575
4576 C->set_has_split_ifs(true); // Has chance for split-if optimization
4577 if (!stopped()) {
4578 set_result(newcopy);
4579 }
4580 return true;
4581 }
4582
4583
4584 //----------------------generate_virtual_guard---------------------------
4585 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4586 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4587 RegionNode* slow_region) {
4588 ciMethod* method = callee();
4589 int vtable_index = method->vtable_index();
4590 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4591 "bad index %d", vtable_index);
4592 // Get the Method* out of the appropriate vtable entry.
4593 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4594 vtable_index*vtableEntry::size_in_bytes() +
4595 in_bytes(vtableEntry::method_offset());
4596 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4597 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4598
4599 // Compare the target method with the expected method (e.g., Object.hashCode).
4600 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4601
4602 Node* native_call = makecon(native_call_addr);
4603 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4604 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4605
4606 return generate_slow_guard(test_native, slow_region);
4607 }
4608
4609 //-----------------------generate_method_call----------------------------
4610 // Use generate_method_call to make a slow-call to the real
4611 // method if the fast path fails. An alternative would be to
4612 // use a stub like OptoRuntime::slow_arraycopy_Java.
4613 // This only works for expanding the current library call,
4614 // not another intrinsic. (E.g., don't use this for making an
4615 // arraycopy call inside of the copyOf intrinsic.)
4616 CallJavaNode*
4617 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4618 // When compiling the intrinsic method itself, do not use this technique.
4619 guarantee(callee() != C->method(), "cannot make slow-call to self");
4620
4621 ciMethod* method = callee();
4622 // ensure the JVMS we have will be correct for this call
4623 guarantee(method_id == method->intrinsic_id(), "must match");
4624
4625 const TypeFunc* tf = TypeFunc::make(method);
4626 if (res_not_null) {
4627 assert(tf->return_type() == T_OBJECT, "");
4628 const TypeTuple* range = tf->range();
4629 const Type** fields = TypeTuple::fields(range->cnt());
4630 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4631 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4632 tf = TypeFunc::make(tf->domain(), new_range);
4633 }
4634 CallJavaNode* slow_call;
4635 if (is_static) {
4636 assert(!is_virtual, "");
4637 slow_call = new CallStaticJavaNode(C, tf,
4638 SharedRuntime::get_resolve_static_call_stub(), method);
4639 } else if (is_virtual) {
4640 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4641 int vtable_index = Method::invalid_vtable_index;
4642 if (UseInlineCaches) {
4643 // Suppress the vtable call
4644 } else {
4645 // hashCode and clone are not a miranda methods,
4646 // so the vtable index is fixed.
4647 // No need to use the linkResolver to get it.
4648 vtable_index = method->vtable_index();
4649 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4650 "bad index %d", vtable_index);
4651 }
4652 slow_call = new CallDynamicJavaNode(tf,
4653 SharedRuntime::get_resolve_virtual_call_stub(),
4654 method, vtable_index);
4655 } else { // neither virtual nor static: opt_virtual
4656 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4657 slow_call = new CallStaticJavaNode(C, tf,
4658 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4659 slow_call->set_optimized_virtual(true);
4660 }
4661 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4662 // To be able to issue a direct call (optimized virtual or virtual)
4663 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4664 // about the method being invoked should be attached to the call site to
4665 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4666 slow_call->set_override_symbolic_info(true);
4667 }
4668 set_arguments_for_java_call(slow_call);
4669 set_edges_for_java_call(slow_call);
4670 return slow_call;
4671 }
4672
4673
4674 /**
4675 * Build special case code for calls to hashCode on an object. This call may
4676 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4677 * slightly different code.
4678 */
4679 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4680 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4681 assert(!(is_virtual && is_static), "either virtual, special, or static");
4682
4683 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4684
4685 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4686 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4687 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4688 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4689 Node* obj = nullptr;
4690 if (!is_static) {
4691 // Check for hashing null object
4692 obj = null_check_receiver();
4693 if (stopped()) return true; // unconditionally null
4694 result_reg->init_req(_null_path, top());
4695 result_val->init_req(_null_path, top());
4696 } else {
4697 // Do a null check, and return zero if null.
4698 // System.identityHashCode(null) == 0
4699 obj = argument(0);
4700 Node* null_ctl = top();
4701 obj = null_check_oop(obj, &null_ctl);
4702 result_reg->init_req(_null_path, null_ctl);
4703 result_val->init_req(_null_path, _gvn.intcon(0));
4704 }
4705
4706 // Unconditionally null? Then return right away.
4707 if (stopped()) {
4708 set_control( result_reg->in(_null_path));
4709 if (!stopped())
4710 set_result(result_val->in(_null_path));
4711 return true;
4712 }
4713
4714 // We only go to the fast case code if we pass a number of guards. The
4715 // paths which do not pass are accumulated in the slow_region.
4716 RegionNode* slow_region = new RegionNode(1);
4717 record_for_igvn(slow_region);
4718
4719 // If this is a virtual call, we generate a funny guard. We pull out
4720 // the vtable entry corresponding to hashCode() from the target object.
4721 // If the target method which we are calling happens to be the native
4722 // Object hashCode() method, we pass the guard. We do not need this
4723 // guard for non-virtual calls -- the caller is known to be the native
4724 // Object hashCode().
4725 if (is_virtual) {
4726 // After null check, get the object's klass.
4727 Node* obj_klass = load_object_klass(obj);
4728 generate_virtual_guard(obj_klass, slow_region);
4729 }
4730
4731 // Get the header out of the object, use LoadMarkNode when available
4732 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4733 // The control of the load must be null. Otherwise, the load can move before
4734 // the null check after castPP removal.
4735 Node* no_ctrl = nullptr;
4736 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4737
4738 if (!UseObjectMonitorTable) {
4739 // Test the header to see if it is safe to read w.r.t. locking.
4740 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4741 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4742 if (LockingMode == LM_LIGHTWEIGHT) {
4743 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4744 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4745 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4746
4747 generate_slow_guard(test_monitor, slow_region);
4748 } else {
4749 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4750 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4751 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4752
4753 generate_slow_guard(test_not_unlocked, slow_region);
4754 }
4755 }
4756
4757 // Get the hash value and check to see that it has been properly assigned.
4758 // We depend on hash_mask being at most 32 bits and avoid the use of
4759 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4760 // vm: see markWord.hpp.
4761 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4762 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4763 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4764 // This hack lets the hash bits live anywhere in the mark object now, as long
4765 // as the shift drops the relevant bits into the low 32 bits. Note that
4766 // Java spec says that HashCode is an int so there's no point in capturing
4767 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4768 hshifted_header = ConvX2I(hshifted_header);
4769 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4770
4771 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4772 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4773 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4774
4775 generate_slow_guard(test_assigned, slow_region);
4776
4777 Node* init_mem = reset_memory();
4778 // fill in the rest of the null path:
4779 result_io ->init_req(_null_path, i_o());
4780 result_mem->init_req(_null_path, init_mem);
4781
4782 result_val->init_req(_fast_path, hash_val);
4783 result_reg->init_req(_fast_path, control());
4784 result_io ->init_req(_fast_path, i_o());
4785 result_mem->init_req(_fast_path, init_mem);
4786
4787 // Generate code for the slow case. We make a call to hashCode().
4788 set_control(_gvn.transform(slow_region));
4789 if (!stopped()) {
4790 // No need for PreserveJVMState, because we're using up the present state.
4791 set_all_memory(init_mem);
4792 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4793 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4794 Node* slow_result = set_results_for_java_call(slow_call);
4795 // this->control() comes from set_results_for_java_call
4796 result_reg->init_req(_slow_path, control());
4797 result_val->init_req(_slow_path, slow_result);
4798 result_io ->set_req(_slow_path, i_o());
4799 result_mem ->set_req(_slow_path, reset_memory());
4800 }
4801
4802 // Return the combined state.
4803 set_i_o( _gvn.transform(result_io) );
4804 set_all_memory( _gvn.transform(result_mem));
4805
4806 set_result(result_reg, result_val);
4807 return true;
4808 }
4809
4810 //---------------------------inline_native_getClass----------------------------
4811 // public final native Class<?> java.lang.Object.getClass();
4812 //
4813 // Build special case code for calls to getClass on an object.
4814 bool LibraryCallKit::inline_native_getClass() {
4815 Node* obj = null_check_receiver();
4816 if (stopped()) return true;
4817 set_result(load_mirror_from_klass(load_object_klass(obj)));
4818 return true;
4819 }
4820
4821 //-----------------inline_native_Reflection_getCallerClass---------------------
4822 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4823 //
4824 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4825 //
4826 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4827 // in that it must skip particular security frames and checks for
4828 // caller sensitive methods.
4829 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4830 #ifndef PRODUCT
4831 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4832 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4833 }
4834 #endif
4835
4836 if (!jvms()->has_method()) {
4837 #ifndef PRODUCT
4838 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4839 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4840 }
4841 #endif
4842 return false;
4843 }
4844
4845 // Walk back up the JVM state to find the caller at the required
4846 // depth.
4847 JVMState* caller_jvms = jvms();
4848
4849 // Cf. JVM_GetCallerClass
4850 // NOTE: Start the loop at depth 1 because the current JVM state does
4851 // not include the Reflection.getCallerClass() frame.
4852 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4853 ciMethod* m = caller_jvms->method();
4854 switch (n) {
4855 case 0:
4856 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4857 break;
4858 case 1:
4859 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4860 if (!m->caller_sensitive()) {
4861 #ifndef PRODUCT
4862 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4863 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4864 }
4865 #endif
4866 return false; // bail-out; let JVM_GetCallerClass do the work
4867 }
4868 break;
4869 default:
4870 if (!m->is_ignored_by_security_stack_walk()) {
4871 // We have reached the desired frame; return the holder class.
4872 // Acquire method holder as java.lang.Class and push as constant.
4873 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4874 ciInstance* caller_mirror = caller_klass->java_mirror();
4875 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4876
4877 #ifndef PRODUCT
4878 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4879 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());
4880 tty->print_cr(" JVM state at this point:");
4881 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4882 ciMethod* m = jvms()->of_depth(i)->method();
4883 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4884 }
4885 }
4886 #endif
4887 return true;
4888 }
4889 break;
4890 }
4891 }
4892
4893 #ifndef PRODUCT
4894 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4895 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4896 tty->print_cr(" JVM state at this point:");
4897 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4898 ciMethod* m = jvms()->of_depth(i)->method();
4899 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4900 }
4901 }
4902 #endif
4903
4904 return false; // bail-out; let JVM_GetCallerClass do the work
4905 }
4906
4907 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4908 Node* arg = argument(0);
4909 Node* result = nullptr;
4910
4911 switch (id) {
4912 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4913 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4914 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4915 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4916 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4917 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4918
4919 case vmIntrinsics::_doubleToLongBits: {
4920 // two paths (plus control) merge in a wood
4921 RegionNode *r = new RegionNode(3);
4922 Node *phi = new PhiNode(r, TypeLong::LONG);
4923
4924 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4925 // Build the boolean node
4926 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4927
4928 // Branch either way.
4929 // NaN case is less traveled, which makes all the difference.
4930 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4931 Node *opt_isnan = _gvn.transform(ifisnan);
4932 assert( opt_isnan->is_If(), "Expect an IfNode");
4933 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4934 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4935
4936 set_control(iftrue);
4937
4938 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4939 Node *slow_result = longcon(nan_bits); // return NaN
4940 phi->init_req(1, _gvn.transform( slow_result ));
4941 r->init_req(1, iftrue);
4942
4943 // Else fall through
4944 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4945 set_control(iffalse);
4946
4947 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4948 r->init_req(2, iffalse);
4949
4950 // Post merge
4951 set_control(_gvn.transform(r));
4952 record_for_igvn(r);
4953
4954 C->set_has_split_ifs(true); // Has chance for split-if optimization
4955 result = phi;
4956 assert(result->bottom_type()->isa_long(), "must be");
4957 break;
4958 }
4959
4960 case vmIntrinsics::_floatToIntBits: {
4961 // two paths (plus control) merge in a wood
4962 RegionNode *r = new RegionNode(3);
4963 Node *phi = new PhiNode(r, TypeInt::INT);
4964
4965 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4966 // Build the boolean node
4967 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4968
4969 // Branch either way.
4970 // NaN case is less traveled, which makes all the difference.
4971 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4972 Node *opt_isnan = _gvn.transform(ifisnan);
4973 assert( opt_isnan->is_If(), "Expect an IfNode");
4974 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4975 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4976
4977 set_control(iftrue);
4978
4979 static const jint nan_bits = 0x7fc00000;
4980 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4981 phi->init_req(1, _gvn.transform( slow_result ));
4982 r->init_req(1, iftrue);
4983
4984 // Else fall through
4985 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4986 set_control(iffalse);
4987
4988 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4989 r->init_req(2, iffalse);
4990
4991 // Post merge
4992 set_control(_gvn.transform(r));
4993 record_for_igvn(r);
4994
4995 C->set_has_split_ifs(true); // Has chance for split-if optimization
4996 result = phi;
4997 assert(result->bottom_type()->isa_int(), "must be");
4998 break;
4999 }
5000
5001 default:
5002 fatal_unexpected_iid(id);
5003 break;
5004 }
5005 set_result(_gvn.transform(result));
5006 return true;
5007 }
5008
5009 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5010 Node* arg = argument(0);
5011 Node* result = nullptr;
5012
5013 switch (id) {
5014 case vmIntrinsics::_floatIsInfinite:
5015 result = new IsInfiniteFNode(arg);
5016 break;
5017 case vmIntrinsics::_floatIsFinite:
5018 result = new IsFiniteFNode(arg);
5019 break;
5020 case vmIntrinsics::_doubleIsInfinite:
5021 result = new IsInfiniteDNode(arg);
5022 break;
5023 case vmIntrinsics::_doubleIsFinite:
5024 result = new IsFiniteDNode(arg);
5025 break;
5026 default:
5027 fatal_unexpected_iid(id);
5028 break;
5029 }
5030 set_result(_gvn.transform(result));
5031 return true;
5032 }
5033
5034 //----------------------inline_unsafe_copyMemory-------------------------
5035 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5036
5037 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5038 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5039 const Type* base_t = gvn.type(base);
5040
5041 bool in_native = (base_t == TypePtr::NULL_PTR);
5042 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5043 bool is_mixed = !in_heap && !in_native;
5044
5045 if (is_mixed) {
5046 return true; // mixed accesses can touch both on-heap and off-heap memory
5047 }
5048 if (in_heap) {
5049 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5050 if (!is_prim_array) {
5051 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5052 // there's not enough type information available to determine proper memory slice for it.
5053 return true;
5054 }
5055 }
5056 return false;
5057 }
5058
5059 bool LibraryCallKit::inline_unsafe_copyMemory() {
5060 if (callee()->is_static()) return false; // caller must have the capability!
5061 null_check_receiver(); // null-check receiver
5062 if (stopped()) return true;
5063
5064 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5065
5066 Node* src_base = argument(1); // type: oop
5067 Node* src_off = ConvL2X(argument(2)); // type: long
5068 Node* dst_base = argument(4); // type: oop
5069 Node* dst_off = ConvL2X(argument(5)); // type: long
5070 Node* size = ConvL2X(argument(7)); // type: long
5071
5072 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5073 "fieldOffset must be byte-scaled");
5074
5075 Node* src_addr = make_unsafe_address(src_base, src_off);
5076 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5077
5078 Node* thread = _gvn.transform(new ThreadLocalNode());
5079 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5080 BasicType doing_unsafe_access_bt = T_BYTE;
5081 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5082
5083 // update volatile field
5084 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5085
5086 int flags = RC_LEAF | RC_NO_FP;
5087
5088 const TypePtr* dst_type = TypePtr::BOTTOM;
5089
5090 // Adjust memory effects of the runtime call based on input values.
5091 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5092 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5093 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5094
5095 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5096 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5097 flags |= RC_NARROW_MEM; // narrow in memory
5098 }
5099 }
5100
5101 // Call it. Note that the length argument is not scaled.
5102 make_runtime_call(flags,
5103 OptoRuntime::fast_arraycopy_Type(),
5104 StubRoutines::unsafe_arraycopy(),
5105 "unsafe_arraycopy",
5106 dst_type,
5107 src_addr, dst_addr, size XTOP);
5108
5109 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5110
5111 return true;
5112 }
5113
5114 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5115 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5116 bool LibraryCallKit::inline_unsafe_setMemory() {
5117 if (callee()->is_static()) return false; // caller must have the capability!
5118 null_check_receiver(); // null-check receiver
5119 if (stopped()) return true;
5120
5121 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5122
5123 Node* dst_base = argument(1); // type: oop
5124 Node* dst_off = ConvL2X(argument(2)); // type: long
5125 Node* size = ConvL2X(argument(4)); // type: long
5126 Node* byte = argument(6); // type: byte
5127
5128 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5129 "fieldOffset must be byte-scaled");
5130
5131 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5132
5133 Node* thread = _gvn.transform(new ThreadLocalNode());
5134 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5135 BasicType doing_unsafe_access_bt = T_BYTE;
5136 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5137
5138 // update volatile field
5139 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5140
5141 int flags = RC_LEAF | RC_NO_FP;
5142
5143 const TypePtr* dst_type = TypePtr::BOTTOM;
5144
5145 // Adjust memory effects of the runtime call based on input values.
5146 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5147 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5148
5149 flags |= RC_NARROW_MEM; // narrow in memory
5150 }
5151
5152 // Call it. Note that the length argument is not scaled.
5153 make_runtime_call(flags,
5154 OptoRuntime::unsafe_setmemory_Type(),
5155 StubRoutines::unsafe_setmemory(),
5156 "unsafe_setmemory",
5157 dst_type,
5158 dst_addr, size XTOP, byte);
5159
5160 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5161
5162 return true;
5163 }
5164
5165 #undef XTOP
5166
5167 //------------------------clone_coping-----------------------------------
5168 // Helper function for inline_native_clone.
5169 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5170 assert(obj_size != nullptr, "");
5171 Node* raw_obj = alloc_obj->in(1);
5172 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5173
5174 AllocateNode* alloc = nullptr;
5175 if (ReduceBulkZeroing &&
5176 // If we are implementing an array clone without knowing its source type
5177 // (can happen when compiling the array-guarded branch of a reflective
5178 // Object.clone() invocation), initialize the array within the allocation.
5179 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5180 // to a runtime clone call that assumes fully initialized source arrays.
5181 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5182 // We will be completely responsible for initializing this object -
5183 // mark Initialize node as complete.
5184 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5185 // The object was just allocated - there should be no any stores!
5186 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5187 // Mark as complete_with_arraycopy so that on AllocateNode
5188 // expansion, we know this AllocateNode is initialized by an array
5189 // copy and a StoreStore barrier exists after the array copy.
5190 alloc->initialization()->set_complete_with_arraycopy();
5191 }
5192
5193 Node* size = _gvn.transform(obj_size);
5194 access_clone(obj, alloc_obj, size, is_array);
5195
5196 // Do not let reads from the cloned object float above the arraycopy.
5197 if (alloc != nullptr) {
5198 // Do not let stores that initialize this object be reordered with
5199 // a subsequent store that would make this object accessible by
5200 // other threads.
5201 // Record what AllocateNode this StoreStore protects so that
5202 // escape analysis can go from the MemBarStoreStoreNode to the
5203 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5204 // based on the escape status of the AllocateNode.
5205 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5206 } else {
5207 insert_mem_bar(Op_MemBarCPUOrder);
5208 }
5209 }
5210
5211 //------------------------inline_native_clone----------------------------
5212 // protected native Object java.lang.Object.clone();
5213 //
5214 // Here are the simple edge cases:
5215 // null receiver => normal trap
5216 // virtual and clone was overridden => slow path to out-of-line clone
5217 // not cloneable or finalizer => slow path to out-of-line Object.clone
5218 //
5219 // The general case has two steps, allocation and copying.
5220 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5221 //
5222 // Copying also has two cases, oop arrays and everything else.
5223 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5224 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5225 //
5226 // These steps fold up nicely if and when the cloned object's klass
5227 // can be sharply typed as an object array, a type array, or an instance.
5228 //
5229 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5230 PhiNode* result_val;
5231
5232 // Set the reexecute bit for the interpreter to reexecute
5233 // the bytecode that invokes Object.clone if deoptimization happens.
5234 { PreserveReexecuteState preexecs(this);
5235 jvms()->set_should_reexecute(true);
5236
5237 Node* obj = null_check_receiver();
5238 if (stopped()) return true;
5239
5240 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5241
5242 // If we are going to clone an instance, we need its exact type to
5243 // know the number and types of fields to convert the clone to
5244 // loads/stores. Maybe a speculative type can help us.
5245 if (!obj_type->klass_is_exact() &&
5246 obj_type->speculative_type() != nullptr &&
5247 obj_type->speculative_type()->is_instance_klass()) {
5248 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5249 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5250 !spec_ik->has_injected_fields()) {
5251 if (!obj_type->isa_instptr() ||
5252 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5253 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5254 }
5255 }
5256 }
5257
5258 // Conservatively insert a memory barrier on all memory slices.
5259 // Do not let writes into the original float below the clone.
5260 insert_mem_bar(Op_MemBarCPUOrder);
5261
5262 // paths into result_reg:
5263 enum {
5264 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5265 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5266 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5267 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5268 PATH_LIMIT
5269 };
5270 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5271 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5272 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5273 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5274 record_for_igvn(result_reg);
5275
5276 Node* obj_klass = load_object_klass(obj);
5277 Node* array_obj = obj;
5278 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5279 if (array_ctl != nullptr) {
5280 // It's an array.
5281 PreserveJVMState pjvms(this);
5282 set_control(array_ctl);
5283 Node* obj_length = load_array_length(array_obj);
5284 Node* array_size = nullptr; // Size of the array without object alignment padding.
5285 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5286
5287 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5288 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5289 // If it is an oop array, it requires very special treatment,
5290 // because gc barriers are required when accessing the array.
5291 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5292 if (is_obja != nullptr) {
5293 PreserveJVMState pjvms2(this);
5294 set_control(is_obja);
5295 // Generate a direct call to the right arraycopy function(s).
5296 // Clones are always tightly coupled.
5297 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5298 ac->set_clone_oop_array();
5299 Node* n = _gvn.transform(ac);
5300 assert(n == ac, "cannot disappear");
5301 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5302
5303 result_reg->init_req(_objArray_path, control());
5304 result_val->init_req(_objArray_path, alloc_obj);
5305 result_i_o ->set_req(_objArray_path, i_o());
5306 result_mem ->set_req(_objArray_path, reset_memory());
5307 }
5308 }
5309 // Otherwise, there are no barriers to worry about.
5310 // (We can dispense with card marks if we know the allocation
5311 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5312 // causes the non-eden paths to take compensating steps to
5313 // simulate a fresh allocation, so that no further
5314 // card marks are required in compiled code to initialize
5315 // the object.)
5316
5317 if (!stopped()) {
5318 copy_to_clone(array_obj, alloc_obj, array_size, true);
5319
5320 // Present the results of the copy.
5321 result_reg->init_req(_array_path, control());
5322 result_val->init_req(_array_path, alloc_obj);
5323 result_i_o ->set_req(_array_path, i_o());
5324 result_mem ->set_req(_array_path, reset_memory());
5325 }
5326 }
5327
5328 // We only go to the instance fast case code if we pass a number of guards.
5329 // The paths which do not pass are accumulated in the slow_region.
5330 RegionNode* slow_region = new RegionNode(1);
5331 record_for_igvn(slow_region);
5332 if (!stopped()) {
5333 // It's an instance (we did array above). Make the slow-path tests.
5334 // If this is a virtual call, we generate a funny guard. We grab
5335 // the vtable entry corresponding to clone() from the target object.
5336 // If the target method which we are calling happens to be the
5337 // Object clone() method, we pass the guard. We do not need this
5338 // guard for non-virtual calls; the caller is known to be the native
5339 // Object clone().
5340 if (is_virtual) {
5341 generate_virtual_guard(obj_klass, slow_region);
5342 }
5343
5344 // The object must be easily cloneable and must not have a finalizer.
5345 // Both of these conditions may be checked in a single test.
5346 // We could optimize the test further, but we don't care.
5347 generate_misc_flags_guard(obj_klass,
5348 // Test both conditions:
5349 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5350 // Must be cloneable but not finalizer:
5351 KlassFlags::_misc_is_cloneable_fast,
5352 slow_region);
5353 }
5354
5355 if (!stopped()) {
5356 // It's an instance, and it passed the slow-path tests.
5357 PreserveJVMState pjvms(this);
5358 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5359 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5360 // is reexecuted if deoptimization occurs and there could be problems when merging
5361 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5362 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5363
5364 copy_to_clone(obj, alloc_obj, obj_size, false);
5365
5366 // Present the results of the slow call.
5367 result_reg->init_req(_instance_path, control());
5368 result_val->init_req(_instance_path, alloc_obj);
5369 result_i_o ->set_req(_instance_path, i_o());
5370 result_mem ->set_req(_instance_path, reset_memory());
5371 }
5372
5373 // Generate code for the slow case. We make a call to clone().
5374 set_control(_gvn.transform(slow_region));
5375 if (!stopped()) {
5376 PreserveJVMState pjvms(this);
5377 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5378 // We need to deoptimize on exception (see comment above)
5379 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5380 // this->control() comes from set_results_for_java_call
5381 result_reg->init_req(_slow_path, control());
5382 result_val->init_req(_slow_path, slow_result);
5383 result_i_o ->set_req(_slow_path, i_o());
5384 result_mem ->set_req(_slow_path, reset_memory());
5385 }
5386
5387 // Return the combined state.
5388 set_control( _gvn.transform(result_reg));
5389 set_i_o( _gvn.transform(result_i_o));
5390 set_all_memory( _gvn.transform(result_mem));
5391 } // original reexecute is set back here
5392
5393 set_result(_gvn.transform(result_val));
5394 return true;
5395 }
5396
5397 // If we have a tightly coupled allocation, the arraycopy may take care
5398 // of the array initialization. If one of the guards we insert between
5399 // the allocation and the arraycopy causes a deoptimization, an
5400 // uninitialized array will escape the compiled method. To prevent that
5401 // we set the JVM state for uncommon traps between the allocation and
5402 // the arraycopy to the state before the allocation so, in case of
5403 // deoptimization, we'll reexecute the allocation and the
5404 // initialization.
5405 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5406 if (alloc != nullptr) {
5407 ciMethod* trap_method = alloc->jvms()->method();
5408 int trap_bci = alloc->jvms()->bci();
5409
5410 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5411 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5412 // Make sure there's no store between the allocation and the
5413 // arraycopy otherwise visible side effects could be rexecuted
5414 // in case of deoptimization and cause incorrect execution.
5415 bool no_interfering_store = true;
5416 Node* mem = alloc->in(TypeFunc::Memory);
5417 if (mem->is_MergeMem()) {
5418 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5419 Node* n = mms.memory();
5420 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5421 assert(n->is_Store(), "what else?");
5422 no_interfering_store = false;
5423 break;
5424 }
5425 }
5426 } else {
5427 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5428 Node* n = mms.memory();
5429 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5430 assert(n->is_Store(), "what else?");
5431 no_interfering_store = false;
5432 break;
5433 }
5434 }
5435 }
5436
5437 if (no_interfering_store) {
5438 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5439
5440 JVMState* saved_jvms = jvms();
5441 saved_reexecute_sp = _reexecute_sp;
5442
5443 set_jvms(sfpt->jvms());
5444 _reexecute_sp = jvms()->sp();
5445
5446 return saved_jvms;
5447 }
5448 }
5449 }
5450 return nullptr;
5451 }
5452
5453 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5454 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5455 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5456 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5457 uint size = alloc->req();
5458 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5459 old_jvms->set_map(sfpt);
5460 for (uint i = 0; i < size; i++) {
5461 sfpt->init_req(i, alloc->in(i));
5462 }
5463 // re-push array length for deoptimization
5464 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5465 old_jvms->set_sp(old_jvms->sp()+1);
5466 old_jvms->set_monoff(old_jvms->monoff()+1);
5467 old_jvms->set_scloff(old_jvms->scloff()+1);
5468 old_jvms->set_endoff(old_jvms->endoff()+1);
5469 old_jvms->set_should_reexecute(true);
5470
5471 sfpt->set_i_o(map()->i_o());
5472 sfpt->set_memory(map()->memory());
5473 sfpt->set_control(map()->control());
5474 return sfpt;
5475 }
5476
5477 // In case of a deoptimization, we restart execution at the
5478 // allocation, allocating a new array. We would leave an uninitialized
5479 // array in the heap that GCs wouldn't expect. Move the allocation
5480 // after the traps so we don't allocate the array if we
5481 // deoptimize. This is possible because tightly_coupled_allocation()
5482 // guarantees there's no observer of the allocated array at this point
5483 // and the control flow is simple enough.
5484 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5485 int saved_reexecute_sp, uint new_idx) {
5486 if (saved_jvms_before_guards != nullptr && !stopped()) {
5487 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5488
5489 assert(alloc != nullptr, "only with a tightly coupled allocation");
5490 // restore JVM state to the state at the arraycopy
5491 saved_jvms_before_guards->map()->set_control(map()->control());
5492 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5493 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5494 // If we've improved the types of some nodes (null check) while
5495 // emitting the guards, propagate them to the current state
5496 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5497 set_jvms(saved_jvms_before_guards);
5498 _reexecute_sp = saved_reexecute_sp;
5499
5500 // Remove the allocation from above the guards
5501 CallProjections callprojs;
5502 alloc->extract_projections(&callprojs, true);
5503 InitializeNode* init = alloc->initialization();
5504 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5505 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5506 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5507
5508 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5509 // the allocation (i.e. is only valid if the allocation succeeds):
5510 // 1) replace CastIINode with AllocateArrayNode's length here
5511 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5512 //
5513 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5514 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5515 Node* init_control = init->proj_out(TypeFunc::Control);
5516 Node* alloc_length = alloc->Ideal_length();
5517 #ifdef ASSERT
5518 Node* prev_cast = nullptr;
5519 #endif
5520 for (uint i = 0; i < init_control->outcnt(); i++) {
5521 Node* init_out = init_control->raw_out(i);
5522 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5523 #ifdef ASSERT
5524 if (prev_cast == nullptr) {
5525 prev_cast = init_out;
5526 } else {
5527 if (prev_cast->cmp(*init_out) == false) {
5528 prev_cast->dump();
5529 init_out->dump();
5530 assert(false, "not equal CastIINode");
5531 }
5532 }
5533 #endif
5534 C->gvn_replace_by(init_out, alloc_length);
5535 }
5536 }
5537 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5538
5539 // move the allocation here (after the guards)
5540 _gvn.hash_delete(alloc);
5541 alloc->set_req(TypeFunc::Control, control());
5542 alloc->set_req(TypeFunc::I_O, i_o());
5543 Node *mem = reset_memory();
5544 set_all_memory(mem);
5545 alloc->set_req(TypeFunc::Memory, mem);
5546 set_control(init->proj_out_or_null(TypeFunc::Control));
5547 set_i_o(callprojs.fallthrough_ioproj);
5548
5549 // Update memory as done in GraphKit::set_output_for_allocation()
5550 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5551 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5552 if (ary_type->isa_aryptr() && length_type != nullptr) {
5553 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5554 }
5555 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5556 int elemidx = C->get_alias_index(telemref);
5557 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5558 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5559
5560 Node* allocx = _gvn.transform(alloc);
5561 assert(allocx == alloc, "where has the allocation gone?");
5562 assert(dest->is_CheckCastPP(), "not an allocation result?");
5563
5564 _gvn.hash_delete(dest);
5565 dest->set_req(0, control());
5566 Node* destx = _gvn.transform(dest);
5567 assert(destx == dest, "where has the allocation result gone?");
5568
5569 array_ideal_length(alloc, ary_type, true);
5570 }
5571 }
5572
5573 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5574 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5575 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5576 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5577 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5578 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5579 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5580 JVMState* saved_jvms_before_guards) {
5581 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5582 // There is at least one unrelated uncommon trap which needs to be replaced.
5583 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5584
5585 JVMState* saved_jvms = jvms();
5586 const int saved_reexecute_sp = _reexecute_sp;
5587 set_jvms(sfpt->jvms());
5588 _reexecute_sp = jvms()->sp();
5589
5590 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5591
5592 // Restore state
5593 set_jvms(saved_jvms);
5594 _reexecute_sp = saved_reexecute_sp;
5595 }
5596 }
5597
5598 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5599 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5600 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5601 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5602 while (if_proj->is_IfProj()) {
5603 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5604 if (uncommon_trap != nullptr) {
5605 create_new_uncommon_trap(uncommon_trap);
5606 }
5607 assert(if_proj->in(0)->is_If(), "must be If");
5608 if_proj = if_proj->in(0)->in(0);
5609 }
5610 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5611 "must have reached control projection of init node");
5612 }
5613
5614 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5615 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5616 assert(trap_request != 0, "no valid UCT trap request");
5617 PreserveJVMState pjvms(this);
5618 set_control(uncommon_trap_call->in(0));
5619 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5620 Deoptimization::trap_request_action(trap_request));
5621 assert(stopped(), "Should be stopped");
5622 _gvn.hash_delete(uncommon_trap_call);
5623 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5624 }
5625
5626 // Common checks for array sorting intrinsics arguments.
5627 // Returns `true` if checks passed.
5628 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5629 // check address of the class
5630 if (elementType == nullptr || elementType->is_top()) {
5631 return false; // dead path
5632 }
5633 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5634 if (elem_klass == nullptr) {
5635 return false; // dead path
5636 }
5637 // java_mirror_type() returns non-null for compile-time Class constants only
5638 ciType* elem_type = elem_klass->java_mirror_type();
5639 if (elem_type == nullptr) {
5640 return false;
5641 }
5642 bt = elem_type->basic_type();
5643 // Disable the intrinsic if the CPU does not support SIMD sort
5644 if (!Matcher::supports_simd_sort(bt)) {
5645 return false;
5646 }
5647 // check address of the array
5648 if (obj == nullptr || obj->is_top()) {
5649 return false; // dead path
5650 }
5651 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5652 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5653 return false; // failed input validation
5654 }
5655 return true;
5656 }
5657
5658 //------------------------------inline_array_partition-----------------------
5659 bool LibraryCallKit::inline_array_partition() {
5660 address stubAddr = StubRoutines::select_array_partition_function();
5661 if (stubAddr == nullptr) {
5662 return false; // Intrinsic's stub is not implemented on this platform
5663 }
5664 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5665
5666 // no receiver because it is a static method
5667 Node* elementType = argument(0);
5668 Node* obj = argument(1);
5669 Node* offset = argument(2); // long
5670 Node* fromIndex = argument(4);
5671 Node* toIndex = argument(5);
5672 Node* indexPivot1 = argument(6);
5673 Node* indexPivot2 = argument(7);
5674 // PartitionOperation: argument(8) is ignored
5675
5676 Node* pivotIndices = nullptr;
5677 BasicType bt = T_ILLEGAL;
5678
5679 if (!check_array_sort_arguments(elementType, obj, bt)) {
5680 return false;
5681 }
5682 null_check(obj);
5683 // If obj is dead, only null-path is taken.
5684 if (stopped()) {
5685 return true;
5686 }
5687 // Set the original stack and the reexecute bit for the interpreter to reexecute
5688 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5689 { PreserveReexecuteState preexecs(this);
5690 jvms()->set_should_reexecute(true);
5691
5692 Node* obj_adr = make_unsafe_address(obj, offset);
5693
5694 // create the pivotIndices array of type int and size = 2
5695 Node* size = intcon(2);
5696 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5697 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5698 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5699 guarantee(alloc != nullptr, "created above");
5700 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5701
5702 // pass the basic type enum to the stub
5703 Node* elemType = intcon(bt);
5704
5705 // Call the stub
5706 const char *stubName = "array_partition_stub";
5707 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5708 stubAddr, stubName, TypePtr::BOTTOM,
5709 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5710 indexPivot1, indexPivot2);
5711
5712 } // original reexecute is set back here
5713
5714 if (!stopped()) {
5715 set_result(pivotIndices);
5716 }
5717
5718 return true;
5719 }
5720
5721
5722 //------------------------------inline_array_sort-----------------------
5723 bool LibraryCallKit::inline_array_sort() {
5724 address stubAddr = StubRoutines::select_arraysort_function();
5725 if (stubAddr == nullptr) {
5726 return false; // Intrinsic's stub is not implemented on this platform
5727 }
5728 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5729
5730 // no receiver because it is a static method
5731 Node* elementType = argument(0);
5732 Node* obj = argument(1);
5733 Node* offset = argument(2); // long
5734 Node* fromIndex = argument(4);
5735 Node* toIndex = argument(5);
5736 // SortOperation: argument(6) is ignored
5737
5738 BasicType bt = T_ILLEGAL;
5739
5740 if (!check_array_sort_arguments(elementType, obj, bt)) {
5741 return false;
5742 }
5743 null_check(obj);
5744 // If obj is dead, only null-path is taken.
5745 if (stopped()) {
5746 return true;
5747 }
5748 Node* obj_adr = make_unsafe_address(obj, offset);
5749
5750 // pass the basic type enum to the stub
5751 Node* elemType = intcon(bt);
5752
5753 // Call the stub.
5754 const char *stubName = "arraysort_stub";
5755 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5756 stubAddr, stubName, TypePtr::BOTTOM,
5757 obj_adr, elemType, fromIndex, toIndex);
5758
5759 return true;
5760 }
5761
5762
5763 //------------------------------inline_arraycopy-----------------------
5764 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5765 // Object dest, int destPos,
5766 // int length);
5767 bool LibraryCallKit::inline_arraycopy() {
5768 // Get the arguments.
5769 Node* src = argument(0); // type: oop
5770 Node* src_offset = argument(1); // type: int
5771 Node* dest = argument(2); // type: oop
5772 Node* dest_offset = argument(3); // type: int
5773 Node* length = argument(4); // type: int
5774
5775 uint new_idx = C->unique();
5776
5777 // Check for allocation before we add nodes that would confuse
5778 // tightly_coupled_allocation()
5779 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5780
5781 int saved_reexecute_sp = -1;
5782 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5783 // See arraycopy_restore_alloc_state() comment
5784 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5785 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5786 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5787 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5788
5789 // The following tests must be performed
5790 // (1) src and dest are arrays.
5791 // (2) src and dest arrays must have elements of the same BasicType
5792 // (3) src and dest must not be null.
5793 // (4) src_offset must not be negative.
5794 // (5) dest_offset must not be negative.
5795 // (6) length must not be negative.
5796 // (7) src_offset + length must not exceed length of src.
5797 // (8) dest_offset + length must not exceed length of dest.
5798 // (9) each element of an oop array must be assignable
5799
5800 // (3) src and dest must not be null.
5801 // always do this here because we need the JVM state for uncommon traps
5802 Node* null_ctl = top();
5803 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5804 assert(null_ctl->is_top(), "no null control here");
5805 dest = null_check(dest, T_ARRAY);
5806
5807 if (!can_emit_guards) {
5808 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5809 // guards but the arraycopy node could still take advantage of a
5810 // tightly allocated allocation. tightly_coupled_allocation() is
5811 // called again to make sure it takes the null check above into
5812 // account: the null check is mandatory and if it caused an
5813 // uncommon trap to be emitted then the allocation can't be
5814 // considered tightly coupled in this context.
5815 alloc = tightly_coupled_allocation(dest);
5816 }
5817
5818 bool validated = false;
5819
5820 const Type* src_type = _gvn.type(src);
5821 const Type* dest_type = _gvn.type(dest);
5822 const TypeAryPtr* top_src = src_type->isa_aryptr();
5823 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5824
5825 // Do we have the type of src?
5826 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5827 // Do we have the type of dest?
5828 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5829 // Is the type for src from speculation?
5830 bool src_spec = false;
5831 // Is the type for dest from speculation?
5832 bool dest_spec = false;
5833
5834 if ((!has_src || !has_dest) && can_emit_guards) {
5835 // We don't have sufficient type information, let's see if
5836 // speculative types can help. We need to have types for both src
5837 // and dest so that it pays off.
5838
5839 // Do we already have or could we have type information for src
5840 bool could_have_src = has_src;
5841 // Do we already have or could we have type information for dest
5842 bool could_have_dest = has_dest;
5843
5844 ciKlass* src_k = nullptr;
5845 if (!has_src) {
5846 src_k = src_type->speculative_type_not_null();
5847 if (src_k != nullptr && src_k->is_array_klass()) {
5848 could_have_src = true;
5849 }
5850 }
5851
5852 ciKlass* dest_k = nullptr;
5853 if (!has_dest) {
5854 dest_k = dest_type->speculative_type_not_null();
5855 if (dest_k != nullptr && dest_k->is_array_klass()) {
5856 could_have_dest = true;
5857 }
5858 }
5859
5860 if (could_have_src && could_have_dest) {
5861 // This is going to pay off so emit the required guards
5862 if (!has_src) {
5863 src = maybe_cast_profiled_obj(src, src_k, true);
5864 src_type = _gvn.type(src);
5865 top_src = src_type->isa_aryptr();
5866 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5867 src_spec = true;
5868 }
5869 if (!has_dest) {
5870 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5871 dest_type = _gvn.type(dest);
5872 top_dest = dest_type->isa_aryptr();
5873 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5874 dest_spec = true;
5875 }
5876 }
5877 }
5878
5879 if (has_src && has_dest && can_emit_guards) {
5880 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5881 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5882 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5883 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5884
5885 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5886 // If both arrays are object arrays then having the exact types
5887 // for both will remove the need for a subtype check at runtime
5888 // before the call and may make it possible to pick a faster copy
5889 // routine (without a subtype check on every element)
5890 // Do we have the exact type of src?
5891 bool could_have_src = src_spec;
5892 // Do we have the exact type of dest?
5893 bool could_have_dest = dest_spec;
5894 ciKlass* src_k = nullptr;
5895 ciKlass* dest_k = nullptr;
5896 if (!src_spec) {
5897 src_k = src_type->speculative_type_not_null();
5898 if (src_k != nullptr && src_k->is_array_klass()) {
5899 could_have_src = true;
5900 }
5901 }
5902 if (!dest_spec) {
5903 dest_k = dest_type->speculative_type_not_null();
5904 if (dest_k != nullptr && dest_k->is_array_klass()) {
5905 could_have_dest = true;
5906 }
5907 }
5908 if (could_have_src && could_have_dest) {
5909 // If we can have both exact types, emit the missing guards
5910 if (could_have_src && !src_spec) {
5911 src = maybe_cast_profiled_obj(src, src_k, true);
5912 }
5913 if (could_have_dest && !dest_spec) {
5914 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5915 }
5916 }
5917 }
5918 }
5919
5920 ciMethod* trap_method = method();
5921 int trap_bci = bci();
5922 if (saved_jvms_before_guards != nullptr) {
5923 trap_method = alloc->jvms()->method();
5924 trap_bci = alloc->jvms()->bci();
5925 }
5926
5927 bool negative_length_guard_generated = false;
5928
5929 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5930 can_emit_guards &&
5931 !src->is_top() && !dest->is_top()) {
5932 // validate arguments: enables transformation the ArrayCopyNode
5933 validated = true;
5934
5935 RegionNode* slow_region = new RegionNode(1);
5936 record_for_igvn(slow_region);
5937
5938 // (1) src and dest are arrays.
5939 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5940 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5941
5942 // (2) src and dest arrays must have elements of the same BasicType
5943 // done at macro expansion or at Ideal transformation time
5944
5945 // (4) src_offset must not be negative.
5946 generate_negative_guard(src_offset, slow_region);
5947
5948 // (5) dest_offset must not be negative.
5949 generate_negative_guard(dest_offset, slow_region);
5950
5951 // (7) src_offset + length must not exceed length of src.
5952 generate_limit_guard(src_offset, length,
5953 load_array_length(src),
5954 slow_region);
5955
5956 // (8) dest_offset + length must not exceed length of dest.
5957 generate_limit_guard(dest_offset, length,
5958 load_array_length(dest),
5959 slow_region);
5960
5961 // (6) length must not be negative.
5962 // This is also checked in generate_arraycopy() during macro expansion, but
5963 // we also have to check it here for the case where the ArrayCopyNode will
5964 // be eliminated by Escape Analysis.
5965 if (EliminateAllocations) {
5966 generate_negative_guard(length, slow_region);
5967 negative_length_guard_generated = true;
5968 }
5969
5970 // (9) each element of an oop array must be assignable
5971 Node* dest_klass = load_object_klass(dest);
5972 if (src != dest) {
5973 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5974
5975 if (not_subtype_ctrl != top()) {
5976 PreserveJVMState pjvms(this);
5977 set_control(not_subtype_ctrl);
5978 uncommon_trap(Deoptimization::Reason_intrinsic,
5979 Deoptimization::Action_make_not_entrant);
5980 assert(stopped(), "Should be stopped");
5981 }
5982 }
5983 {
5984 PreserveJVMState pjvms(this);
5985 set_control(_gvn.transform(slow_region));
5986 uncommon_trap(Deoptimization::Reason_intrinsic,
5987 Deoptimization::Action_make_not_entrant);
5988 assert(stopped(), "Should be stopped");
5989 }
5990
5991 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5992 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5993 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5994 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5995 }
5996
5997 if (stopped()) {
5998 return true;
5999 }
6000
6001 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6002 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6003 // so the compiler has a chance to eliminate them: during macro expansion,
6004 // we have to set their control (CastPP nodes are eliminated).
6005 load_object_klass(src), load_object_klass(dest),
6006 load_array_length(src), load_array_length(dest));
6007
6008 ac->set_arraycopy(validated);
6009
6010 Node* n = _gvn.transform(ac);
6011 if (n == ac) {
6012 ac->connect_outputs(this);
6013 } else {
6014 assert(validated, "shouldn't transform if all arguments not validated");
6015 set_all_memory(n);
6016 }
6017 clear_upper_avx();
6018
6019
6020 return true;
6021 }
6022
6023
6024 // Helper function which determines if an arraycopy immediately follows
6025 // an allocation, with no intervening tests or other escapes for the object.
6026 AllocateArrayNode*
6027 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6028 if (stopped()) return nullptr; // no fast path
6029 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6030
6031 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6032 if (alloc == nullptr) return nullptr;
6033
6034 Node* rawmem = memory(Compile::AliasIdxRaw);
6035 // Is the allocation's memory state untouched?
6036 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6037 // Bail out if there have been raw-memory effects since the allocation.
6038 // (Example: There might have been a call or safepoint.)
6039 return nullptr;
6040 }
6041 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6042 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6043 return nullptr;
6044 }
6045
6046 // There must be no unexpected observers of this allocation.
6047 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6048 Node* obs = ptr->fast_out(i);
6049 if (obs != this->map()) {
6050 return nullptr;
6051 }
6052 }
6053
6054 // This arraycopy must unconditionally follow the allocation of the ptr.
6055 Node* alloc_ctl = ptr->in(0);
6056 Node* ctl = control();
6057 while (ctl != alloc_ctl) {
6058 // There may be guards which feed into the slow_region.
6059 // Any other control flow means that we might not get a chance
6060 // to finish initializing the allocated object.
6061 // Various low-level checks bottom out in uncommon traps. These
6062 // are considered safe since we've already checked above that
6063 // there is no unexpected observer of this allocation.
6064 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6065 assert(ctl->in(0)->is_If(), "must be If");
6066 ctl = ctl->in(0)->in(0);
6067 } else {
6068 return nullptr;
6069 }
6070 }
6071
6072 // If we get this far, we have an allocation which immediately
6073 // precedes the arraycopy, and we can take over zeroing the new object.
6074 // The arraycopy will finish the initialization, and provide
6075 // a new control state to which we will anchor the destination pointer.
6076
6077 return alloc;
6078 }
6079
6080 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6081 if (node->is_IfProj()) {
6082 Node* other_proj = node->as_IfProj()->other_if_proj();
6083 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6084 Node* obs = other_proj->fast_out(j);
6085 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6086 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6087 return obs->as_CallStaticJava();
6088 }
6089 }
6090 }
6091 return nullptr;
6092 }
6093
6094 //-------------inline_encodeISOArray-----------------------------------
6095 // encode char[] to byte[] in ISO_8859_1 or ASCII
6096 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6097 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6098 // no receiver since it is static method
6099 Node *src = argument(0);
6100 Node *src_offset = argument(1);
6101 Node *dst = argument(2);
6102 Node *dst_offset = argument(3);
6103 Node *length = argument(4);
6104
6105 src = must_be_not_null(src, true);
6106 dst = must_be_not_null(dst, true);
6107
6108 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6109 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6110 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6111 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6112 // failed array check
6113 return false;
6114 }
6115
6116 // Figure out the size and type of the elements we will be copying.
6117 BasicType src_elem = src_type->elem()->array_element_basic_type();
6118 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6119 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6120 return false;
6121 }
6122
6123 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6124 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6125 // 'src_start' points to src array + scaled offset
6126 // 'dst_start' points to dst array + scaled offset
6127
6128 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6129 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6130 enc = _gvn.transform(enc);
6131 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6132 set_memory(res_mem, mtype);
6133 set_result(enc);
6134 clear_upper_avx();
6135
6136 return true;
6137 }
6138
6139 //-------------inline_multiplyToLen-----------------------------------
6140 bool LibraryCallKit::inline_multiplyToLen() {
6141 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6142
6143 address stubAddr = StubRoutines::multiplyToLen();
6144 if (stubAddr == nullptr) {
6145 return false; // Intrinsic's stub is not implemented on this platform
6146 }
6147 const char* stubName = "multiplyToLen";
6148
6149 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6150
6151 // no receiver because it is a static method
6152 Node* x = argument(0);
6153 Node* xlen = argument(1);
6154 Node* y = argument(2);
6155 Node* ylen = argument(3);
6156 Node* z = argument(4);
6157
6158 x = must_be_not_null(x, true);
6159 y = must_be_not_null(y, true);
6160
6161 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6162 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6163 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6164 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6165 // failed array check
6166 return false;
6167 }
6168
6169 BasicType x_elem = x_type->elem()->array_element_basic_type();
6170 BasicType y_elem = y_type->elem()->array_element_basic_type();
6171 if (x_elem != T_INT || y_elem != T_INT) {
6172 return false;
6173 }
6174
6175 Node* x_start = array_element_address(x, intcon(0), x_elem);
6176 Node* y_start = array_element_address(y, intcon(0), y_elem);
6177 // 'x_start' points to x array + scaled xlen
6178 // 'y_start' points to y array + scaled ylen
6179
6180 Node* z_start = array_element_address(z, intcon(0), T_INT);
6181
6182 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6183 OptoRuntime::multiplyToLen_Type(),
6184 stubAddr, stubName, TypePtr::BOTTOM,
6185 x_start, xlen, y_start, ylen, z_start);
6186
6187 C->set_has_split_ifs(true); // Has chance for split-if optimization
6188 set_result(z);
6189 return true;
6190 }
6191
6192 //-------------inline_squareToLen------------------------------------
6193 bool LibraryCallKit::inline_squareToLen() {
6194 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6195
6196 address stubAddr = StubRoutines::squareToLen();
6197 if (stubAddr == nullptr) {
6198 return false; // Intrinsic's stub is not implemented on this platform
6199 }
6200 const char* stubName = "squareToLen";
6201
6202 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6203
6204 Node* x = argument(0);
6205 Node* len = argument(1);
6206 Node* z = argument(2);
6207 Node* zlen = argument(3);
6208
6209 x = must_be_not_null(x, true);
6210 z = must_be_not_null(z, true);
6211
6212 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6213 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6214 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6215 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6216 // failed array check
6217 return false;
6218 }
6219
6220 BasicType x_elem = x_type->elem()->array_element_basic_type();
6221 BasicType z_elem = z_type->elem()->array_element_basic_type();
6222 if (x_elem != T_INT || z_elem != T_INT) {
6223 return false;
6224 }
6225
6226
6227 Node* x_start = array_element_address(x, intcon(0), x_elem);
6228 Node* z_start = array_element_address(z, intcon(0), z_elem);
6229
6230 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6231 OptoRuntime::squareToLen_Type(),
6232 stubAddr, stubName, TypePtr::BOTTOM,
6233 x_start, len, z_start, zlen);
6234
6235 set_result(z);
6236 return true;
6237 }
6238
6239 //-------------inline_mulAdd------------------------------------------
6240 bool LibraryCallKit::inline_mulAdd() {
6241 assert(UseMulAddIntrinsic, "not implemented on this platform");
6242
6243 address stubAddr = StubRoutines::mulAdd();
6244 if (stubAddr == nullptr) {
6245 return false; // Intrinsic's stub is not implemented on this platform
6246 }
6247 const char* stubName = "mulAdd";
6248
6249 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6250
6251 Node* out = argument(0);
6252 Node* in = argument(1);
6253 Node* offset = argument(2);
6254 Node* len = argument(3);
6255 Node* k = argument(4);
6256
6257 in = must_be_not_null(in, true);
6258 out = must_be_not_null(out, true);
6259
6260 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6261 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6262 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6263 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6264 // failed array check
6265 return false;
6266 }
6267
6268 BasicType out_elem = out_type->elem()->array_element_basic_type();
6269 BasicType in_elem = in_type->elem()->array_element_basic_type();
6270 if (out_elem != T_INT || in_elem != T_INT) {
6271 return false;
6272 }
6273
6274 Node* outlen = load_array_length(out);
6275 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6276 Node* out_start = array_element_address(out, intcon(0), out_elem);
6277 Node* in_start = array_element_address(in, intcon(0), in_elem);
6278
6279 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6280 OptoRuntime::mulAdd_Type(),
6281 stubAddr, stubName, TypePtr::BOTTOM,
6282 out_start,in_start, new_offset, len, k);
6283 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6284 set_result(result);
6285 return true;
6286 }
6287
6288 //-------------inline_montgomeryMultiply-----------------------------------
6289 bool LibraryCallKit::inline_montgomeryMultiply() {
6290 address stubAddr = StubRoutines::montgomeryMultiply();
6291 if (stubAddr == nullptr) {
6292 return false; // Intrinsic's stub is not implemented on this platform
6293 }
6294
6295 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6296 const char* stubName = "montgomery_multiply";
6297
6298 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6299
6300 Node* a = argument(0);
6301 Node* b = argument(1);
6302 Node* n = argument(2);
6303 Node* len = argument(3);
6304 Node* inv = argument(4);
6305 Node* m = argument(6);
6306
6307 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6308 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6309 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6310 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6311 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6312 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6313 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6314 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6315 // failed array check
6316 return false;
6317 }
6318
6319 BasicType a_elem = a_type->elem()->array_element_basic_type();
6320 BasicType b_elem = b_type->elem()->array_element_basic_type();
6321 BasicType n_elem = n_type->elem()->array_element_basic_type();
6322 BasicType m_elem = m_type->elem()->array_element_basic_type();
6323 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6324 return false;
6325 }
6326
6327 // Make the call
6328 {
6329 Node* a_start = array_element_address(a, intcon(0), a_elem);
6330 Node* b_start = array_element_address(b, intcon(0), b_elem);
6331 Node* n_start = array_element_address(n, intcon(0), n_elem);
6332 Node* m_start = array_element_address(m, intcon(0), m_elem);
6333
6334 Node* call = make_runtime_call(RC_LEAF,
6335 OptoRuntime::montgomeryMultiply_Type(),
6336 stubAddr, stubName, TypePtr::BOTTOM,
6337 a_start, b_start, n_start, len, inv, top(),
6338 m_start);
6339 set_result(m);
6340 }
6341
6342 return true;
6343 }
6344
6345 bool LibraryCallKit::inline_montgomerySquare() {
6346 address stubAddr = StubRoutines::montgomerySquare();
6347 if (stubAddr == nullptr) {
6348 return false; // Intrinsic's stub is not implemented on this platform
6349 }
6350
6351 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6352 const char* stubName = "montgomery_square";
6353
6354 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6355
6356 Node* a = argument(0);
6357 Node* n = argument(1);
6358 Node* len = argument(2);
6359 Node* inv = argument(3);
6360 Node* m = argument(5);
6361
6362 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6363 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6364 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6365 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6366 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6367 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6368 // failed array check
6369 return false;
6370 }
6371
6372 BasicType a_elem = a_type->elem()->array_element_basic_type();
6373 BasicType n_elem = n_type->elem()->array_element_basic_type();
6374 BasicType m_elem = m_type->elem()->array_element_basic_type();
6375 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6376 return false;
6377 }
6378
6379 // Make the call
6380 {
6381 Node* a_start = array_element_address(a, intcon(0), a_elem);
6382 Node* n_start = array_element_address(n, intcon(0), n_elem);
6383 Node* m_start = array_element_address(m, intcon(0), m_elem);
6384
6385 Node* call = make_runtime_call(RC_LEAF,
6386 OptoRuntime::montgomerySquare_Type(),
6387 stubAddr, stubName, TypePtr::BOTTOM,
6388 a_start, n_start, len, inv, top(),
6389 m_start);
6390 set_result(m);
6391 }
6392
6393 return true;
6394 }
6395
6396 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6397 address stubAddr = nullptr;
6398 const char* stubName = nullptr;
6399
6400 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6401 if (stubAddr == nullptr) {
6402 return false; // Intrinsic's stub is not implemented on this platform
6403 }
6404
6405 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6406
6407 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6408
6409 Node* newArr = argument(0);
6410 Node* oldArr = argument(1);
6411 Node* newIdx = argument(2);
6412 Node* shiftCount = argument(3);
6413 Node* numIter = argument(4);
6414
6415 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6416 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6417 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6418 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6419 return false;
6420 }
6421
6422 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6423 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6424 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6425 return false;
6426 }
6427
6428 // Make the call
6429 {
6430 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6431 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6432
6433 Node* call = make_runtime_call(RC_LEAF,
6434 OptoRuntime::bigIntegerShift_Type(),
6435 stubAddr,
6436 stubName,
6437 TypePtr::BOTTOM,
6438 newArr_start,
6439 oldArr_start,
6440 newIdx,
6441 shiftCount,
6442 numIter);
6443 }
6444
6445 return true;
6446 }
6447
6448 //-------------inline_vectorizedMismatch------------------------------
6449 bool LibraryCallKit::inline_vectorizedMismatch() {
6450 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6451
6452 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6453 Node* obja = argument(0); // Object
6454 Node* aoffset = argument(1); // long
6455 Node* objb = argument(3); // Object
6456 Node* boffset = argument(4); // long
6457 Node* length = argument(6); // int
6458 Node* scale = argument(7); // int
6459
6460 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6461 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6462 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6463 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6464 scale == top()) {
6465 return false; // failed input validation
6466 }
6467
6468 Node* obja_adr = make_unsafe_address(obja, aoffset);
6469 Node* objb_adr = make_unsafe_address(objb, boffset);
6470
6471 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6472 //
6473 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6474 // if (length <= inline_limit) {
6475 // inline_path:
6476 // vmask = VectorMaskGen length
6477 // vload1 = LoadVectorMasked obja, vmask
6478 // vload2 = LoadVectorMasked objb, vmask
6479 // result1 = VectorCmpMasked vload1, vload2, vmask
6480 // } else {
6481 // call_stub_path:
6482 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6483 // }
6484 // exit_block:
6485 // return Phi(result1, result2);
6486 //
6487 enum { inline_path = 1, // input is small enough to process it all at once
6488 stub_path = 2, // input is too large; call into the VM
6489 PATH_LIMIT = 3
6490 };
6491
6492 Node* exit_block = new RegionNode(PATH_LIMIT);
6493 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6494 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6495
6496 Node* call_stub_path = control();
6497
6498 BasicType elem_bt = T_ILLEGAL;
6499
6500 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6501 if (scale_t->is_con()) {
6502 switch (scale_t->get_con()) {
6503 case 0: elem_bt = T_BYTE; break;
6504 case 1: elem_bt = T_SHORT; break;
6505 case 2: elem_bt = T_INT; break;
6506 case 3: elem_bt = T_LONG; break;
6507
6508 default: elem_bt = T_ILLEGAL; break; // not supported
6509 }
6510 }
6511
6512 int inline_limit = 0;
6513 bool do_partial_inline = false;
6514
6515 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6516 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6517 do_partial_inline = inline_limit >= 16;
6518 }
6519
6520 if (do_partial_inline) {
6521 assert(elem_bt != T_ILLEGAL, "sanity");
6522
6523 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6524 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6525 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6526
6527 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6528 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6529 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6530
6531 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6532
6533 if (!stopped()) {
6534 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6535
6536 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6537 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6538 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6539 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6540
6541 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6542 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6543 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6544 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6545
6546 exit_block->init_req(inline_path, control());
6547 memory_phi->init_req(inline_path, map()->memory());
6548 result_phi->init_req(inline_path, result);
6549
6550 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6551 clear_upper_avx();
6552 }
6553 }
6554 }
6555
6556 if (call_stub_path != nullptr) {
6557 set_control(call_stub_path);
6558
6559 Node* call = make_runtime_call(RC_LEAF,
6560 OptoRuntime::vectorizedMismatch_Type(),
6561 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6562 obja_adr, objb_adr, length, scale);
6563
6564 exit_block->init_req(stub_path, control());
6565 memory_phi->init_req(stub_path, map()->memory());
6566 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6567 }
6568
6569 exit_block = _gvn.transform(exit_block);
6570 memory_phi = _gvn.transform(memory_phi);
6571 result_phi = _gvn.transform(result_phi);
6572
6573 set_control(exit_block);
6574 set_all_memory(memory_phi);
6575 set_result(result_phi);
6576
6577 return true;
6578 }
6579
6580 //------------------------------inline_vectorizedHashcode----------------------------
6581 bool LibraryCallKit::inline_vectorizedHashCode() {
6582 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6583
6584 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6585 Node* array = argument(0);
6586 Node* offset = argument(1);
6587 Node* length = argument(2);
6588 Node* initialValue = argument(3);
6589 Node* basic_type = argument(4);
6590
6591 if (basic_type == top()) {
6592 return false; // failed input validation
6593 }
6594
6595 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6596 if (!basic_type_t->is_con()) {
6597 return false; // Only intrinsify if mode argument is constant
6598 }
6599
6600 array = must_be_not_null(array, true);
6601
6602 BasicType bt = (BasicType)basic_type_t->get_con();
6603
6604 // Resolve address of first element
6605 Node* array_start = array_element_address(array, offset, bt);
6606
6607 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6608 array_start, length, initialValue, basic_type)));
6609 clear_upper_avx();
6610
6611 return true;
6612 }
6613
6614 /**
6615 * Calculate CRC32 for byte.
6616 * int java.util.zip.CRC32.update(int crc, int b)
6617 */
6618 bool LibraryCallKit::inline_updateCRC32() {
6619 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6620 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6621 // no receiver since it is static method
6622 Node* crc = argument(0); // type: int
6623 Node* b = argument(1); // type: int
6624
6625 /*
6626 * int c = ~ crc;
6627 * b = timesXtoThe32[(b ^ c) & 0xFF];
6628 * b = b ^ (c >>> 8);
6629 * crc = ~b;
6630 */
6631
6632 Node* M1 = intcon(-1);
6633 crc = _gvn.transform(new XorINode(crc, M1));
6634 Node* result = _gvn.transform(new XorINode(crc, b));
6635 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6636
6637 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6638 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6639 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6640 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6641
6642 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6643 result = _gvn.transform(new XorINode(crc, result));
6644 result = _gvn.transform(new XorINode(result, M1));
6645 set_result(result);
6646 return true;
6647 }
6648
6649 /**
6650 * Calculate CRC32 for byte[] array.
6651 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6652 */
6653 bool LibraryCallKit::inline_updateBytesCRC32() {
6654 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6655 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6656 // no receiver since it is static method
6657 Node* crc = argument(0); // type: int
6658 Node* src = argument(1); // type: oop
6659 Node* offset = argument(2); // type: int
6660 Node* length = argument(3); // type: int
6661
6662 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6663 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6664 // failed array check
6665 return false;
6666 }
6667
6668 // Figure out the size and type of the elements we will be copying.
6669 BasicType src_elem = src_type->elem()->array_element_basic_type();
6670 if (src_elem != T_BYTE) {
6671 return false;
6672 }
6673
6674 // 'src_start' points to src array + scaled offset
6675 src = must_be_not_null(src, true);
6676 Node* src_start = array_element_address(src, offset, src_elem);
6677
6678 // We assume that range check is done by caller.
6679 // TODO: generate range check (offset+length < src.length) in debug VM.
6680
6681 // Call the stub.
6682 address stubAddr = StubRoutines::updateBytesCRC32();
6683 const char *stubName = "updateBytesCRC32";
6684
6685 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6686 stubAddr, stubName, TypePtr::BOTTOM,
6687 crc, src_start, length);
6688 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6689 set_result(result);
6690 return true;
6691 }
6692
6693 /**
6694 * Calculate CRC32 for ByteBuffer.
6695 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6696 */
6697 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6698 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6699 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6700 // no receiver since it is static method
6701 Node* crc = argument(0); // type: int
6702 Node* src = argument(1); // type: long
6703 Node* offset = argument(3); // type: int
6704 Node* length = argument(4); // type: int
6705
6706 src = ConvL2X(src); // adjust Java long to machine word
6707 Node* base = _gvn.transform(new CastX2PNode(src));
6708 offset = ConvI2X(offset);
6709
6710 // 'src_start' points to src array + scaled offset
6711 Node* src_start = basic_plus_adr(top(), base, offset);
6712
6713 // Call the stub.
6714 address stubAddr = StubRoutines::updateBytesCRC32();
6715 const char *stubName = "updateBytesCRC32";
6716
6717 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6718 stubAddr, stubName, TypePtr::BOTTOM,
6719 crc, src_start, length);
6720 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6721 set_result(result);
6722 return true;
6723 }
6724
6725 //------------------------------get_table_from_crc32c_class-----------------------
6726 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6727 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6728 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6729
6730 return table;
6731 }
6732
6733 //------------------------------inline_updateBytesCRC32C-----------------------
6734 //
6735 // Calculate CRC32C for byte[] array.
6736 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6737 //
6738 bool LibraryCallKit::inline_updateBytesCRC32C() {
6739 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6740 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6741 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6742 // no receiver since it is a static method
6743 Node* crc = argument(0); // type: int
6744 Node* src = argument(1); // type: oop
6745 Node* offset = argument(2); // type: int
6746 Node* end = argument(3); // type: int
6747
6748 Node* length = _gvn.transform(new SubINode(end, offset));
6749
6750 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6751 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6752 // failed array check
6753 return false;
6754 }
6755
6756 // Figure out the size and type of the elements we will be copying.
6757 BasicType src_elem = src_type->elem()->array_element_basic_type();
6758 if (src_elem != T_BYTE) {
6759 return false;
6760 }
6761
6762 // 'src_start' points to src array + scaled offset
6763 src = must_be_not_null(src, true);
6764 Node* src_start = array_element_address(src, offset, src_elem);
6765
6766 // static final int[] byteTable in class CRC32C
6767 Node* table = get_table_from_crc32c_class(callee()->holder());
6768 table = must_be_not_null(table, true);
6769 Node* table_start = array_element_address(table, intcon(0), T_INT);
6770
6771 // We assume that range check is done by caller.
6772 // TODO: generate range check (offset+length < src.length) in debug VM.
6773
6774 // Call the stub.
6775 address stubAddr = StubRoutines::updateBytesCRC32C();
6776 const char *stubName = "updateBytesCRC32C";
6777
6778 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6779 stubAddr, stubName, TypePtr::BOTTOM,
6780 crc, src_start, length, table_start);
6781 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6782 set_result(result);
6783 return true;
6784 }
6785
6786 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6787 //
6788 // Calculate CRC32C for DirectByteBuffer.
6789 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6790 //
6791 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6792 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6793 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6794 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6795 // no receiver since it is a static method
6796 Node* crc = argument(0); // type: int
6797 Node* src = argument(1); // type: long
6798 Node* offset = argument(3); // type: int
6799 Node* end = argument(4); // type: int
6800
6801 Node* length = _gvn.transform(new SubINode(end, offset));
6802
6803 src = ConvL2X(src); // adjust Java long to machine word
6804 Node* base = _gvn.transform(new CastX2PNode(src));
6805 offset = ConvI2X(offset);
6806
6807 // 'src_start' points to src array + scaled offset
6808 Node* src_start = basic_plus_adr(top(), base, offset);
6809
6810 // static final int[] byteTable in class CRC32C
6811 Node* table = get_table_from_crc32c_class(callee()->holder());
6812 table = must_be_not_null(table, true);
6813 Node* table_start = array_element_address(table, intcon(0), T_INT);
6814
6815 // Call the stub.
6816 address stubAddr = StubRoutines::updateBytesCRC32C();
6817 const char *stubName = "updateBytesCRC32C";
6818
6819 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6820 stubAddr, stubName, TypePtr::BOTTOM,
6821 crc, src_start, length, table_start);
6822 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6823 set_result(result);
6824 return true;
6825 }
6826
6827 //------------------------------inline_updateBytesAdler32----------------------
6828 //
6829 // Calculate Adler32 checksum for byte[] array.
6830 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6831 //
6832 bool LibraryCallKit::inline_updateBytesAdler32() {
6833 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6834 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6835 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6836 // no receiver since it is static method
6837 Node* crc = argument(0); // type: int
6838 Node* src = argument(1); // type: oop
6839 Node* offset = argument(2); // type: int
6840 Node* length = argument(3); // type: int
6841
6842 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6843 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6844 // failed array check
6845 return false;
6846 }
6847
6848 // Figure out the size and type of the elements we will be copying.
6849 BasicType src_elem = src_type->elem()->array_element_basic_type();
6850 if (src_elem != T_BYTE) {
6851 return false;
6852 }
6853
6854 // 'src_start' points to src array + scaled offset
6855 Node* src_start = array_element_address(src, offset, src_elem);
6856
6857 // We assume that range check is done by caller.
6858 // TODO: generate range check (offset+length < src.length) in debug VM.
6859
6860 // Call the stub.
6861 address stubAddr = StubRoutines::updateBytesAdler32();
6862 const char *stubName = "updateBytesAdler32";
6863
6864 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6865 stubAddr, stubName, TypePtr::BOTTOM,
6866 crc, src_start, length);
6867 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6868 set_result(result);
6869 return true;
6870 }
6871
6872 //------------------------------inline_updateByteBufferAdler32---------------
6873 //
6874 // Calculate Adler32 checksum for DirectByteBuffer.
6875 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6876 //
6877 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6878 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6879 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6880 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6881 // no receiver since it is static method
6882 Node* crc = argument(0); // type: int
6883 Node* src = argument(1); // type: long
6884 Node* offset = argument(3); // type: int
6885 Node* length = argument(4); // type: int
6886
6887 src = ConvL2X(src); // adjust Java long to machine word
6888 Node* base = _gvn.transform(new CastX2PNode(src));
6889 offset = ConvI2X(offset);
6890
6891 // 'src_start' points to src array + scaled offset
6892 Node* src_start = basic_plus_adr(top(), base, offset);
6893
6894 // Call the stub.
6895 address stubAddr = StubRoutines::updateBytesAdler32();
6896 const char *stubName = "updateBytesAdler32";
6897
6898 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6899 stubAddr, stubName, TypePtr::BOTTOM,
6900 crc, src_start, length);
6901
6902 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6903 set_result(result);
6904 return true;
6905 }
6906
6907 //----------------------------inline_reference_get----------------------------
6908 // public T java.lang.ref.Reference.get();
6909 bool LibraryCallKit::inline_reference_get() {
6910 const int referent_offset = java_lang_ref_Reference::referent_offset();
6911
6912 // Get the argument:
6913 Node* reference_obj = null_check_receiver();
6914 if (stopped()) return true;
6915
6916 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6917 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6918 decorators, /*is_static*/ false, nullptr);
6919 if (result == nullptr) return false;
6920
6921 // Add memory barrier to prevent commoning reads from this field
6922 // across safepoint since GC can change its value.
6923 insert_mem_bar(Op_MemBarCPUOrder);
6924
6925 set_result(result);
6926 return true;
6927 }
6928
6929 //----------------------------inline_reference_refersTo0----------------------------
6930 // bool java.lang.ref.Reference.refersTo0();
6931 // bool java.lang.ref.PhantomReference.refersTo0();
6932 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6933 // Get arguments:
6934 Node* reference_obj = null_check_receiver();
6935 Node* other_obj = argument(1);
6936 if (stopped()) return true;
6937
6938 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6939 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6940 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6941 decorators, /*is_static*/ false, nullptr);
6942 if (referent == nullptr) return false;
6943
6944 // Add memory barrier to prevent commoning reads from this field
6945 // across safepoint since GC can change its value.
6946 insert_mem_bar(Op_MemBarCPUOrder);
6947
6948 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6949 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6950 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6951
6952 RegionNode* region = new RegionNode(3);
6953 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6954
6955 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6956 region->init_req(1, if_true);
6957 phi->init_req(1, intcon(1));
6958
6959 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6960 region->init_req(2, if_false);
6961 phi->init_req(2, intcon(0));
6962
6963 set_control(_gvn.transform(region));
6964 record_for_igvn(region);
6965 set_result(_gvn.transform(phi));
6966 return true;
6967 }
6968
6969 //----------------------------inline_reference_clear0----------------------------
6970 // void java.lang.ref.Reference.clear0();
6971 // void java.lang.ref.PhantomReference.clear0();
6972 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6973 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6974
6975 // Get arguments
6976 Node* reference_obj = null_check_receiver();
6977 if (stopped()) return true;
6978
6979 // Common access parameters
6980 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6981 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6982 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6983 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6984 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6985
6986 Node* referent = access_load_at(reference_obj,
6987 referent_field_addr,
6988 referent_field_addr_type,
6989 val_type,
6990 T_OBJECT,
6991 decorators);
6992
6993 IdealKit ideal(this);
6994 #define __ ideal.
6995 __ if_then(referent, BoolTest::ne, null());
6996 sync_kit(ideal);
6997 access_store_at(reference_obj,
6998 referent_field_addr,
6999 referent_field_addr_type,
7000 null(),
7001 val_type,
7002 T_OBJECT,
7003 decorators);
7004 __ sync_kit(this);
7005 __ end_if();
7006 final_sync(ideal);
7007 #undef __
7008
7009 return true;
7010 }
7011
7012 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7013 DecoratorSet decorators, bool is_static,
7014 ciInstanceKlass* fromKls) {
7015 if (fromKls == nullptr) {
7016 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7017 assert(tinst != nullptr, "obj is null");
7018 assert(tinst->is_loaded(), "obj is not loaded");
7019 fromKls = tinst->instance_klass();
7020 } else {
7021 assert(is_static, "only for static field access");
7022 }
7023 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7024 ciSymbol::make(fieldTypeString),
7025 is_static);
7026
7027 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7028 if (field == nullptr) return (Node *) nullptr;
7029
7030 if (is_static) {
7031 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7032 fromObj = makecon(tip);
7033 }
7034
7035 // Next code copied from Parse::do_get_xxx():
7036
7037 // Compute address and memory type.
7038 int offset = field->offset_in_bytes();
7039 bool is_vol = field->is_volatile();
7040 ciType* field_klass = field->type();
7041 assert(field_klass->is_loaded(), "should be loaded");
7042 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7043 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7044 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7045 "slice of address and input slice don't match");
7046 BasicType bt = field->layout_type();
7047
7048 // Build the resultant type of the load
7049 const Type *type;
7050 if (bt == T_OBJECT) {
7051 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7052 } else {
7053 type = Type::get_const_basic_type(bt);
7054 }
7055
7056 if (is_vol) {
7057 decorators |= MO_SEQ_CST;
7058 }
7059
7060 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7061 }
7062
7063 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7064 bool is_exact /* true */, bool is_static /* false */,
7065 ciInstanceKlass * fromKls /* nullptr */) {
7066 if (fromKls == nullptr) {
7067 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7068 assert(tinst != nullptr, "obj is null");
7069 assert(tinst->is_loaded(), "obj is not loaded");
7070 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7071 fromKls = tinst->instance_klass();
7072 }
7073 else {
7074 assert(is_static, "only for static field access");
7075 }
7076 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7077 ciSymbol::make(fieldTypeString),
7078 is_static);
7079
7080 assert(field != nullptr, "undefined field");
7081 assert(!field->is_volatile(), "not defined for volatile fields");
7082
7083 if (is_static) {
7084 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7085 fromObj = makecon(tip);
7086 }
7087
7088 // Next code copied from Parse::do_get_xxx():
7089
7090 // Compute address and memory type.
7091 int offset = field->offset_in_bytes();
7092 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7093
7094 return adr;
7095 }
7096
7097 //------------------------------inline_aescrypt_Block-----------------------
7098 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7099 address stubAddr = nullptr;
7100 const char *stubName;
7101 assert(UseAES, "need AES instruction support");
7102
7103 switch(id) {
7104 case vmIntrinsics::_aescrypt_encryptBlock:
7105 stubAddr = StubRoutines::aescrypt_encryptBlock();
7106 stubName = "aescrypt_encryptBlock";
7107 break;
7108 case vmIntrinsics::_aescrypt_decryptBlock:
7109 stubAddr = StubRoutines::aescrypt_decryptBlock();
7110 stubName = "aescrypt_decryptBlock";
7111 break;
7112 default:
7113 break;
7114 }
7115 if (stubAddr == nullptr) return false;
7116
7117 Node* aescrypt_object = argument(0);
7118 Node* src = argument(1);
7119 Node* src_offset = argument(2);
7120 Node* dest = argument(3);
7121 Node* dest_offset = argument(4);
7122
7123 src = must_be_not_null(src, true);
7124 dest = must_be_not_null(dest, true);
7125
7126 // (1) src and dest are arrays.
7127 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7128 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7129 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7130 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7131
7132 // for the quick and dirty code we will skip all the checks.
7133 // we are just trying to get the call to be generated.
7134 Node* src_start = src;
7135 Node* dest_start = dest;
7136 if (src_offset != nullptr || dest_offset != nullptr) {
7137 assert(src_offset != nullptr && dest_offset != nullptr, "");
7138 src_start = array_element_address(src, src_offset, T_BYTE);
7139 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7140 }
7141
7142 // now need to get the start of its expanded key array
7143 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7144 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7145 if (k_start == nullptr) return false;
7146
7147 // Call the stub.
7148 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7149 stubAddr, stubName, TypePtr::BOTTOM,
7150 src_start, dest_start, k_start);
7151
7152 return true;
7153 }
7154
7155 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7156 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7157 address stubAddr = nullptr;
7158 const char *stubName = nullptr;
7159
7160 assert(UseAES, "need AES instruction support");
7161
7162 switch(id) {
7163 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7164 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7165 stubName = "cipherBlockChaining_encryptAESCrypt";
7166 break;
7167 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7168 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7169 stubName = "cipherBlockChaining_decryptAESCrypt";
7170 break;
7171 default:
7172 break;
7173 }
7174 if (stubAddr == nullptr) return false;
7175
7176 Node* cipherBlockChaining_object = argument(0);
7177 Node* src = argument(1);
7178 Node* src_offset = argument(2);
7179 Node* len = argument(3);
7180 Node* dest = argument(4);
7181 Node* dest_offset = argument(5);
7182
7183 src = must_be_not_null(src, false);
7184 dest = must_be_not_null(dest, false);
7185
7186 // (1) src and dest are arrays.
7187 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7188 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7189 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7190 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7191
7192 // checks are the responsibility of the caller
7193 Node* src_start = src;
7194 Node* dest_start = dest;
7195 if (src_offset != nullptr || dest_offset != nullptr) {
7196 assert(src_offset != nullptr && dest_offset != nullptr, "");
7197 src_start = array_element_address(src, src_offset, T_BYTE);
7198 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7199 }
7200
7201 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7202 // (because of the predicated logic executed earlier).
7203 // so we cast it here safely.
7204 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7205
7206 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7207 if (embeddedCipherObj == nullptr) return false;
7208
7209 // cast it to what we know it will be at runtime
7210 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7211 assert(tinst != nullptr, "CBC obj is null");
7212 assert(tinst->is_loaded(), "CBC obj is not loaded");
7213 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7214 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7215
7216 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7217 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7218 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7219 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7220 aescrypt_object = _gvn.transform(aescrypt_object);
7221
7222 // we need to get the start of the aescrypt_object's expanded key array
7223 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7224 if (k_start == nullptr) return false;
7225
7226 // similarly, get the start address of the r vector
7227 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7228 if (objRvec == nullptr) return false;
7229 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7230
7231 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7232 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7233 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7234 stubAddr, stubName, TypePtr::BOTTOM,
7235 src_start, dest_start, k_start, r_start, len);
7236
7237 // return cipher length (int)
7238 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7239 set_result(retvalue);
7240 return true;
7241 }
7242
7243 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7244 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7245 address stubAddr = nullptr;
7246 const char *stubName = nullptr;
7247
7248 assert(UseAES, "need AES instruction support");
7249
7250 switch (id) {
7251 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7252 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7253 stubName = "electronicCodeBook_encryptAESCrypt";
7254 break;
7255 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7256 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7257 stubName = "electronicCodeBook_decryptAESCrypt";
7258 break;
7259 default:
7260 break;
7261 }
7262
7263 if (stubAddr == nullptr) return false;
7264
7265 Node* electronicCodeBook_object = argument(0);
7266 Node* src = argument(1);
7267 Node* src_offset = argument(2);
7268 Node* len = argument(3);
7269 Node* dest = argument(4);
7270 Node* dest_offset = argument(5);
7271
7272 // (1) src and dest are arrays.
7273 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7274 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7275 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7276 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7277
7278 // checks are the responsibility of the caller
7279 Node* src_start = src;
7280 Node* dest_start = dest;
7281 if (src_offset != nullptr || dest_offset != nullptr) {
7282 assert(src_offset != nullptr && dest_offset != nullptr, "");
7283 src_start = array_element_address(src, src_offset, T_BYTE);
7284 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7285 }
7286
7287 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7288 // (because of the predicated logic executed earlier).
7289 // so we cast it here safely.
7290 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7291
7292 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7293 if (embeddedCipherObj == nullptr) return false;
7294
7295 // cast it to what we know it will be at runtime
7296 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7297 assert(tinst != nullptr, "ECB obj is null");
7298 assert(tinst->is_loaded(), "ECB obj is not loaded");
7299 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7300 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7301
7302 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7303 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7304 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7305 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7306 aescrypt_object = _gvn.transform(aescrypt_object);
7307
7308 // we need to get the start of the aescrypt_object's expanded key array
7309 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7310 if (k_start == nullptr) return false;
7311
7312 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7313 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7314 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7315 stubAddr, stubName, TypePtr::BOTTOM,
7316 src_start, dest_start, k_start, len);
7317
7318 // return cipher length (int)
7319 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7320 set_result(retvalue);
7321 return true;
7322 }
7323
7324 //------------------------------inline_counterMode_AESCrypt-----------------------
7325 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7326 assert(UseAES, "need AES instruction support");
7327 if (!UseAESCTRIntrinsics) return false;
7328
7329 address stubAddr = nullptr;
7330 const char *stubName = nullptr;
7331 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7332 stubAddr = StubRoutines::counterMode_AESCrypt();
7333 stubName = "counterMode_AESCrypt";
7334 }
7335 if (stubAddr == nullptr) return false;
7336
7337 Node* counterMode_object = argument(0);
7338 Node* src = argument(1);
7339 Node* src_offset = argument(2);
7340 Node* len = argument(3);
7341 Node* dest = argument(4);
7342 Node* dest_offset = argument(5);
7343
7344 // (1) src and dest are arrays.
7345 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7346 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7347 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7348 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7349
7350 // checks are the responsibility of the caller
7351 Node* src_start = src;
7352 Node* dest_start = dest;
7353 if (src_offset != nullptr || dest_offset != nullptr) {
7354 assert(src_offset != nullptr && dest_offset != nullptr, "");
7355 src_start = array_element_address(src, src_offset, T_BYTE);
7356 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7357 }
7358
7359 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7360 // (because of the predicated logic executed earlier).
7361 // so we cast it here safely.
7362 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7363 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7364 if (embeddedCipherObj == nullptr) return false;
7365 // cast it to what we know it will be at runtime
7366 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7367 assert(tinst != nullptr, "CTR obj is null");
7368 assert(tinst->is_loaded(), "CTR obj is not loaded");
7369 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7370 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7371 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7372 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7373 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7374 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7375 aescrypt_object = _gvn.transform(aescrypt_object);
7376 // we need to get the start of the aescrypt_object's expanded key array
7377 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7378 if (k_start == nullptr) return false;
7379 // similarly, get the start address of the r vector
7380 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7381 if (obj_counter == nullptr) return false;
7382 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7383
7384 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7385 if (saved_encCounter == nullptr) return false;
7386 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7387 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7388
7389 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7390 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7391 OptoRuntime::counterMode_aescrypt_Type(),
7392 stubAddr, stubName, TypePtr::BOTTOM,
7393 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7394
7395 // return cipher length (int)
7396 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7397 set_result(retvalue);
7398 return true;
7399 }
7400
7401 //------------------------------get_key_start_from_aescrypt_object-----------------------
7402 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7403 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7404 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7405 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7406 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7407 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7408 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7409 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7410 if (objSessionK == nullptr) {
7411 return (Node *) nullptr;
7412 }
7413 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7414 #else
7415 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7416 #endif // PPC64
7417 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7418 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7419
7420 // now have the array, need to get the start address of the K array
7421 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7422 return k_start;
7423 }
7424
7425 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7426 // Return node representing slow path of predicate check.
7427 // the pseudo code we want to emulate with this predicate is:
7428 // for encryption:
7429 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7430 // for decryption:
7431 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7432 // note cipher==plain is more conservative than the original java code but that's OK
7433 //
7434 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7435 // The receiver was checked for null already.
7436 Node* objCBC = argument(0);
7437
7438 Node* src = argument(1);
7439 Node* dest = argument(4);
7440
7441 // Load embeddedCipher field of CipherBlockChaining object.
7442 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7443
7444 // get AESCrypt klass for instanceOf check
7445 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7446 // will have same classloader as CipherBlockChaining object
7447 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7448 assert(tinst != nullptr, "CBCobj is null");
7449 assert(tinst->is_loaded(), "CBCobj is not loaded");
7450
7451 // we want to do an instanceof comparison against the AESCrypt class
7452 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7453 if (!klass_AESCrypt->is_loaded()) {
7454 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7455 Node* ctrl = control();
7456 set_control(top()); // no regular fast path
7457 return ctrl;
7458 }
7459
7460 src = must_be_not_null(src, true);
7461 dest = must_be_not_null(dest, true);
7462
7463 // Resolve oops to stable for CmpP below.
7464 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7465
7466 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7467 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7468 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7469
7470 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7471
7472 // for encryption, we are done
7473 if (!decrypting)
7474 return instof_false; // even if it is null
7475
7476 // for decryption, we need to add a further check to avoid
7477 // taking the intrinsic path when cipher and plain are the same
7478 // see the original java code for why.
7479 RegionNode* region = new RegionNode(3);
7480 region->init_req(1, instof_false);
7481
7482 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7483 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7484 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7485 region->init_req(2, src_dest_conjoint);
7486
7487 record_for_igvn(region);
7488 return _gvn.transform(region);
7489 }
7490
7491 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7492 // Return node representing slow path of predicate check.
7493 // the pseudo code we want to emulate with this predicate is:
7494 // for encryption:
7495 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7496 // for decryption:
7497 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7498 // note cipher==plain is more conservative than the original java code but that's OK
7499 //
7500 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7501 // The receiver was checked for null already.
7502 Node* objECB = argument(0);
7503
7504 // Load embeddedCipher field of ElectronicCodeBook object.
7505 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7506
7507 // get AESCrypt klass for instanceOf check
7508 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7509 // will have same classloader as ElectronicCodeBook object
7510 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7511 assert(tinst != nullptr, "ECBobj is null");
7512 assert(tinst->is_loaded(), "ECBobj is not loaded");
7513
7514 // we want to do an instanceof comparison against the AESCrypt class
7515 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7516 if (!klass_AESCrypt->is_loaded()) {
7517 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7518 Node* ctrl = control();
7519 set_control(top()); // no regular fast path
7520 return ctrl;
7521 }
7522 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7523
7524 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7525 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7526 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7527
7528 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7529
7530 // for encryption, we are done
7531 if (!decrypting)
7532 return instof_false; // even if it is null
7533
7534 // for decryption, we need to add a further check to avoid
7535 // taking the intrinsic path when cipher and plain are the same
7536 // see the original java code for why.
7537 RegionNode* region = new RegionNode(3);
7538 region->init_req(1, instof_false);
7539 Node* src = argument(1);
7540 Node* dest = argument(4);
7541 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7542 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7543 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7544 region->init_req(2, src_dest_conjoint);
7545
7546 record_for_igvn(region);
7547 return _gvn.transform(region);
7548 }
7549
7550 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7551 // Return node representing slow path of predicate check.
7552 // the pseudo code we want to emulate with this predicate is:
7553 // for encryption:
7554 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7555 // for decryption:
7556 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7557 // note cipher==plain is more conservative than the original java code but that's OK
7558 //
7559
7560 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7561 // The receiver was checked for null already.
7562 Node* objCTR = argument(0);
7563
7564 // Load embeddedCipher field of CipherBlockChaining object.
7565 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7566
7567 // get AESCrypt klass for instanceOf check
7568 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7569 // will have same classloader as CipherBlockChaining object
7570 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7571 assert(tinst != nullptr, "CTRobj is null");
7572 assert(tinst->is_loaded(), "CTRobj is not loaded");
7573
7574 // we want to do an instanceof comparison against the AESCrypt class
7575 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7576 if (!klass_AESCrypt->is_loaded()) {
7577 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7578 Node* ctrl = control();
7579 set_control(top()); // no regular fast path
7580 return ctrl;
7581 }
7582
7583 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7584 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7585 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7586 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7587 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7588
7589 return instof_false; // even if it is null
7590 }
7591
7592 //------------------------------inline_ghash_processBlocks
7593 bool LibraryCallKit::inline_ghash_processBlocks() {
7594 address stubAddr;
7595 const char *stubName;
7596 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7597
7598 stubAddr = StubRoutines::ghash_processBlocks();
7599 stubName = "ghash_processBlocks";
7600
7601 Node* data = argument(0);
7602 Node* offset = argument(1);
7603 Node* len = argument(2);
7604 Node* state = argument(3);
7605 Node* subkeyH = argument(4);
7606
7607 state = must_be_not_null(state, true);
7608 subkeyH = must_be_not_null(subkeyH, true);
7609 data = must_be_not_null(data, true);
7610
7611 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7612 assert(state_start, "state is null");
7613 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7614 assert(subkeyH_start, "subkeyH is null");
7615 Node* data_start = array_element_address(data, offset, T_BYTE);
7616 assert(data_start, "data is null");
7617
7618 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7619 OptoRuntime::ghash_processBlocks_Type(),
7620 stubAddr, stubName, TypePtr::BOTTOM,
7621 state_start, subkeyH_start, data_start, len);
7622 return true;
7623 }
7624
7625 //------------------------------inline_chacha20Block
7626 bool LibraryCallKit::inline_chacha20Block() {
7627 address stubAddr;
7628 const char *stubName;
7629 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7630
7631 stubAddr = StubRoutines::chacha20Block();
7632 stubName = "chacha20Block";
7633
7634 Node* state = argument(0);
7635 Node* result = argument(1);
7636
7637 state = must_be_not_null(state, true);
7638 result = must_be_not_null(result, true);
7639
7640 Node* state_start = array_element_address(state, intcon(0), T_INT);
7641 assert(state_start, "state is null");
7642 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7643 assert(result_start, "result is null");
7644
7645 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7646 OptoRuntime::chacha20Block_Type(),
7647 stubAddr, stubName, TypePtr::BOTTOM,
7648 state_start, result_start);
7649 // return key stream length (int)
7650 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7651 set_result(retvalue);
7652 return true;
7653 }
7654
7655 //------------------------------inline_dilithiumAlmostNtt
7656 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7657 address stubAddr;
7658 const char *stubName;
7659 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7660 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7661
7662 stubAddr = StubRoutines::dilithiumAlmostNtt();
7663 stubName = "dilithiumAlmostNtt";
7664 if (!stubAddr) return false;
7665
7666 Node* coeffs = argument(0);
7667 Node* ntt_zetas = argument(1);
7668
7669 coeffs = must_be_not_null(coeffs, true);
7670 ntt_zetas = must_be_not_null(ntt_zetas, true);
7671
7672 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7673 assert(coeffs_start, "coeffs is null");
7674 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7675 assert(ntt_zetas_start, "ntt_zetas is null");
7676 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7677 OptoRuntime::dilithiumAlmostNtt_Type(),
7678 stubAddr, stubName, TypePtr::BOTTOM,
7679 coeffs_start, ntt_zetas_start);
7680 // return an int
7681 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7682 set_result(retvalue);
7683 return true;
7684 }
7685
7686 //------------------------------inline_dilithiumAlmostInverseNtt
7687 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7688 address stubAddr;
7689 const char *stubName;
7690 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7691 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7692
7693 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7694 stubName = "dilithiumAlmostInverseNtt";
7695 if (!stubAddr) return false;
7696
7697 Node* coeffs = argument(0);
7698 Node* zetas = argument(1);
7699
7700 coeffs = must_be_not_null(coeffs, true);
7701 zetas = must_be_not_null(zetas, true);
7702
7703 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7704 assert(coeffs_start, "coeffs is null");
7705 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7706 assert(zetas_start, "inverseNtt_zetas is null");
7707 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7708 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7709 stubAddr, stubName, TypePtr::BOTTOM,
7710 coeffs_start, zetas_start);
7711
7712 // return an int
7713 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7714 set_result(retvalue);
7715 return true;
7716 }
7717
7718 //------------------------------inline_dilithiumNttMult
7719 bool LibraryCallKit::inline_dilithiumNttMult() {
7720 address stubAddr;
7721 const char *stubName;
7722 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7723 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7724
7725 stubAddr = StubRoutines::dilithiumNttMult();
7726 stubName = "dilithiumNttMult";
7727 if (!stubAddr) return false;
7728
7729 Node* result = argument(0);
7730 Node* ntta = argument(1);
7731 Node* nttb = argument(2);
7732
7733 result = must_be_not_null(result, true);
7734 ntta = must_be_not_null(ntta, true);
7735 nttb = must_be_not_null(nttb, true);
7736
7737 Node* result_start = array_element_address(result, intcon(0), T_INT);
7738 assert(result_start, "result is null");
7739 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
7740 assert(ntta_start, "ntta is null");
7741 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
7742 assert(nttb_start, "nttb is null");
7743 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7744 OptoRuntime::dilithiumNttMult_Type(),
7745 stubAddr, stubName, TypePtr::BOTTOM,
7746 result_start, ntta_start, nttb_start);
7747
7748 // return an int
7749 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
7750 set_result(retvalue);
7751
7752 return true;
7753 }
7754
7755 //------------------------------inline_dilithiumMontMulByConstant
7756 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
7757 address stubAddr;
7758 const char *stubName;
7759 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7760 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
7761
7762 stubAddr = StubRoutines::dilithiumMontMulByConstant();
7763 stubName = "dilithiumMontMulByConstant";
7764 if (!stubAddr) return false;
7765
7766 Node* coeffs = argument(0);
7767 Node* constant = argument(1);
7768
7769 coeffs = must_be_not_null(coeffs, true);
7770
7771 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7772 assert(coeffs_start, "coeffs is null");
7773 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
7774 OptoRuntime::dilithiumMontMulByConstant_Type(),
7775 stubAddr, stubName, TypePtr::BOTTOM,
7776 coeffs_start, constant);
7777
7778 // return an int
7779 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
7780 set_result(retvalue);
7781 return true;
7782 }
7783
7784
7785 //------------------------------inline_dilithiumDecomposePoly
7786 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
7787 address stubAddr;
7788 const char *stubName;
7789 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7790 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
7791
7792 stubAddr = StubRoutines::dilithiumDecomposePoly();
7793 stubName = "dilithiumDecomposePoly";
7794 if (!stubAddr) return false;
7795
7796 Node* input = argument(0);
7797 Node* lowPart = argument(1);
7798 Node* highPart = argument(2);
7799 Node* twoGamma2 = argument(3);
7800 Node* multiplier = argument(4);
7801
7802 input = must_be_not_null(input, true);
7803 lowPart = must_be_not_null(lowPart, true);
7804 highPart = must_be_not_null(highPart, true);
7805
7806 Node* input_start = array_element_address(input, intcon(0), T_INT);
7807 assert(input_start, "input is null");
7808 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
7809 assert(lowPart_start, "lowPart is null");
7810 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
7811 assert(highPart_start, "highPart is null");
7812
7813 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
7814 OptoRuntime::dilithiumDecomposePoly_Type(),
7815 stubAddr, stubName, TypePtr::BOTTOM,
7816 input_start, lowPart_start, highPart_start,
7817 twoGamma2, multiplier);
7818
7819 // return an int
7820 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
7821 set_result(retvalue);
7822 return true;
7823 }
7824
7825 bool LibraryCallKit::inline_base64_encodeBlock() {
7826 address stubAddr;
7827 const char *stubName;
7828 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7829 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7830 stubAddr = StubRoutines::base64_encodeBlock();
7831 stubName = "encodeBlock";
7832
7833 if (!stubAddr) return false;
7834 Node* base64obj = argument(0);
7835 Node* src = argument(1);
7836 Node* offset = argument(2);
7837 Node* len = argument(3);
7838 Node* dest = argument(4);
7839 Node* dp = argument(5);
7840 Node* isURL = argument(6);
7841
7842 src = must_be_not_null(src, true);
7843 dest = must_be_not_null(dest, true);
7844
7845 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7846 assert(src_start, "source array is null");
7847 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7848 assert(dest_start, "destination array is null");
7849
7850 Node* base64 = make_runtime_call(RC_LEAF,
7851 OptoRuntime::base64_encodeBlock_Type(),
7852 stubAddr, stubName, TypePtr::BOTTOM,
7853 src_start, offset, len, dest_start, dp, isURL);
7854 return true;
7855 }
7856
7857 bool LibraryCallKit::inline_base64_decodeBlock() {
7858 address stubAddr;
7859 const char *stubName;
7860 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7861 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7862 stubAddr = StubRoutines::base64_decodeBlock();
7863 stubName = "decodeBlock";
7864
7865 if (!stubAddr) return false;
7866 Node* base64obj = argument(0);
7867 Node* src = argument(1);
7868 Node* src_offset = argument(2);
7869 Node* len = argument(3);
7870 Node* dest = argument(4);
7871 Node* dest_offset = argument(5);
7872 Node* isURL = argument(6);
7873 Node* isMIME = argument(7);
7874
7875 src = must_be_not_null(src, true);
7876 dest = must_be_not_null(dest, true);
7877
7878 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7879 assert(src_start, "source array is null");
7880 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7881 assert(dest_start, "destination array is null");
7882
7883 Node* call = make_runtime_call(RC_LEAF,
7884 OptoRuntime::base64_decodeBlock_Type(),
7885 stubAddr, stubName, TypePtr::BOTTOM,
7886 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7887 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7888 set_result(result);
7889 return true;
7890 }
7891
7892 bool LibraryCallKit::inline_poly1305_processBlocks() {
7893 address stubAddr;
7894 const char *stubName;
7895 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7896 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7897 stubAddr = StubRoutines::poly1305_processBlocks();
7898 stubName = "poly1305_processBlocks";
7899
7900 if (!stubAddr) return false;
7901 null_check_receiver(); // null-check receiver
7902 if (stopped()) return true;
7903
7904 Node* input = argument(1);
7905 Node* input_offset = argument(2);
7906 Node* len = argument(3);
7907 Node* alimbs = argument(4);
7908 Node* rlimbs = argument(5);
7909
7910 input = must_be_not_null(input, true);
7911 alimbs = must_be_not_null(alimbs, true);
7912 rlimbs = must_be_not_null(rlimbs, true);
7913
7914 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7915 assert(input_start, "input array is null");
7916 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7917 assert(acc_start, "acc array is null");
7918 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7919 assert(r_start, "r array is null");
7920
7921 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7922 OptoRuntime::poly1305_processBlocks_Type(),
7923 stubAddr, stubName, TypePtr::BOTTOM,
7924 input_start, len, acc_start, r_start);
7925 return true;
7926 }
7927
7928 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7929 address stubAddr;
7930 const char *stubName;
7931 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7932 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7933 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7934 stubName = "intpoly_montgomeryMult_P256";
7935
7936 if (!stubAddr) return false;
7937 null_check_receiver(); // null-check receiver
7938 if (stopped()) return true;
7939
7940 Node* a = argument(1);
7941 Node* b = argument(2);
7942 Node* r = argument(3);
7943
7944 a = must_be_not_null(a, true);
7945 b = must_be_not_null(b, true);
7946 r = must_be_not_null(r, true);
7947
7948 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7949 assert(a_start, "a array is null");
7950 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7951 assert(b_start, "b array is null");
7952 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7953 assert(r_start, "r array is null");
7954
7955 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7956 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7957 stubAddr, stubName, TypePtr::BOTTOM,
7958 a_start, b_start, r_start);
7959 return true;
7960 }
7961
7962 bool LibraryCallKit::inline_intpoly_assign() {
7963 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7964 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7965 const char *stubName = "intpoly_assign";
7966 address stubAddr = StubRoutines::intpoly_assign();
7967 if (!stubAddr) return false;
7968
7969 Node* set = argument(0);
7970 Node* a = argument(1);
7971 Node* b = argument(2);
7972 Node* arr_length = load_array_length(a);
7973
7974 a = must_be_not_null(a, true);
7975 b = must_be_not_null(b, true);
7976
7977 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7978 assert(a_start, "a array is null");
7979 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7980 assert(b_start, "b array is null");
7981
7982 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7983 OptoRuntime::intpoly_assign_Type(),
7984 stubAddr, stubName, TypePtr::BOTTOM,
7985 set, a_start, b_start, arr_length);
7986 return true;
7987 }
7988
7989 //------------------------------inline_digestBase_implCompress-----------------------
7990 //
7991 // Calculate MD5 for single-block byte[] array.
7992 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7993 //
7994 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7995 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7996 //
7997 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7998 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7999 //
8000 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8001 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8002 //
8003 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8004 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8005 //
8006 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8007 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8008
8009 Node* digestBase_obj = argument(0);
8010 Node* src = argument(1); // type oop
8011 Node* ofs = argument(2); // type int
8012
8013 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8014 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8015 // failed array check
8016 return false;
8017 }
8018 // Figure out the size and type of the elements we will be copying.
8019 BasicType src_elem = src_type->elem()->array_element_basic_type();
8020 if (src_elem != T_BYTE) {
8021 return false;
8022 }
8023 // 'src_start' points to src array + offset
8024 src = must_be_not_null(src, true);
8025 Node* src_start = array_element_address(src, ofs, src_elem);
8026 Node* state = nullptr;
8027 Node* block_size = nullptr;
8028 address stubAddr;
8029 const char *stubName;
8030
8031 switch(id) {
8032 case vmIntrinsics::_md5_implCompress:
8033 assert(UseMD5Intrinsics, "need MD5 instruction support");
8034 state = get_state_from_digest_object(digestBase_obj, T_INT);
8035 stubAddr = StubRoutines::md5_implCompress();
8036 stubName = "md5_implCompress";
8037 break;
8038 case vmIntrinsics::_sha_implCompress:
8039 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8040 state = get_state_from_digest_object(digestBase_obj, T_INT);
8041 stubAddr = StubRoutines::sha1_implCompress();
8042 stubName = "sha1_implCompress";
8043 break;
8044 case vmIntrinsics::_sha2_implCompress:
8045 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8046 state = get_state_from_digest_object(digestBase_obj, T_INT);
8047 stubAddr = StubRoutines::sha256_implCompress();
8048 stubName = "sha256_implCompress";
8049 break;
8050 case vmIntrinsics::_sha5_implCompress:
8051 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8052 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8053 stubAddr = StubRoutines::sha512_implCompress();
8054 stubName = "sha512_implCompress";
8055 break;
8056 case vmIntrinsics::_sha3_implCompress:
8057 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8058 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8059 stubAddr = StubRoutines::sha3_implCompress();
8060 stubName = "sha3_implCompress";
8061 block_size = get_block_size_from_digest_object(digestBase_obj);
8062 if (block_size == nullptr) return false;
8063 break;
8064 default:
8065 fatal_unexpected_iid(id);
8066 return false;
8067 }
8068 if (state == nullptr) return false;
8069
8070 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8071 if (stubAddr == nullptr) return false;
8072
8073 // Call the stub.
8074 Node* call;
8075 if (block_size == nullptr) {
8076 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8077 stubAddr, stubName, TypePtr::BOTTOM,
8078 src_start, state);
8079 } else {
8080 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8081 stubAddr, stubName, TypePtr::BOTTOM,
8082 src_start, state, block_size);
8083 }
8084
8085 return true;
8086 }
8087
8088 //------------------------------inline_double_keccak
8089 bool LibraryCallKit::inline_double_keccak() {
8090 address stubAddr;
8091 const char *stubName;
8092 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8093 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8094
8095 stubAddr = StubRoutines::double_keccak();
8096 stubName = "double_keccak";
8097 if (!stubAddr) return false;
8098
8099 Node* status0 = argument(0);
8100 Node* status1 = argument(1);
8101
8102 status0 = must_be_not_null(status0, true);
8103 status1 = must_be_not_null(status1, true);
8104
8105 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8106 assert(status0_start, "status0 is null");
8107 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8108 assert(status1_start, "status1 is null");
8109 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8110 OptoRuntime::double_keccak_Type(),
8111 stubAddr, stubName, TypePtr::BOTTOM,
8112 status0_start, status1_start);
8113 // return an int
8114 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8115 set_result(retvalue);
8116 return true;
8117 }
8118
8119
8120 //------------------------------inline_digestBase_implCompressMB-----------------------
8121 //
8122 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8123 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8124 //
8125 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8126 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8127 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8128 assert((uint)predicate < 5, "sanity");
8129 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8130
8131 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8132 Node* src = argument(1); // byte[] array
8133 Node* ofs = argument(2); // type int
8134 Node* limit = argument(3); // type int
8135
8136 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8137 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8138 // failed array check
8139 return false;
8140 }
8141 // Figure out the size and type of the elements we will be copying.
8142 BasicType src_elem = src_type->elem()->array_element_basic_type();
8143 if (src_elem != T_BYTE) {
8144 return false;
8145 }
8146 // 'src_start' points to src array + offset
8147 src = must_be_not_null(src, false);
8148 Node* src_start = array_element_address(src, ofs, src_elem);
8149
8150 const char* klass_digestBase_name = nullptr;
8151 const char* stub_name = nullptr;
8152 address stub_addr = nullptr;
8153 BasicType elem_type = T_INT;
8154
8155 switch (predicate) {
8156 case 0:
8157 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8158 klass_digestBase_name = "sun/security/provider/MD5";
8159 stub_name = "md5_implCompressMB";
8160 stub_addr = StubRoutines::md5_implCompressMB();
8161 }
8162 break;
8163 case 1:
8164 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8165 klass_digestBase_name = "sun/security/provider/SHA";
8166 stub_name = "sha1_implCompressMB";
8167 stub_addr = StubRoutines::sha1_implCompressMB();
8168 }
8169 break;
8170 case 2:
8171 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8172 klass_digestBase_name = "sun/security/provider/SHA2";
8173 stub_name = "sha256_implCompressMB";
8174 stub_addr = StubRoutines::sha256_implCompressMB();
8175 }
8176 break;
8177 case 3:
8178 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8179 klass_digestBase_name = "sun/security/provider/SHA5";
8180 stub_name = "sha512_implCompressMB";
8181 stub_addr = StubRoutines::sha512_implCompressMB();
8182 elem_type = T_LONG;
8183 }
8184 break;
8185 case 4:
8186 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8187 klass_digestBase_name = "sun/security/provider/SHA3";
8188 stub_name = "sha3_implCompressMB";
8189 stub_addr = StubRoutines::sha3_implCompressMB();
8190 elem_type = T_LONG;
8191 }
8192 break;
8193 default:
8194 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8195 }
8196 if (klass_digestBase_name != nullptr) {
8197 assert(stub_addr != nullptr, "Stub is generated");
8198 if (stub_addr == nullptr) return false;
8199
8200 // get DigestBase klass to lookup for SHA klass
8201 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8202 assert(tinst != nullptr, "digestBase_obj is not instance???");
8203 assert(tinst->is_loaded(), "DigestBase is not loaded");
8204
8205 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8206 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8207 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8208 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8209 }
8210 return false;
8211 }
8212
8213 //------------------------------inline_digestBase_implCompressMB-----------------------
8214 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8215 BasicType elem_type, address stubAddr, const char *stubName,
8216 Node* src_start, Node* ofs, Node* limit) {
8217 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8218 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8219 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8220 digest_obj = _gvn.transform(digest_obj);
8221
8222 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8223 if (state == nullptr) return false;
8224
8225 Node* block_size = nullptr;
8226 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8227 block_size = get_block_size_from_digest_object(digest_obj);
8228 if (block_size == nullptr) return false;
8229 }
8230
8231 // Call the stub.
8232 Node* call;
8233 if (block_size == nullptr) {
8234 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8235 OptoRuntime::digestBase_implCompressMB_Type(false),
8236 stubAddr, stubName, TypePtr::BOTTOM,
8237 src_start, state, ofs, limit);
8238 } else {
8239 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8240 OptoRuntime::digestBase_implCompressMB_Type(true),
8241 stubAddr, stubName, TypePtr::BOTTOM,
8242 src_start, state, block_size, ofs, limit);
8243 }
8244
8245 // return ofs (int)
8246 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8247 set_result(result);
8248
8249 return true;
8250 }
8251
8252 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8253 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8254 assert(UseAES, "need AES instruction support");
8255 address stubAddr = nullptr;
8256 const char *stubName = nullptr;
8257 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8258 stubName = "galoisCounterMode_AESCrypt";
8259
8260 if (stubAddr == nullptr) return false;
8261
8262 Node* in = argument(0);
8263 Node* inOfs = argument(1);
8264 Node* len = argument(2);
8265 Node* ct = argument(3);
8266 Node* ctOfs = argument(4);
8267 Node* out = argument(5);
8268 Node* outOfs = argument(6);
8269 Node* gctr_object = argument(7);
8270 Node* ghash_object = argument(8);
8271
8272 // (1) in, ct and out are arrays.
8273 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8274 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8275 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8276 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8277 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8278 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8279
8280 // checks are the responsibility of the caller
8281 Node* in_start = in;
8282 Node* ct_start = ct;
8283 Node* out_start = out;
8284 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8285 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8286 in_start = array_element_address(in, inOfs, T_BYTE);
8287 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8288 out_start = array_element_address(out, outOfs, T_BYTE);
8289 }
8290
8291 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8292 // (because of the predicated logic executed earlier).
8293 // so we cast it here safely.
8294 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8295 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8296 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8297 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8298 Node* state = load_field_from_object(ghash_object, "state", "[J");
8299
8300 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8301 return false;
8302 }
8303 // cast it to what we know it will be at runtime
8304 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8305 assert(tinst != nullptr, "GCTR obj is null");
8306 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8307 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8308 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8309 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8310 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8311 const TypeOopPtr* xtype = aklass->as_instance_type();
8312 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8313 aescrypt_object = _gvn.transform(aescrypt_object);
8314 // we need to get the start of the aescrypt_object's expanded key array
8315 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8316 if (k_start == nullptr) return false;
8317 // similarly, get the start address of the r vector
8318 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8319 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8320 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8321
8322
8323 // Call the stub, passing params
8324 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8325 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8326 stubAddr, stubName, TypePtr::BOTTOM,
8327 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8328
8329 // return cipher length (int)
8330 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8331 set_result(retvalue);
8332
8333 return true;
8334 }
8335
8336 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8337 // Return node representing slow path of predicate check.
8338 // the pseudo code we want to emulate with this predicate is:
8339 // for encryption:
8340 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8341 // for decryption:
8342 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8343 // note cipher==plain is more conservative than the original java code but that's OK
8344 //
8345
8346 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8347 // The receiver was checked for null already.
8348 Node* objGCTR = argument(7);
8349 // Load embeddedCipher field of GCTR object.
8350 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8351 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8352
8353 // get AESCrypt klass for instanceOf check
8354 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8355 // will have same classloader as CipherBlockChaining object
8356 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8357 assert(tinst != nullptr, "GCTR obj is null");
8358 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8359
8360 // we want to do an instanceof comparison against the AESCrypt class
8361 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8362 if (!klass_AESCrypt->is_loaded()) {
8363 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8364 Node* ctrl = control();
8365 set_control(top()); // no regular fast path
8366 return ctrl;
8367 }
8368
8369 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8370 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8371 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8372 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8373 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8374
8375 return instof_false; // even if it is null
8376 }
8377
8378 //------------------------------get_state_from_digest_object-----------------------
8379 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8380 const char* state_type;
8381 switch (elem_type) {
8382 case T_BYTE: state_type = "[B"; break;
8383 case T_INT: state_type = "[I"; break;
8384 case T_LONG: state_type = "[J"; break;
8385 default: ShouldNotReachHere();
8386 }
8387 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8388 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8389 if (digest_state == nullptr) return (Node *) nullptr;
8390
8391 // now have the array, need to get the start address of the state array
8392 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8393 return state;
8394 }
8395
8396 //------------------------------get_block_size_from_sha3_object----------------------------------
8397 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8398 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8399 assert (block_size != nullptr, "sanity");
8400 return block_size;
8401 }
8402
8403 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8404 // Return node representing slow path of predicate check.
8405 // the pseudo code we want to emulate with this predicate is:
8406 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8407 //
8408 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8409 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8410 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8411 assert((uint)predicate < 5, "sanity");
8412
8413 // The receiver was checked for null already.
8414 Node* digestBaseObj = argument(0);
8415
8416 // get DigestBase klass for instanceOf check
8417 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8418 assert(tinst != nullptr, "digestBaseObj is null");
8419 assert(tinst->is_loaded(), "DigestBase is not loaded");
8420
8421 const char* klass_name = nullptr;
8422 switch (predicate) {
8423 case 0:
8424 if (UseMD5Intrinsics) {
8425 // we want to do an instanceof comparison against the MD5 class
8426 klass_name = "sun/security/provider/MD5";
8427 }
8428 break;
8429 case 1:
8430 if (UseSHA1Intrinsics) {
8431 // we want to do an instanceof comparison against the SHA class
8432 klass_name = "sun/security/provider/SHA";
8433 }
8434 break;
8435 case 2:
8436 if (UseSHA256Intrinsics) {
8437 // we want to do an instanceof comparison against the SHA2 class
8438 klass_name = "sun/security/provider/SHA2";
8439 }
8440 break;
8441 case 3:
8442 if (UseSHA512Intrinsics) {
8443 // we want to do an instanceof comparison against the SHA5 class
8444 klass_name = "sun/security/provider/SHA5";
8445 }
8446 break;
8447 case 4:
8448 if (UseSHA3Intrinsics) {
8449 // we want to do an instanceof comparison against the SHA3 class
8450 klass_name = "sun/security/provider/SHA3";
8451 }
8452 break;
8453 default:
8454 fatal("unknown SHA intrinsic predicate: %d", predicate);
8455 }
8456
8457 ciKlass* klass = nullptr;
8458 if (klass_name != nullptr) {
8459 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8460 }
8461 if ((klass == nullptr) || !klass->is_loaded()) {
8462 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8463 Node* ctrl = control();
8464 set_control(top()); // no intrinsic path
8465 return ctrl;
8466 }
8467 ciInstanceKlass* instklass = klass->as_instance_klass();
8468
8469 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8470 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8471 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8472 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8473
8474 return instof_false; // even if it is null
8475 }
8476
8477 //-------------inline_fma-----------------------------------
8478 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8479 Node *a = nullptr;
8480 Node *b = nullptr;
8481 Node *c = nullptr;
8482 Node* result = nullptr;
8483 switch (id) {
8484 case vmIntrinsics::_fmaD:
8485 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8486 // no receiver since it is static method
8487 a = round_double_node(argument(0));
8488 b = round_double_node(argument(2));
8489 c = round_double_node(argument(4));
8490 result = _gvn.transform(new FmaDNode(a, b, c));
8491 break;
8492 case vmIntrinsics::_fmaF:
8493 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8494 a = argument(0);
8495 b = argument(1);
8496 c = argument(2);
8497 result = _gvn.transform(new FmaFNode(a, b, c));
8498 break;
8499 default:
8500 fatal_unexpected_iid(id); break;
8501 }
8502 set_result(result);
8503 return true;
8504 }
8505
8506 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8507 // argument(0) is receiver
8508 Node* codePoint = argument(1);
8509 Node* n = nullptr;
8510
8511 switch (id) {
8512 case vmIntrinsics::_isDigit :
8513 n = new DigitNode(control(), codePoint);
8514 break;
8515 case vmIntrinsics::_isLowerCase :
8516 n = new LowerCaseNode(control(), codePoint);
8517 break;
8518 case vmIntrinsics::_isUpperCase :
8519 n = new UpperCaseNode(control(), codePoint);
8520 break;
8521 case vmIntrinsics::_isWhitespace :
8522 n = new WhitespaceNode(control(), codePoint);
8523 break;
8524 default:
8525 fatal_unexpected_iid(id);
8526 }
8527
8528 set_result(_gvn.transform(n));
8529 return true;
8530 }
8531
8532 bool LibraryCallKit::inline_profileBoolean() {
8533 Node* counts = argument(1);
8534 const TypeAryPtr* ary = nullptr;
8535 ciArray* aobj = nullptr;
8536 if (counts->is_Con()
8537 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8538 && (aobj = ary->const_oop()->as_array()) != nullptr
8539 && (aobj->length() == 2)) {
8540 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8541 jint false_cnt = aobj->element_value(0).as_int();
8542 jint true_cnt = aobj->element_value(1).as_int();
8543
8544 if (C->log() != nullptr) {
8545 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8546 false_cnt, true_cnt);
8547 }
8548
8549 if (false_cnt + true_cnt == 0) {
8550 // According to profile, never executed.
8551 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8552 Deoptimization::Action_reinterpret);
8553 return true;
8554 }
8555
8556 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8557 // is a number of each value occurrences.
8558 Node* result = argument(0);
8559 if (false_cnt == 0 || true_cnt == 0) {
8560 // According to profile, one value has been never seen.
8561 int expected_val = (false_cnt == 0) ? 1 : 0;
8562
8563 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8564 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8565
8566 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8567 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8568 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8569
8570 { // Slow path: uncommon trap for never seen value and then reexecute
8571 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8572 // the value has been seen at least once.
8573 PreserveJVMState pjvms(this);
8574 PreserveReexecuteState preexecs(this);
8575 jvms()->set_should_reexecute(true);
8576
8577 set_control(slow_path);
8578 set_i_o(i_o());
8579
8580 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8581 Deoptimization::Action_reinterpret);
8582 }
8583 // The guard for never seen value enables sharpening of the result and
8584 // returning a constant. It allows to eliminate branches on the same value
8585 // later on.
8586 set_control(fast_path);
8587 result = intcon(expected_val);
8588 }
8589 // Stop profiling.
8590 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8591 // By replacing method body with profile data (represented as ProfileBooleanNode
8592 // on IR level) we effectively disable profiling.
8593 // It enables full speed execution once optimized code is generated.
8594 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8595 C->record_for_igvn(profile);
8596 set_result(profile);
8597 return true;
8598 } else {
8599 // Continue profiling.
8600 // Profile data isn't available at the moment. So, execute method's bytecode version.
8601 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8602 // is compiled and counters aren't available since corresponding MethodHandle
8603 // isn't a compile-time constant.
8604 return false;
8605 }
8606 }
8607
8608 bool LibraryCallKit::inline_isCompileConstant() {
8609 Node* n = argument(0);
8610 set_result(n->is_Con() ? intcon(1) : intcon(0));
8611 return true;
8612 }
8613
8614 //------------------------------- inline_getObjectSize --------------------------------------
8615 //
8616 // Calculate the runtime size of the object/array.
8617 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8618 //
8619 bool LibraryCallKit::inline_getObjectSize() {
8620 Node* obj = argument(3);
8621 Node* klass_node = load_object_klass(obj);
8622
8623 jint layout_con = Klass::_lh_neutral_value;
8624 Node* layout_val = get_layout_helper(klass_node, layout_con);
8625 int layout_is_con = (layout_val == nullptr);
8626
8627 if (layout_is_con) {
8628 // Layout helper is constant, can figure out things at compile time.
8629
8630 if (Klass::layout_helper_is_instance(layout_con)) {
8631 // Instance case: layout_con contains the size itself.
8632 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8633 set_result(size);
8634 } else {
8635 // Array case: size is round(header + element_size*arraylength).
8636 // Since arraylength is different for every array instance, we have to
8637 // compute the whole thing at runtime.
8638
8639 Node* arr_length = load_array_length(obj);
8640
8641 int round_mask = MinObjAlignmentInBytes - 1;
8642 int hsize = Klass::layout_helper_header_size(layout_con);
8643 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8644
8645 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8646 round_mask = 0; // strength-reduce it if it goes away completely
8647 }
8648 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8649 Node* header_size = intcon(hsize + round_mask);
8650
8651 Node* lengthx = ConvI2X(arr_length);
8652 Node* headerx = ConvI2X(header_size);
8653
8654 Node* abody = lengthx;
8655 if (eshift != 0) {
8656 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8657 }
8658 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8659 if (round_mask != 0) {
8660 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8661 }
8662 size = ConvX2L(size);
8663 set_result(size);
8664 }
8665 } else {
8666 // Layout helper is not constant, need to test for array-ness at runtime.
8667
8668 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8669 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8670 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8671 record_for_igvn(result_reg);
8672
8673 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8674 if (array_ctl != nullptr) {
8675 // Array case: size is round(header + element_size*arraylength).
8676 // Since arraylength is different for every array instance, we have to
8677 // compute the whole thing at runtime.
8678
8679 PreserveJVMState pjvms(this);
8680 set_control(array_ctl);
8681 Node* arr_length = load_array_length(obj);
8682
8683 int round_mask = MinObjAlignmentInBytes - 1;
8684 Node* mask = intcon(round_mask);
8685
8686 Node* hss = intcon(Klass::_lh_header_size_shift);
8687 Node* hsm = intcon(Klass::_lh_header_size_mask);
8688 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8689 header_size = _gvn.transform(new AndINode(header_size, hsm));
8690 header_size = _gvn.transform(new AddINode(header_size, mask));
8691
8692 // There is no need to mask or shift this value.
8693 // The semantics of LShiftINode include an implicit mask to 0x1F.
8694 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8695 Node* elem_shift = layout_val;
8696
8697 Node* lengthx = ConvI2X(arr_length);
8698 Node* headerx = ConvI2X(header_size);
8699
8700 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8701 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8702 if (round_mask != 0) {
8703 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8704 }
8705 size = ConvX2L(size);
8706
8707 result_reg->init_req(_array_path, control());
8708 result_val->init_req(_array_path, size);
8709 }
8710
8711 if (!stopped()) {
8712 // Instance case: the layout helper gives us instance size almost directly,
8713 // but we need to mask out the _lh_instance_slow_path_bit.
8714 Node* size = ConvI2X(layout_val);
8715 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8716 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8717 size = _gvn.transform(new AndXNode(size, mask));
8718 size = ConvX2L(size);
8719
8720 result_reg->init_req(_instance_path, control());
8721 result_val->init_req(_instance_path, size);
8722 }
8723
8724 set_result(result_reg, result_val);
8725 }
8726
8727 return true;
8728 }
8729
8730 //------------------------------- inline_blackhole --------------------------------------
8731 //
8732 // Make sure all arguments to this node are alive.
8733 // This matches methods that were requested to be blackholed through compile commands.
8734 //
8735 bool LibraryCallKit::inline_blackhole() {
8736 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8737 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8738 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8739
8740 // Blackhole node pinches only the control, not memory. This allows
8741 // the blackhole to be pinned in the loop that computes blackholed
8742 // values, but have no other side effects, like breaking the optimizations
8743 // across the blackhole.
8744
8745 Node* bh = _gvn.transform(new BlackholeNode(control()));
8746 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8747
8748 // Bind call arguments as blackhole arguments to keep them alive
8749 uint nargs = callee()->arg_size();
8750 for (uint i = 0; i < nargs; i++) {
8751 bh->add_req(argument(i));
8752 }
8753
8754 return true;
8755 }
8756
8757 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
8758 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
8759 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
8760 return nullptr; // box klass is not Float16
8761 }
8762
8763 // Null check; get notnull casted pointer
8764 Node* null_ctl = top();
8765 Node* not_null_box = null_check_oop(box, &null_ctl, true);
8766 // If not_null_box is dead, only null-path is taken
8767 if (stopped()) {
8768 set_control(null_ctl);
8769 return nullptr;
8770 }
8771 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
8772 const TypePtr* adr_type = C->alias_type(field)->adr_type();
8773 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
8774 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
8775 }
8776
8777 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
8778 PreserveReexecuteState preexecs(this);
8779 jvms()->set_should_reexecute(true);
8780
8781 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
8782 Node* klass_node = makecon(klass_type);
8783 Node* box = new_instance(klass_node);
8784
8785 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
8786 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
8787
8788 Node* field_store = _gvn.transform(access_store_at(box,
8789 value_field,
8790 value_adr_type,
8791 value,
8792 TypeInt::SHORT,
8793 T_SHORT,
8794 IN_HEAP));
8795 set_memory(field_store, value_adr_type);
8796 return box;
8797 }
8798
8799 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
8800 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
8801 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
8802 return false;
8803 }
8804
8805 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
8806 if (box_type == nullptr || box_type->const_oop() == nullptr) {
8807 return false;
8808 }
8809
8810 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
8811 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
8812 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
8813 ciSymbols::short_signature(),
8814 false);
8815 assert(field != nullptr, "");
8816
8817 // Transformed nodes
8818 Node* fld1 = nullptr;
8819 Node* fld2 = nullptr;
8820 Node* fld3 = nullptr;
8821 switch(num_args) {
8822 case 3:
8823 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
8824 if (fld3 == nullptr) {
8825 return false;
8826 }
8827 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
8828 // fall-through
8829 case 2:
8830 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
8831 if (fld2 == nullptr) {
8832 return false;
8833 }
8834 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
8835 // fall-through
8836 case 1:
8837 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
8838 if (fld1 == nullptr) {
8839 return false;
8840 }
8841 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
8842 break;
8843 default: fatal("Unsupported number of arguments %d", num_args);
8844 }
8845
8846 Node* result = nullptr;
8847 switch (id) {
8848 // Unary operations
8849 case vmIntrinsics::_sqrt_float16:
8850 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
8851 break;
8852 // Ternary operations
8853 case vmIntrinsics::_fma_float16:
8854 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
8855 break;
8856 default:
8857 fatal_unexpected_iid(id);
8858 break;
8859 }
8860 result = _gvn.transform(new ReinterpretHF2SNode(result));
8861 set_result(box_fp16_value(float16_box_type, field, result));
8862 return true;
8863 }
8864