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