1 /*
2 * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/mountUnmountDisabler.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
572 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
573 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
574
575 case vmIntrinsics::_Class_cast: return inline_Class_cast();
576
577 case vmIntrinsics::_aescrypt_encryptBlock:
578 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
579
580 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
581 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
582 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
585 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
586 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_counterMode_AESCrypt:
589 return inline_counterMode_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_galoisCounterMode_AESCrypt:
592 return inline_galoisCounterMode_AESCrypt();
593
594 case vmIntrinsics::_md5_implCompress:
595 case vmIntrinsics::_sha_implCompress:
596 case vmIntrinsics::_sha2_implCompress:
597 case vmIntrinsics::_sha5_implCompress:
598 case vmIntrinsics::_sha3_implCompress:
599 return inline_digestBase_implCompress(intrinsic_id());
600 case vmIntrinsics::_double_keccak:
601 return inline_double_keccak();
602
603 case vmIntrinsics::_digestBase_implCompressMB:
604 return inline_digestBase_implCompressMB(predicate);
605
606 case vmIntrinsics::_multiplyToLen:
607 return inline_multiplyToLen();
608
609 case vmIntrinsics::_squareToLen:
610 return inline_squareToLen();
611
612 case vmIntrinsics::_mulAdd:
613 return inline_mulAdd();
614
615 case vmIntrinsics::_montgomeryMultiply:
616 return inline_montgomeryMultiply();
617 case vmIntrinsics::_montgomerySquare:
618 return inline_montgomerySquare();
619
620 case vmIntrinsics::_bigIntegerRightShiftWorker:
621 return inline_bigIntegerShift(true);
622 case vmIntrinsics::_bigIntegerLeftShiftWorker:
623 return inline_bigIntegerShift(false);
624
625 case vmIntrinsics::_vectorizedMismatch:
626 return inline_vectorizedMismatch();
627
628 case vmIntrinsics::_ghash_processBlocks:
629 return inline_ghash_processBlocks();
630 case vmIntrinsics::_chacha20Block:
631 return inline_chacha20Block();
632 case vmIntrinsics::_kyberNtt:
633 return inline_kyberNtt();
634 case vmIntrinsics::_kyberInverseNtt:
635 return inline_kyberInverseNtt();
636 case vmIntrinsics::_kyberNttMult:
637 return inline_kyberNttMult();
638 case vmIntrinsics::_kyberAddPoly_2:
639 return inline_kyberAddPoly_2();
640 case vmIntrinsics::_kyberAddPoly_3:
641 return inline_kyberAddPoly_3();
642 case vmIntrinsics::_kyber12To16:
643 return inline_kyber12To16();
644 case vmIntrinsics::_kyberBarrettReduce:
645 return inline_kyberBarrettReduce();
646 case vmIntrinsics::_dilithiumAlmostNtt:
647 return inline_dilithiumAlmostNtt();
648 case vmIntrinsics::_dilithiumAlmostInverseNtt:
649 return inline_dilithiumAlmostInverseNtt();
650 case vmIntrinsics::_dilithiumNttMult:
651 return inline_dilithiumNttMult();
652 case vmIntrinsics::_dilithiumMontMulByConstant:
653 return inline_dilithiumMontMulByConstant();
654 case vmIntrinsics::_dilithiumDecomposePoly:
655 return inline_dilithiumDecomposePoly();
656 case vmIntrinsics::_base64_encodeBlock:
657 return inline_base64_encodeBlock();
658 case vmIntrinsics::_base64_decodeBlock:
659 return inline_base64_decodeBlock();
660 case vmIntrinsics::_poly1305_processBlocks:
661 return inline_poly1305_processBlocks();
662 case vmIntrinsics::_intpoly_montgomeryMult_P256:
663 return inline_intpoly_montgomeryMult_P256();
664 case vmIntrinsics::_intpoly_assign:
665 return inline_intpoly_assign();
666 case vmIntrinsics::_encodeISOArray:
667 case vmIntrinsics::_encodeByteISOArray:
668 return inline_encodeISOArray(false);
669 case vmIntrinsics::_encodeAsciiArray:
670 return inline_encodeISOArray(true);
671
672 case vmIntrinsics::_updateCRC32:
673 return inline_updateCRC32();
674 case vmIntrinsics::_updateBytesCRC32:
675 return inline_updateBytesCRC32();
676 case vmIntrinsics::_updateByteBufferCRC32:
677 return inline_updateByteBufferCRC32();
678
679 case vmIntrinsics::_updateBytesCRC32C:
680 return inline_updateBytesCRC32C();
681 case vmIntrinsics::_updateDirectByteBufferCRC32C:
682 return inline_updateDirectByteBufferCRC32C();
683
684 case vmIntrinsics::_updateBytesAdler32:
685 return inline_updateBytesAdler32();
686 case vmIntrinsics::_updateByteBufferAdler32:
687 return inline_updateByteBufferAdler32();
688
689 case vmIntrinsics::_profileBoolean:
690 return inline_profileBoolean();
691 case vmIntrinsics::_isCompileConstant:
692 return inline_isCompileConstant();
693
694 case vmIntrinsics::_countPositives:
695 return inline_countPositives();
696
697 case vmIntrinsics::_fmaD:
698 case vmIntrinsics::_fmaF:
699 return inline_fma(intrinsic_id());
700
701 case vmIntrinsics::_isDigit:
702 case vmIntrinsics::_isLowerCase:
703 case vmIntrinsics::_isUpperCase:
704 case vmIntrinsics::_isWhitespace:
705 return inline_character_compare(intrinsic_id());
706
707 case vmIntrinsics::_min:
708 case vmIntrinsics::_max:
709 case vmIntrinsics::_min_strict:
710 case vmIntrinsics::_max_strict:
711 case vmIntrinsics::_minL:
712 case vmIntrinsics::_maxL:
713 case vmIntrinsics::_minF:
714 case vmIntrinsics::_maxF:
715 case vmIntrinsics::_minD:
716 case vmIntrinsics::_maxD:
717 case vmIntrinsics::_minF_strict:
718 case vmIntrinsics::_maxF_strict:
719 case vmIntrinsics::_minD_strict:
720 case vmIntrinsics::_maxD_strict:
721 return inline_min_max(intrinsic_id());
722
723 case vmIntrinsics::_VectorUnaryOp:
724 return inline_vector_nary_operation(1);
725 case vmIntrinsics::_VectorBinaryOp:
726 return inline_vector_nary_operation(2);
727 case vmIntrinsics::_VectorUnaryLibOp:
728 return inline_vector_call(1);
729 case vmIntrinsics::_VectorBinaryLibOp:
730 return inline_vector_call(2);
731 case vmIntrinsics::_VectorTernaryOp:
732 return inline_vector_nary_operation(3);
733 case vmIntrinsics::_VectorFromBitsCoerced:
734 return inline_vector_frombits_coerced();
735 case vmIntrinsics::_VectorMaskOp:
736 return inline_vector_mask_operation();
737 case vmIntrinsics::_VectorLoadOp:
738 return inline_vector_mem_operation(/*is_store=*/false);
739 case vmIntrinsics::_VectorLoadMaskedOp:
740 return inline_vector_mem_masked_operation(/*is_store*/false);
741 case vmIntrinsics::_VectorStoreOp:
742 return inline_vector_mem_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorStoreMaskedOp:
744 return inline_vector_mem_masked_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorGatherOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ false);
747 case vmIntrinsics::_VectorScatterOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ true);
749 case vmIntrinsics::_VectorReductionCoerced:
750 return inline_vector_reduction();
751 case vmIntrinsics::_VectorTest:
752 return inline_vector_test();
753 case vmIntrinsics::_VectorBlend:
754 return inline_vector_blend();
755 case vmIntrinsics::_VectorRearrange:
756 return inline_vector_rearrange();
757 case vmIntrinsics::_VectorSelectFrom:
758 return inline_vector_select_from();
759 case vmIntrinsics::_VectorCompare:
760 return inline_vector_compare();
761 case vmIntrinsics::_VectorBroadcastInt:
762 return inline_vector_broadcast_int();
763 case vmIntrinsics::_VectorConvert:
764 return inline_vector_convert();
765 case vmIntrinsics::_VectorInsert:
766 return inline_vector_insert();
767 case vmIntrinsics::_VectorExtract:
768 return inline_vector_extract();
769 case vmIntrinsics::_VectorCompressExpand:
770 return inline_vector_compress_expand();
771 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
772 return inline_vector_select_from_two_vectors();
773 case vmIntrinsics::_IndexVector:
774 return inline_index_vector();
775 case vmIntrinsics::_IndexPartiallyInUpperRange:
776 return inline_index_partially_in_upper_range();
777
778 case vmIntrinsics::_getObjectSize:
779 return inline_getObjectSize();
780
781 case vmIntrinsics::_blackhole:
782 return inline_blackhole();
783
784 default:
785 // If you get here, it may be that someone has added a new intrinsic
786 // to the list in vmIntrinsics.hpp without implementing it here.
787 #ifndef PRODUCT
788 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
789 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
790 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
791 }
792 #endif
793 return false;
794 }
795 }
796
797 Node* LibraryCallKit::try_to_predicate(int predicate) {
798 if (!jvms()->has_method()) {
799 // Root JVMState has a null method.
800 assert(map()->memory()->Opcode() == Op_Parm, "");
801 // Insert the memory aliasing node
802 set_all_memory(reset_memory());
803 }
804 assert(merged_memory(), "");
805
806 switch (intrinsic_id()) {
807 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(false);
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(true);
811 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(false);
813 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(true);
815 case vmIntrinsics::_counterMode_AESCrypt:
816 return inline_counterMode_AESCrypt_predicate();
817 case vmIntrinsics::_digestBase_implCompressMB:
818 return inline_digestBase_implCompressMB_predicate(predicate);
819 case vmIntrinsics::_galoisCounterMode_AESCrypt:
820 return inline_galoisCounterMode_AESCrypt_predicate();
821
822 default:
823 // If you get here, it may be that someone has added a new intrinsic
824 // to the list in vmIntrinsics.hpp without implementing it here.
825 #ifndef PRODUCT
826 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
827 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
828 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
829 }
830 #endif
831 Node* slow_ctl = control();
832 set_control(top()); // No fast path intrinsic
833 return slow_ctl;
834 }
835 }
836
837 //------------------------------set_result-------------------------------
838 // Helper function for finishing intrinsics.
839 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
840 record_for_igvn(region);
841 set_control(_gvn.transform(region));
842 set_result( _gvn.transform(value));
843 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
844 }
845
846 RegionNode* LibraryCallKit::create_bailout() {
847 RegionNode* bailout = new RegionNode(1);
848 record_for_igvn(bailout);
849 return bailout;
850 }
851
852 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
853 if (bailout->req() > 1) {
854 bailout = _gvn.transform(bailout)->as_Region();
855 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
856 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
857 C->root()->add_req(halt);
858 }
859 return stopped();
860 }
861
862 //------------------------------generate_guard---------------------------
863 // Helper function for generating guarded fast-slow graph structures.
864 // The given 'test', if true, guards a slow path. If the test fails
865 // then a fast path can be taken. (We generally hope it fails.)
866 // In all cases, GraphKit::control() is updated to the fast path.
867 // The returned value represents the control for the slow path.
868 // The return value is never 'top'; it is either a valid control
869 // or null if it is obvious that the slow path can never be taken.
870 // Also, if region and the slow control are not null, the slow edge
871 // is appended to the region.
872 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
873 if (stopped()) {
874 // Already short circuited.
875 return nullptr;
876 }
877
878 // Build an if node and its projections.
879 // If test is true we take the slow path, which we assume is uncommon.
880 if (_gvn.type(test) == TypeInt::ZERO) {
881 // The slow branch is never taken. No need to build this guard.
882 return nullptr;
883 }
884
885 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
886
887 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
888 if (if_slow == top()) {
889 // The slow branch is never taken. No need to build this guard.
890 return nullptr;
891 }
892
893 if (region != nullptr)
894 region->add_req(if_slow);
895
896 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
897 set_control(if_fast);
898
899 return if_slow;
900 }
901
902 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
903 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
904 }
905 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
906 return generate_guard(test, region, PROB_FAIR);
907 }
908
909 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
910 Node** pos_index, bool with_opaque) {
911 if (stopped())
912 return nullptr; // already stopped
913 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
914 return nullptr; // index is already adequately typed
915 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
916 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
917 if (with_opaque) {
918 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
919 }
920 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
921 if (is_neg != nullptr && pos_index != nullptr) {
922 // Emulate effect of Parse::adjust_map_after_if.
923 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
924 (*pos_index) = _gvn.transform(ccast);
925 }
926 return is_neg;
927 }
928
929 // Make sure that 'position' is a valid limit index, in [0..length].
930 // There are two equivalent plans for checking this:
931 // A. (offset + copyLength) unsigned<= arrayLength
932 // B. offset <= (arrayLength - copyLength)
933 // We require that all of the values above, except for the sum and
934 // difference, are already known to be non-negative.
935 // Plan A is robust in the face of overflow, if offset and copyLength
936 // are both hugely positive.
937 //
938 // Plan B is less direct and intuitive, but it does not overflow at
939 // all, since the difference of two non-negatives is always
940 // representable. Whenever Java methods must perform the equivalent
941 // check they generally use Plan B instead of Plan A.
942 // For the moment we use Plan A.
943 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
944 Node* subseq_length,
945 Node* array_length,
946 RegionNode* region,
947 bool with_opaque) {
948 if (stopped())
949 return nullptr; // already stopped
950 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
951 if (zero_offset && subseq_length->eqv_uncast(array_length))
952 return nullptr; // common case of whole-array copy
953 Node* last = subseq_length;
954 if (!zero_offset) // last += offset
955 last = _gvn.transform(new AddINode(last, offset));
956 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
957 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
958 if (with_opaque) {
959 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
960 }
961 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
962 return is_over;
963 }
964
965 // Emit range checks for the given String.value byte array
966 void LibraryCallKit::generate_string_range_check(Node* array,
967 Node* offset,
968 Node* count,
969 bool char_count,
970 RegionNode* region) {
971 if (stopped()) {
972 return; // already stopped
973 }
974 if (char_count) {
975 // Convert char count to byte count
976 count = _gvn.transform(new LShiftINode(count, intcon(1)));
977 }
978 // Offset and count must not be negative
979 generate_negative_guard(offset, region, nullptr, true);
980 generate_negative_guard(count, region, nullptr, true);
981 // Offset + count must not exceed length of array
982 generate_limit_guard(offset, count, load_array_length(array), region, true);
983 }
984
985 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
986 bool is_immutable) {
987 ciKlass* thread_klass = env()->Thread_klass();
988 const Type* thread_type
989 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
990
991 Node* thread = _gvn.transform(new ThreadLocalNode());
992 Node* p = off_heap_plus_addr(thread, in_bytes(handle_offset));
993 tls_output = thread;
994
995 Node* thread_obj_handle
996 = (is_immutable
997 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
998 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
999 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1000 thread_obj_handle = _gvn.transform(thread_obj_handle);
1001
1002 DecoratorSet decorators = IN_NATIVE;
1003 if (is_immutable) {
1004 decorators |= C2_IMMUTABLE_MEMORY;
1005 }
1006 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1007 }
1008
1009 //--------------------------generate_current_thread--------------------
1010 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1011 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1012 /*is_immutable*/false);
1013 }
1014
1015 //--------------------------generate_virtual_thread--------------------
1016 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1017 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1018 !C->method()->changes_current_thread());
1019 }
1020
1021 //------------------------------make_string_method_node------------------------
1022 // Helper method for String intrinsic functions. This version is called with
1023 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1024 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1025 // containing the lengths of str1 and str2.
1026 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1027 Node* result = nullptr;
1028 switch (opcode) {
1029 case Op_StrIndexOf:
1030 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1031 str1_start, cnt1, str2_start, cnt2, ae);
1032 break;
1033 case Op_StrComp:
1034 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1035 str1_start, cnt1, str2_start, cnt2, ae);
1036 break;
1037 case Op_StrEquals:
1038 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1039 // Use the constant length if there is one because optimized match rule may exist.
1040 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1042 break;
1043 default:
1044 ShouldNotReachHere();
1045 return nullptr;
1046 }
1047
1048 // All these intrinsics have checks.
1049 C->set_has_split_ifs(true); // Has chance for split-if optimization
1050 clear_upper_avx();
1051
1052 return _gvn.transform(result);
1053 }
1054
1055 //------------------------------inline_string_compareTo------------------------
1056 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1057 Node* arg1 = argument(0);
1058 Node* arg2 = argument(1);
1059
1060 arg1 = must_be_not_null(arg1, true);
1061 arg2 = must_be_not_null(arg2, true);
1062
1063 // Get start addr and length of first argument
1064 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1065 Node* arg1_cnt = load_array_length(arg1);
1066
1067 // Get start addr and length of second argument
1068 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1069 Node* arg2_cnt = load_array_length(arg2);
1070
1071 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1072 set_result(result);
1073 return true;
1074 }
1075
1076 //------------------------------inline_string_equals------------------------
1077 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1078 Node* arg1 = argument(0);
1079 Node* arg2 = argument(1);
1080
1081 // paths (plus control) merge
1082 RegionNode* region = new RegionNode(3);
1083 Node* phi = new PhiNode(region, TypeInt::BOOL);
1084
1085 if (!stopped()) {
1086
1087 arg1 = must_be_not_null(arg1, true);
1088 arg2 = must_be_not_null(arg2, true);
1089
1090 // Get start addr and length of first argument
1091 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1092 Node* arg1_cnt = load_array_length(arg1);
1093
1094 // Get start addr and length of second argument
1095 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1096 Node* arg2_cnt = load_array_length(arg2);
1097
1098 // Check for arg1_cnt != arg2_cnt
1099 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1100 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1101 Node* if_ne = generate_slow_guard(bol, nullptr);
1102 if (if_ne != nullptr) {
1103 phi->init_req(2, intcon(0));
1104 region->init_req(2, if_ne);
1105 }
1106
1107 // Check for count == 0 is done by assembler code for StrEquals.
1108
1109 if (!stopped()) {
1110 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1111 phi->init_req(1, equals);
1112 region->init_req(1, control());
1113 }
1114 }
1115
1116 // post merge
1117 set_control(_gvn.transform(region));
1118 record_for_igvn(region);
1119
1120 set_result(_gvn.transform(phi));
1121 return true;
1122 }
1123
1124 //------------------------------inline_array_equals----------------------------
1125 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1126 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1127 Node* arg1 = argument(0);
1128 Node* arg2 = argument(1);
1129
1130 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1131 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1132 clear_upper_avx();
1133
1134 return true;
1135 }
1136
1137
1138 //------------------------------inline_countPositives------------------------------
1139 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1140 bool LibraryCallKit::inline_countPositives() {
1141 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1142 // no receiver since it is static method
1143 Node* ba = argument(0);
1144 Node* offset = argument(1);
1145 Node* len = argument(2);
1146
1147 ba = must_be_not_null(ba, true);
1148 RegionNode* bailout = create_bailout();
1149 generate_string_range_check(ba, offset, len, false, bailout);
1150 if (check_bailout(bailout)) {
1151 return true;
1152 }
1153
1154 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1155 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1156 set_result(_gvn.transform(result));
1157 clear_upper_avx();
1158 return true;
1159 }
1160
1161 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1162 Node* index = argument(0);
1163 Node* length = bt == T_INT ? argument(1) : argument(2);
1164 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1165 return false;
1166 }
1167
1168 // check that length is positive
1169 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1170 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1171
1172 {
1173 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1174 uncommon_trap(Deoptimization::Reason_intrinsic,
1175 Deoptimization::Action_make_not_entrant);
1176 }
1177
1178 if (stopped()) {
1179 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1180 return true;
1181 }
1182
1183 // length is now known positive, add a cast node to make this explicit
1184 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1185 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1186 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1187 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1188 casted_length = _gvn.transform(casted_length);
1189 replace_in_map(length, casted_length);
1190 length = casted_length;
1191
1192 // Use an unsigned comparison for the range check itself
1193 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1194 BoolTest::mask btest = BoolTest::lt;
1195 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1196 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1197 _gvn.set_type(rc, rc->Value(&_gvn));
1198 if (!rc_bool->is_Con()) {
1199 record_for_igvn(rc);
1200 }
1201 set_control(_gvn.transform(new IfTrueNode(rc)));
1202 {
1203 PreserveJVMState pjvms(this);
1204 set_control(_gvn.transform(new IfFalseNode(rc)));
1205 uncommon_trap(Deoptimization::Reason_range_check,
1206 Deoptimization::Action_make_not_entrant);
1207 }
1208
1209 if (stopped()) {
1210 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1211 return true;
1212 }
1213
1214 // index is now known to be >= 0 and < length, cast it
1215 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1216 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1217 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1218 result = _gvn.transform(result);
1219 set_result(result);
1220 replace_in_map(index, result);
1221 return true;
1222 }
1223
1224 //------------------------------inline_string_indexOf------------------------
1225 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1226 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1227 return false;
1228 }
1229 Node* src = argument(0);
1230 Node* tgt = argument(1);
1231
1232 // Make the merge point
1233 RegionNode* result_rgn = new RegionNode(4);
1234 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1235
1236 src = must_be_not_null(src, true);
1237 tgt = must_be_not_null(tgt, true);
1238
1239 // Get start addr and length of source string
1240 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1241 Node* src_count = load_array_length(src);
1242
1243 // Get start addr and length of substring
1244 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1245 Node* tgt_count = load_array_length(tgt);
1246
1247 Node* result = nullptr;
1248 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1249
1250 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1251 // Divide src size by 2 if String is UTF16 encoded
1252 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1253 }
1254 if (ae == StrIntrinsicNode::UU) {
1255 // Divide substring size by 2 if String is UTF16 encoded
1256 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1257 }
1258
1259 if (call_opt_stub) {
1260 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1261 StubRoutines::_string_indexof_array[ae],
1262 "stringIndexOf", TypePtr::BOTTOM, src_start,
1263 src_count, tgt_start, tgt_count);
1264 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1265 } else {
1266 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1267 result_rgn, result_phi, ae);
1268 }
1269 if (result != nullptr) {
1270 result_phi->init_req(3, result);
1271 result_rgn->init_req(3, control());
1272 }
1273 set_control(_gvn.transform(result_rgn));
1274 record_for_igvn(result_rgn);
1275 set_result(_gvn.transform(result_phi));
1276
1277 return true;
1278 }
1279
1280 //-----------------------------inline_string_indexOfI-----------------------
1281 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1282 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1283 return false;
1284 }
1285
1286 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1287 Node* src = argument(0); // byte[]
1288 Node* src_count = argument(1); // char count
1289 Node* tgt = argument(2); // byte[]
1290 Node* tgt_count = argument(3); // char count
1291 Node* from_index = argument(4); // char index
1292
1293 src = must_be_not_null(src, true);
1294 tgt = must_be_not_null(tgt, true);
1295
1296 // Multiply byte array index by 2 if String is UTF16 encoded
1297 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1298 src_count = _gvn.transform(new SubINode(src_count, from_index));
1299 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1300 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1301
1302 // Range checks
1303 RegionNode* bailout = create_bailout();
1304 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1305 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1306 if (check_bailout(bailout)) {
1307 return true;
1308 }
1309
1310 RegionNode* region = new RegionNode(5);
1311 Node* phi = new PhiNode(region, TypeInt::INT);
1312 Node* result = nullptr;
1313
1314 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1315
1316 if (call_opt_stub) {
1317 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1318 StubRoutines::_string_indexof_array[ae],
1319 "stringIndexOf", TypePtr::BOTTOM, src_start,
1320 src_count, tgt_start, tgt_count);
1321 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1322 } else {
1323 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1324 region, phi, ae);
1325 }
1326 if (result != nullptr) {
1327 // The result is index relative to from_index if substring was found, -1 otherwise.
1328 // Generate code which will fold into cmove.
1329 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1330 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1331
1332 Node* if_lt = generate_slow_guard(bol, nullptr);
1333 if (if_lt != nullptr) {
1334 // result == -1
1335 phi->init_req(3, result);
1336 region->init_req(3, if_lt);
1337 }
1338 if (!stopped()) {
1339 result = _gvn.transform(new AddINode(result, from_index));
1340 phi->init_req(4, result);
1341 region->init_req(4, control());
1342 }
1343 }
1344
1345 set_control(_gvn.transform(region));
1346 record_for_igvn(region);
1347 set_result(_gvn.transform(phi));
1348 clear_upper_avx();
1349
1350 return true;
1351 }
1352
1353 // Create StrIndexOfNode with fast path checks
1354 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1355 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1356 // Check for substr count > string count
1357 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1359 Node* if_gt = generate_slow_guard(bol, nullptr);
1360 if (if_gt != nullptr) {
1361 phi->init_req(1, intcon(-1));
1362 region->init_req(1, if_gt);
1363 }
1364 if (!stopped()) {
1365 // Check for substr count == 0
1366 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1367 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1368 Node* if_zero = generate_slow_guard(bol, nullptr);
1369 if (if_zero != nullptr) {
1370 phi->init_req(2, intcon(0));
1371 region->init_req(2, if_zero);
1372 }
1373 }
1374 if (!stopped()) {
1375 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1376 }
1377 return nullptr;
1378 }
1379
1380 //-----------------------------inline_string_indexOfChar-----------------------
1381 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1382 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1383 return false;
1384 }
1385 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1386 return false;
1387 }
1388 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1389 Node* src = argument(0); // byte[]
1390 Node* int_ch = argument(1);
1391 Node* from_index = argument(2);
1392 Node* max = argument(3);
1393
1394 src = must_be_not_null(src, true);
1395
1396 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1397 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1398 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1399
1400 // Range checks
1401 RegionNode* bailout = create_bailout();
1402 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1403 if (check_bailout(bailout)) {
1404 return true;
1405 }
1406
1407 // Check for int_ch >= 0
1408 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1409 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1410 {
1411 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1412 uncommon_trap(Deoptimization::Reason_intrinsic,
1413 Deoptimization::Action_maybe_recompile);
1414 }
1415 if (stopped()) {
1416 return true;
1417 }
1418
1419 RegionNode* region = new RegionNode(3);
1420 Node* phi = new PhiNode(region, TypeInt::INT);
1421
1422 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1423 C->set_has_split_ifs(true); // Has chance for split-if optimization
1424 _gvn.transform(result);
1425
1426 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1427 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1428
1429 Node* if_lt = generate_slow_guard(bol, nullptr);
1430 if (if_lt != nullptr) {
1431 // result == -1
1432 phi->init_req(2, result);
1433 region->init_req(2, if_lt);
1434 }
1435 if (!stopped()) {
1436 result = _gvn.transform(new AddINode(result, from_index));
1437 phi->init_req(1, result);
1438 region->init_req(1, control());
1439 }
1440 set_control(_gvn.transform(region));
1441 record_for_igvn(region);
1442 set_result(_gvn.transform(phi));
1443 clear_upper_avx();
1444
1445 return true;
1446 }
1447 //---------------------------inline_string_copy---------------------
1448 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1449 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1450 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1451 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1452 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1453 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1454 bool LibraryCallKit::inline_string_copy(bool compress) {
1455 int nargs = 5; // 2 oops, 3 ints
1456 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1457
1458 Node* src = argument(0);
1459 Node* src_offset = argument(1);
1460 Node* dst = argument(2);
1461 Node* dst_offset = argument(3);
1462 Node* length = argument(4);
1463
1464 // Check for allocation before we add nodes that would confuse
1465 // tightly_coupled_allocation()
1466 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1467
1468 // Figure out the size and type of the elements we will be copying.
1469 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1470 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1471 if (src_type == nullptr || dst_type == nullptr) {
1472 return false;
1473 }
1474 BasicType src_elem = src_type->elem()->array_element_basic_type();
1475 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1476 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1477 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1478 "Unsupported array types for inline_string_copy");
1479
1480 src = must_be_not_null(src, true);
1481 dst = must_be_not_null(dst, true);
1482
1483 // Convert char[] offsets to byte[] offsets
1484 bool convert_src = (compress && src_elem == T_BYTE);
1485 bool convert_dst = (!compress && dst_elem == T_BYTE);
1486 if (convert_src) {
1487 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1488 } else if (convert_dst) {
1489 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1490 }
1491
1492 // Range checks
1493 RegionNode* bailout = create_bailout();
1494 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1495 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1496 if (check_bailout(bailout)) {
1497 return true;
1498 }
1499
1500 Node* src_start = array_element_address(src, src_offset, src_elem);
1501 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1502 // 'src_start' points to src array + scaled offset
1503 // 'dst_start' points to dst array + scaled offset
1504 Node* count = nullptr;
1505 if (compress) {
1506 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1507 } else {
1508 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1509 }
1510
1511 if (alloc != nullptr) {
1512 if (alloc->maybe_set_complete(&_gvn)) {
1513 // "You break it, you buy it."
1514 InitializeNode* init = alloc->initialization();
1515 assert(init->is_complete(), "we just did this");
1516 init->set_complete_with_arraycopy();
1517 assert(dst->is_CheckCastPP(), "sanity");
1518 assert(dst->in(0)->in(0) == init, "dest pinned");
1519 }
1520 // Do not let stores that initialize this object be reordered with
1521 // a subsequent store that would make this object accessible by
1522 // other threads.
1523 // Record what AllocateNode this StoreStore protects so that
1524 // escape analysis can go from the MemBarStoreStoreNode to the
1525 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1526 // based on the escape status of the AllocateNode.
1527 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1528 }
1529 if (compress) {
1530 set_result(_gvn.transform(count));
1531 }
1532 clear_upper_avx();
1533
1534 return true;
1535 }
1536
1537 #ifdef _LP64
1538 #define XTOP ,top() /*additional argument*/
1539 #else //_LP64
1540 #define XTOP /*no additional argument*/
1541 #endif //_LP64
1542
1543 //------------------------inline_string_toBytesU--------------------------
1544 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1545 bool LibraryCallKit::inline_string_toBytesU() {
1546 // Get the arguments.
1547 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1548 Node* value = argument(0);
1549 Node* offset = argument(1);
1550 Node* length = argument(2);
1551
1552 Node* newcopy = nullptr;
1553
1554 // Set the original stack and the reexecute bit for the interpreter to reexecute
1555 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1556 { PreserveReexecuteState preexecs(this);
1557 jvms()->set_should_reexecute(true);
1558
1559 value = must_be_not_null(value, true);
1560 RegionNode* bailout = create_bailout();
1561 generate_negative_guard(offset, bailout, nullptr, true);
1562 generate_negative_guard(length, bailout, nullptr, true);
1563 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1564 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1565 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1566 if (check_bailout(bailout)) {
1567 return true;
1568 }
1569
1570 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1571 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1572 newcopy = new_array(klass_node, size, 0); // no arguments to push
1573 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1574 guarantee(alloc != nullptr, "created above");
1575
1576 // Calculate starting addresses.
1577 Node* src_start = array_element_address(value, offset, T_CHAR);
1578 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1579
1580 // Check if dst array address is aligned to HeapWordSize
1581 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1582 // If true, then check if src array address is aligned to HeapWordSize
1583 if (aligned) {
1584 const TypeInt* toffset = gvn().type(offset)->is_int();
1585 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1586 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1587 }
1588
1589 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1590 const char* copyfunc_name = "arraycopy";
1591 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1592 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1593 OptoRuntime::fast_arraycopy_Type(),
1594 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1595 src_start, dst_start, ConvI2X(length) XTOP);
1596 // Do not let reads from the cloned object float above the arraycopy.
1597 if (alloc->maybe_set_complete(&_gvn)) {
1598 // "You break it, you buy it."
1599 InitializeNode* init = alloc->initialization();
1600 assert(init->is_complete(), "we just did this");
1601 init->set_complete_with_arraycopy();
1602 assert(newcopy->is_CheckCastPP(), "sanity");
1603 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1604 }
1605 // Do not let stores that initialize this object be reordered with
1606 // a subsequent store that would make this object accessible by
1607 // other threads.
1608 // Record what AllocateNode this StoreStore protects so that
1609 // escape analysis can go from the MemBarStoreStoreNode to the
1610 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1611 // based on the escape status of the AllocateNode.
1612 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1613 } // original reexecute is set back here
1614
1615 C->set_has_split_ifs(true); // Has chance for split-if optimization
1616 if (!stopped()) {
1617 set_result(newcopy);
1618 }
1619 clear_upper_avx();
1620
1621 return true;
1622 }
1623
1624 //------------------------inline_string_getCharsU--------------------------
1625 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1626 bool LibraryCallKit::inline_string_getCharsU() {
1627 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1628 // Get the arguments.
1629 Node* src = argument(0);
1630 Node* src_begin = argument(1);
1631 Node* src_end = argument(2); // exclusive offset (i < src_end)
1632 Node* dst = argument(3);
1633 Node* dst_begin = argument(4);
1634
1635 // Check for allocation before we add nodes that would confuse
1636 // tightly_coupled_allocation()
1637 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1638
1639 // Check if a null path was taken unconditionally.
1640 src = must_be_not_null(src, true);
1641 dst = must_be_not_null(dst, true);
1642 if (stopped()) {
1643 return true;
1644 }
1645
1646 // Get length and convert char[] offset to byte[] offset
1647 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1648 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1649
1650 // Range checks
1651 RegionNode* bailout = create_bailout();
1652 generate_string_range_check(src, src_begin, length, true, bailout);
1653 generate_string_range_check(dst, dst_begin, length, false, bailout);
1654 if (check_bailout(bailout)) {
1655 return true;
1656 }
1657
1658 // Calculate starting addresses.
1659 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1660 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1661
1662 // Check if array addresses are aligned to HeapWordSize
1663 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1664 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1665 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1666 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1667
1668 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1669 const char* copyfunc_name = "arraycopy";
1670 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1671 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1672 OptoRuntime::fast_arraycopy_Type(),
1673 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1674 src_start, dst_start, ConvI2X(length) XTOP);
1675 // Do not let reads from the cloned object float above the arraycopy.
1676 if (alloc != nullptr) {
1677 if (alloc->maybe_set_complete(&_gvn)) {
1678 // "You break it, you buy it."
1679 InitializeNode* init = alloc->initialization();
1680 assert(init->is_complete(), "we just did this");
1681 init->set_complete_with_arraycopy();
1682 assert(dst->is_CheckCastPP(), "sanity");
1683 assert(dst->in(0)->in(0) == init, "dest pinned");
1684 }
1685 // Do not let stores that initialize this object be reordered with
1686 // a subsequent store that would make this object accessible by
1687 // other threads.
1688 // Record what AllocateNode this StoreStore protects so that
1689 // escape analysis can go from the MemBarStoreStoreNode to the
1690 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1691 // based on the escape status of the AllocateNode.
1692 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1693 } else {
1694 insert_mem_bar(Op_MemBarCPUOrder);
1695 }
1696
1697 C->set_has_split_ifs(true); // Has chance for split-if optimization
1698 return true;
1699 }
1700
1701 //----------------------inline_string_char_access----------------------------
1702 // Store/Load char to/from byte[] array.
1703 // static void StringUTF16.putChar(byte[] val, int index, int c)
1704 // static char StringUTF16.getChar(byte[] val, int index)
1705 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1706 Node* ch;
1707 if (is_store) {
1708 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1709 ch = argument(2);
1710 } else {
1711 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1712 ch = nullptr;
1713 }
1714 Node* value = argument(0);
1715 Node* index = argument(1);
1716
1717 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1718 // correctly requires matched array shapes.
1719 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1720 "sanity: byte[] and char[] bases agree");
1721 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1722 "sanity: byte[] and char[] scales agree");
1723
1724 // Bail when getChar over constants is requested: constant folding would
1725 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1726 // Java method would constant fold nicely instead.
1727 if (!is_store && value->is_Con() && index->is_Con()) {
1728 return false;
1729 }
1730
1731 // Save state and restore on bailout
1732 SavedState old_state(this);
1733
1734 value = must_be_not_null(value, true);
1735
1736 Node* adr = array_element_address(value, index, T_CHAR);
1737 if (adr->is_top()) {
1738 return false;
1739 }
1740 old_state.discard();
1741 if (is_store) {
1742 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1743 } else {
1744 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);
1745 set_result(ch);
1746 }
1747 return true;
1748 }
1749
1750
1751 //------------------------------inline_math-----------------------------------
1752 // public static double Math.abs(double)
1753 // public static double Math.sqrt(double)
1754 // public static double Math.log(double)
1755 // public static double Math.log10(double)
1756 // public static double Math.round(double)
1757 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1758 Node* arg = argument(0);
1759 Node* n = nullptr;
1760 switch (id) {
1761 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1762 case vmIntrinsics::_dsqrt:
1763 case vmIntrinsics::_dsqrt_strict:
1764 n = new SqrtDNode(C, control(), arg); break;
1765 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1766 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1767 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1768 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1769 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1770 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1771 default: fatal_unexpected_iid(id); break;
1772 }
1773 set_result(_gvn.transform(n));
1774 return true;
1775 }
1776
1777 //------------------------------inline_math-----------------------------------
1778 // public static float Math.abs(float)
1779 // public static int Math.abs(int)
1780 // public static long Math.abs(long)
1781 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1782 Node* arg = argument(0);
1783 Node* n = nullptr;
1784 switch (id) {
1785 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1786 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1787 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1788 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1789 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1790 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1791 default: fatal_unexpected_iid(id); break;
1792 }
1793 set_result(_gvn.transform(n));
1794 return true;
1795 }
1796
1797 //------------------------------runtime_math-----------------------------
1798 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1799 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1800 "must be (DD)D or (D)D type");
1801
1802 // Inputs
1803 Node* a = argument(0);
1804 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1805
1806 const TypePtr* no_memory_effects = nullptr;
1807 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1808 no_memory_effects,
1809 a, top(), b, b ? top() : nullptr);
1810 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1811 #ifdef ASSERT
1812 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1813 assert(value_top == top(), "second value must be top");
1814 #endif
1815
1816 set_result(value);
1817 return true;
1818 }
1819
1820 //------------------------------inline_math_pow-----------------------------
1821 bool LibraryCallKit::inline_math_pow() {
1822 Node* base = argument(0);
1823 Node* exp = argument(2);
1824
1825 CallNode* pow = new PowDNode(C, base, exp);
1826 set_predefined_input_for_runtime_call(pow);
1827 pow = _gvn.transform(pow)->as_CallLeafPure();
1828 set_predefined_output_for_runtime_call(pow);
1829 Node* result = _gvn.transform(new ProjNode(pow, TypeFunc::Parms + 0));
1830 record_for_igvn(pow);
1831 set_result(result);
1832 return true;
1833 }
1834
1835 //------------------------------inline_math_native-----------------------------
1836 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1837 switch (id) {
1838 case vmIntrinsics::_dsin:
1839 return StubRoutines::dsin() != nullptr ?
1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1841 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1842 case vmIntrinsics::_dcos:
1843 return StubRoutines::dcos() != nullptr ?
1844 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1845 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1846 case vmIntrinsics::_dtan:
1847 return StubRoutines::dtan() != nullptr ?
1848 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1849 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1850 case vmIntrinsics::_dsinh:
1851 return StubRoutines::dsinh() != nullptr ?
1852 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1853 case vmIntrinsics::_dtanh:
1854 return StubRoutines::dtanh() != nullptr ?
1855 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1856 case vmIntrinsics::_dcbrt:
1857 return StubRoutines::dcbrt() != nullptr ?
1858 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1859 case vmIntrinsics::_dexp:
1860 return StubRoutines::dexp() != nullptr ?
1861 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1862 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1863 case vmIntrinsics::_dlog:
1864 return StubRoutines::dlog() != nullptr ?
1865 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1866 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1867 case vmIntrinsics::_dlog10:
1868 return StubRoutines::dlog10() != nullptr ?
1869 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1870 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1871
1872 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1873 case vmIntrinsics::_ceil:
1874 case vmIntrinsics::_floor:
1875 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1876
1877 case vmIntrinsics::_dsqrt:
1878 case vmIntrinsics::_dsqrt_strict:
1879 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1880 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1881 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1882 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1883 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1884
1885 case vmIntrinsics::_dpow: return inline_math_pow();
1886 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1887 case vmIntrinsics::_fcopySign: return inline_math(id);
1888 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1889 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1890 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1891
1892 // These intrinsics are not yet correctly implemented
1893 case vmIntrinsics::_datan2:
1894 return false;
1895
1896 default:
1897 fatal_unexpected_iid(id);
1898 return false;
1899 }
1900 }
1901
1902 //----------------------------inline_notify-----------------------------------*
1903 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1904 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1905 address func;
1906 if (id == vmIntrinsics::_notify) {
1907 func = OptoRuntime::monitor_notify_Java();
1908 } else {
1909 func = OptoRuntime::monitor_notifyAll_Java();
1910 }
1911 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1912 make_slow_call_ex(call, env()->Throwable_klass(), false);
1913 return true;
1914 }
1915
1916
1917 //----------------------------inline_min_max-----------------------------------
1918 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1919 Node* a = nullptr;
1920 Node* b = nullptr;
1921 Node* n = nullptr;
1922 switch (id) {
1923 case vmIntrinsics::_min:
1924 case vmIntrinsics::_max:
1925 case vmIntrinsics::_minF:
1926 case vmIntrinsics::_maxF:
1927 case vmIntrinsics::_minF_strict:
1928 case vmIntrinsics::_maxF_strict:
1929 case vmIntrinsics::_min_strict:
1930 case vmIntrinsics::_max_strict:
1931 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1932 a = argument(0);
1933 b = argument(1);
1934 break;
1935 case vmIntrinsics::_minD:
1936 case vmIntrinsics::_maxD:
1937 case vmIntrinsics::_minD_strict:
1938 case vmIntrinsics::_maxD_strict:
1939 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1940 a = argument(0);
1941 b = argument(2);
1942 break;
1943 case vmIntrinsics::_minL:
1944 case vmIntrinsics::_maxL:
1945 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1946 a = argument(0);
1947 b = argument(2);
1948 break;
1949 default:
1950 fatal_unexpected_iid(id);
1951 break;
1952 }
1953
1954 switch (id) {
1955 case vmIntrinsics::_min:
1956 case vmIntrinsics::_min_strict:
1957 n = new MinINode(a, b);
1958 break;
1959 case vmIntrinsics::_max:
1960 case vmIntrinsics::_max_strict:
1961 n = new MaxINode(a, b);
1962 break;
1963 case vmIntrinsics::_minF:
1964 case vmIntrinsics::_minF_strict:
1965 n = new MinFNode(a, b);
1966 break;
1967 case vmIntrinsics::_maxF:
1968 case vmIntrinsics::_maxF_strict:
1969 n = new MaxFNode(a, b);
1970 break;
1971 case vmIntrinsics::_minD:
1972 case vmIntrinsics::_minD_strict:
1973 n = new MinDNode(a, b);
1974 break;
1975 case vmIntrinsics::_maxD:
1976 case vmIntrinsics::_maxD_strict:
1977 n = new MaxDNode(a, b);
1978 break;
1979 case vmIntrinsics::_minL:
1980 n = new MinLNode(_gvn.C, a, b);
1981 break;
1982 case vmIntrinsics::_maxL:
1983 n = new MaxLNode(_gvn.C, a, b);
1984 break;
1985 default:
1986 fatal_unexpected_iid(id);
1987 break;
1988 }
1989
1990 set_result(_gvn.transform(n));
1991 return true;
1992 }
1993
1994 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
1995 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
1996 env()->ArithmeticException_instance())) {
1997 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
1998 // so let's bail out intrinsic rather than risking deopting again.
1999 return false;
2000 }
2001
2002 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2003 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2004 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2005 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2006
2007 {
2008 PreserveJVMState pjvms(this);
2009 PreserveReexecuteState preexecs(this);
2010 jvms()->set_should_reexecute(true);
2011
2012 set_control(slow_path);
2013 set_i_o(i_o());
2014
2015 builtin_throw(Deoptimization::Reason_intrinsic,
2016 env()->ArithmeticException_instance(),
2017 /*allow_too_many_traps*/ false);
2018 }
2019
2020 set_control(fast_path);
2021 set_result(math);
2022 return true;
2023 }
2024
2025 template <typename OverflowOp>
2026 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2027 typedef typename OverflowOp::MathOp MathOp;
2028
2029 MathOp* mathOp = new MathOp(arg1, arg2);
2030 Node* operation = _gvn.transform( mathOp );
2031 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2032 return inline_math_mathExact(operation, ofcheck);
2033 }
2034
2035 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2036 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2037 }
2038
2039 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2040 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2041 }
2042
2043 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2044 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2045 }
2046
2047 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2048 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2049 }
2050
2051 bool LibraryCallKit::inline_math_negateExactI() {
2052 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2053 }
2054
2055 bool LibraryCallKit::inline_math_negateExactL() {
2056 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2057 }
2058
2059 bool LibraryCallKit::inline_math_multiplyExactI() {
2060 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2061 }
2062
2063 bool LibraryCallKit::inline_math_multiplyExactL() {
2064 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2065 }
2066
2067 bool LibraryCallKit::inline_math_multiplyHigh() {
2068 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2069 return true;
2070 }
2071
2072 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2073 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2074 return true;
2075 }
2076
2077 inline int
2078 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2079 const TypePtr* base_type = TypePtr::NULL_PTR;
2080 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2081 if (base_type == nullptr) {
2082 // Unknown type.
2083 return Type::AnyPtr;
2084 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2085 // Since this is a null+long form, we have to switch to a rawptr.
2086 base = _gvn.transform(new CastX2PNode(offset));
2087 offset = MakeConX(0);
2088 return Type::RawPtr;
2089 } else if (base_type->base() == Type::RawPtr) {
2090 return Type::RawPtr;
2091 } else if (base_type->isa_oopptr()) {
2092 // Base is never null => always a heap address.
2093 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2094 return Type::OopPtr;
2095 }
2096 // Offset is small => always a heap address.
2097 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2098 if (offset_type != nullptr &&
2099 base_type->offset() == 0 && // (should always be?)
2100 offset_type->_lo >= 0 &&
2101 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2102 return Type::OopPtr;
2103 } else if (type == T_OBJECT) {
2104 // off heap access to an oop doesn't make any sense. Has to be on
2105 // heap.
2106 return Type::OopPtr;
2107 }
2108 // Otherwise, it might either be oop+off or null+addr.
2109 return Type::AnyPtr;
2110 } else {
2111 // No information:
2112 return Type::AnyPtr;
2113 }
2114 }
2115
2116 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2117 Node* uncasted_base = base;
2118 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2119 if (kind == Type::RawPtr) {
2120 return off_heap_plus_addr(uncasted_base, offset);
2121 } else if (kind == Type::AnyPtr) {
2122 assert(base == uncasted_base, "unexpected base change");
2123 if (can_cast) {
2124 if (!_gvn.type(base)->speculative_maybe_null() &&
2125 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2126 // According to profiling, this access is always on
2127 // heap. Casting the base to not null and thus avoiding membars
2128 // around the access should allow better optimizations
2129 Node* null_ctl = top();
2130 base = null_check_oop(base, &null_ctl, true, true, true);
2131 assert(null_ctl->is_top(), "no null control here");
2132 return basic_plus_adr(base, offset);
2133 } else if (_gvn.type(base)->speculative_always_null() &&
2134 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2135 // According to profiling, this access is always off
2136 // heap.
2137 base = null_assert(base);
2138 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2139 offset = MakeConX(0);
2140 return off_heap_plus_addr(raw_base, offset);
2141 }
2142 }
2143 // We don't know if it's an on heap or off heap access. Fall back
2144 // to raw memory access.
2145 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2146 return off_heap_plus_addr(raw, offset);
2147 } else {
2148 assert(base == uncasted_base, "unexpected base change");
2149 // We know it's an on heap access so base can't be null
2150 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2151 base = must_be_not_null(base, true);
2152 }
2153 return basic_plus_adr(base, offset);
2154 }
2155 }
2156
2157 //--------------------------inline_number_methods-----------------------------
2158 // inline int Integer.numberOfLeadingZeros(int)
2159 // inline int Long.numberOfLeadingZeros(long)
2160 //
2161 // inline int Integer.numberOfTrailingZeros(int)
2162 // inline int Long.numberOfTrailingZeros(long)
2163 //
2164 // inline int Integer.bitCount(int)
2165 // inline int Long.bitCount(long)
2166 //
2167 // inline char Character.reverseBytes(char)
2168 // inline short Short.reverseBytes(short)
2169 // inline int Integer.reverseBytes(int)
2170 // inline long Long.reverseBytes(long)
2171 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2172 Node* arg = argument(0);
2173 Node* n = nullptr;
2174 switch (id) {
2175 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2176 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2177 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2178 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2179 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2180 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2181 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2182 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2183 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2184 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2185 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2186 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2187 default: fatal_unexpected_iid(id); break;
2188 }
2189 set_result(_gvn.transform(n));
2190 return true;
2191 }
2192
2193 //--------------------------inline_bitshuffle_methods-----------------------------
2194 // inline int Integer.compress(int, int)
2195 // inline int Integer.expand(int, int)
2196 // inline long Long.compress(long, long)
2197 // inline long Long.expand(long, long)
2198 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2199 Node* n = nullptr;
2200 switch (id) {
2201 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2202 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2203 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2204 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2205 default: fatal_unexpected_iid(id); break;
2206 }
2207 set_result(_gvn.transform(n));
2208 return true;
2209 }
2210
2211 //--------------------------inline_number_methods-----------------------------
2212 // inline int Integer.compareUnsigned(int, int)
2213 // inline int Long.compareUnsigned(long, long)
2214 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2215 Node* arg1 = argument(0);
2216 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2217 Node* n = nullptr;
2218 switch (id) {
2219 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2220 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2221 default: fatal_unexpected_iid(id); break;
2222 }
2223 set_result(_gvn.transform(n));
2224 return true;
2225 }
2226
2227 //--------------------------inline_unsigned_divmod_methods-----------------------------
2228 // inline int Integer.divideUnsigned(int, int)
2229 // inline int Integer.remainderUnsigned(int, int)
2230 // inline long Long.divideUnsigned(long, long)
2231 // inline long Long.remainderUnsigned(long, long)
2232 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2233 Node* n = nullptr;
2234 switch (id) {
2235 case vmIntrinsics::_divideUnsigned_i: {
2236 zero_check_int(argument(1));
2237 // Compile-time detect of null-exception
2238 if (stopped()) {
2239 return true; // keep the graph constructed so far
2240 }
2241 n = new UDivINode(control(), argument(0), argument(1));
2242 break;
2243 }
2244 case vmIntrinsics::_divideUnsigned_l: {
2245 zero_check_long(argument(2));
2246 // Compile-time detect of null-exception
2247 if (stopped()) {
2248 return true; // keep the graph constructed so far
2249 }
2250 n = new UDivLNode(control(), argument(0), argument(2));
2251 break;
2252 }
2253 case vmIntrinsics::_remainderUnsigned_i: {
2254 zero_check_int(argument(1));
2255 // Compile-time detect of null-exception
2256 if (stopped()) {
2257 return true; // keep the graph constructed so far
2258 }
2259 n = new UModINode(control(), argument(0), argument(1));
2260 break;
2261 }
2262 case vmIntrinsics::_remainderUnsigned_l: {
2263 zero_check_long(argument(2));
2264 // Compile-time detect of null-exception
2265 if (stopped()) {
2266 return true; // keep the graph constructed so far
2267 }
2268 n = new UModLNode(control(), argument(0), argument(2));
2269 break;
2270 }
2271 default: fatal_unexpected_iid(id); break;
2272 }
2273 set_result(_gvn.transform(n));
2274 return true;
2275 }
2276
2277 //----------------------------inline_unsafe_access----------------------------
2278
2279 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2280 // Attempt to infer a sharper value type from the offset and base type.
2281 ciKlass* sharpened_klass = nullptr;
2282
2283 // See if it is an instance field, with an object type.
2284 if (alias_type->field() != nullptr) {
2285 if (alias_type->field()->type()->is_klass()) {
2286 sharpened_klass = alias_type->field()->type()->as_klass();
2287 }
2288 }
2289
2290 const TypeOopPtr* result = nullptr;
2291 // See if it is a narrow oop array.
2292 if (adr_type->isa_aryptr()) {
2293 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2294 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2295 if (elem_type != nullptr && elem_type->is_loaded()) {
2296 // Sharpen the value type.
2297 result = elem_type;
2298 }
2299 }
2300 }
2301
2302 // The sharpened class might be unloaded if there is no class loader
2303 // contraint in place.
2304 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2305 // Sharpen the value type.
2306 result = TypeOopPtr::make_from_klass(sharpened_klass);
2307 }
2308 if (result != nullptr) {
2309 #ifndef PRODUCT
2310 if (C->print_intrinsics() || C->print_inlining()) {
2311 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2312 tty->print(" sharpened value: "); result->dump(); tty->cr();
2313 }
2314 #endif
2315 }
2316 return result;
2317 }
2318
2319 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2320 switch (kind) {
2321 case Relaxed:
2322 return MO_UNORDERED;
2323 case Opaque:
2324 return MO_RELAXED;
2325 case Acquire:
2326 return MO_ACQUIRE;
2327 case Release:
2328 return MO_RELEASE;
2329 case Volatile:
2330 return MO_SEQ_CST;
2331 default:
2332 ShouldNotReachHere();
2333 return 0;
2334 }
2335 }
2336
2337 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2338 if (callee()->is_static()) return false; // caller must have the capability!
2339 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2340 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2341 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2342 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2343
2344 if (is_reference_type(type)) {
2345 decorators |= ON_UNKNOWN_OOP_REF;
2346 }
2347
2348 if (unaligned) {
2349 decorators |= C2_UNALIGNED;
2350 }
2351
2352 #ifndef PRODUCT
2353 {
2354 ResourceMark rm;
2355 // Check the signatures.
2356 ciSignature* sig = callee()->signature();
2357 #ifdef ASSERT
2358 if (!is_store) {
2359 // Object getReference(Object base, int/long offset), etc.
2360 BasicType rtype = sig->return_type()->basic_type();
2361 assert(rtype == type, "getter must return the expected value");
2362 assert(sig->count() == 2, "oop getter has 2 arguments");
2363 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2364 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2365 } else {
2366 // void putReference(Object base, int/long offset, Object x), etc.
2367 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2368 assert(sig->count() == 3, "oop putter has 3 arguments");
2369 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2370 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2371 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2372 assert(vtype == type, "putter must accept the expected value");
2373 }
2374 #endif // ASSERT
2375 }
2376 #endif //PRODUCT
2377
2378 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2379
2380 Node* receiver = argument(0); // type: oop
2381
2382 // Build address expression.
2383 Node* heap_base_oop = top();
2384
2385 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2386 Node* base = argument(1); // type: oop
2387 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2388 Node* offset = argument(2); // type: long
2389 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2390 // to be plain byte offsets, which are also the same as those accepted
2391 // by oopDesc::field_addr.
2392 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2393 "fieldOffset must be byte-scaled");
2394 // 32-bit machines ignore the high half!
2395 offset = ConvL2X(offset);
2396
2397 // Save state and restore on bailout
2398 SavedState old_state(this);
2399
2400 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2401 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2402
2403 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2404 if (type != T_OBJECT) {
2405 decorators |= IN_NATIVE; // off-heap primitive access
2406 } else {
2407 return false; // off-heap oop accesses are not supported
2408 }
2409 } else {
2410 heap_base_oop = base; // on-heap or mixed access
2411 }
2412
2413 // Can base be null? Otherwise, always on-heap access.
2414 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2415
2416 if (!can_access_non_heap) {
2417 decorators |= IN_HEAP;
2418 }
2419
2420 Node* val = is_store ? argument(4) : nullptr;
2421
2422 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2423 if (adr_type == TypePtr::NULL_PTR) {
2424 return false; // off-heap access with zero address
2425 }
2426
2427 // Try to categorize the address.
2428 Compile::AliasType* alias_type = C->alias_type(adr_type);
2429 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2430
2431 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2432 alias_type->adr_type() == TypeAryPtr::RANGE) {
2433 return false; // not supported
2434 }
2435
2436 bool mismatched = false;
2437 BasicType bt = alias_type->basic_type();
2438 if (bt != T_ILLEGAL) {
2439 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2440 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2441 // Alias type doesn't differentiate between byte[] and boolean[]).
2442 // Use address type to get the element type.
2443 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2444 }
2445 if (is_reference_type(bt, true)) {
2446 // accessing an array field with getReference is not a mismatch
2447 bt = T_OBJECT;
2448 }
2449 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2450 // Don't intrinsify mismatched object accesses
2451 return false;
2452 }
2453 mismatched = (bt != type);
2454 } else if (alias_type->adr_type()->isa_oopptr()) {
2455 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2456 }
2457
2458 old_state.discard();
2459 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2460
2461 if (mismatched) {
2462 decorators |= C2_MISMATCHED;
2463 }
2464
2465 // First guess at the value type.
2466 const Type *value_type = Type::get_const_basic_type(type);
2467
2468 // Figure out the memory ordering.
2469 decorators |= mo_decorator_for_access_kind(kind);
2470
2471 if (!is_store && type == T_OBJECT) {
2472 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2473 if (tjp != nullptr) {
2474 value_type = tjp;
2475 }
2476 }
2477
2478 receiver = null_check(receiver);
2479 if (stopped()) {
2480 return true;
2481 }
2482 // Heap pointers get a null-check from the interpreter,
2483 // as a courtesy. However, this is not guaranteed by Unsafe,
2484 // and it is not possible to fully distinguish unintended nulls
2485 // from intended ones in this API.
2486
2487 if (!is_store) {
2488 Node* p = nullptr;
2489 // Try to constant fold a load from a constant field
2490 ciField* field = alias_type->field();
2491 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2492 // final or stable field
2493 p = make_constant_from_field(field, heap_base_oop);
2494 }
2495
2496 if (p == nullptr) { // Could not constant fold the load
2497 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2498 // Normalize the value returned by getBoolean in the following cases
2499 if (type == T_BOOLEAN &&
2500 (mismatched ||
2501 heap_base_oop == top() || // - heap_base_oop is null or
2502 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2503 // and the unsafe access is made to large offset
2504 // (i.e., larger than the maximum offset necessary for any
2505 // field access)
2506 ) {
2507 IdealKit ideal = IdealKit(this);
2508 #define __ ideal.
2509 IdealVariable normalized_result(ideal);
2510 __ declarations_done();
2511 __ set(normalized_result, p);
2512 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2513 __ set(normalized_result, ideal.ConI(1));
2514 ideal.end_if();
2515 final_sync(ideal);
2516 p = __ value(normalized_result);
2517 #undef __
2518 }
2519 }
2520 if (type == T_ADDRESS) {
2521 p = gvn().transform(new CastP2XNode(nullptr, p));
2522 p = ConvX2UL(p);
2523 }
2524 // The load node has the control of the preceding MemBarCPUOrder. All
2525 // following nodes will have the control of the MemBarCPUOrder inserted at
2526 // the end of this method. So, pushing the load onto the stack at a later
2527 // point is fine.
2528 set_result(p);
2529 } else {
2530 if (bt == T_ADDRESS) {
2531 // Repackage the long as a pointer.
2532 val = ConvL2X(val);
2533 val = gvn().transform(new CastX2PNode(val));
2534 }
2535 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2536 }
2537
2538 return true;
2539 }
2540
2541 //----------------------------inline_unsafe_load_store----------------------------
2542 // This method serves a couple of different customers (depending on LoadStoreKind):
2543 //
2544 // LS_cmp_swap:
2545 //
2546 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2547 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2548 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2549 //
2550 // LS_cmp_swap_weak:
2551 //
2552 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2553 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2554 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2555 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2556 //
2557 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2558 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2559 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2560 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2561 //
2562 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2563 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2564 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2565 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2566 //
2567 // LS_cmp_exchange:
2568 //
2569 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2570 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2571 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2572 //
2573 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2574 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2575 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2576 //
2577 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2578 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2579 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2580 //
2581 // LS_get_add:
2582 //
2583 // int getAndAddInt( Object o, long offset, int delta)
2584 // long getAndAddLong(Object o, long offset, long delta)
2585 //
2586 // LS_get_set:
2587 //
2588 // int getAndSet(Object o, long offset, int newValue)
2589 // long getAndSet(Object o, long offset, long newValue)
2590 // Object getAndSet(Object o, long offset, Object newValue)
2591 //
2592 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2593 // This basic scheme here is the same as inline_unsafe_access, but
2594 // differs in enough details that combining them would make the code
2595 // overly confusing. (This is a true fact! I originally combined
2596 // them, but even I was confused by it!) As much code/comments as
2597 // possible are retained from inline_unsafe_access though to make
2598 // the correspondences clearer. - dl
2599
2600 if (callee()->is_static()) return false; // caller must have the capability!
2601
2602 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2603 decorators |= mo_decorator_for_access_kind(access_kind);
2604
2605 #ifndef PRODUCT
2606 BasicType rtype;
2607 {
2608 ResourceMark rm;
2609 // Check the signatures.
2610 ciSignature* sig = callee()->signature();
2611 rtype = sig->return_type()->basic_type();
2612 switch(kind) {
2613 case LS_get_add:
2614 case LS_get_set: {
2615 // Check the signatures.
2616 #ifdef ASSERT
2617 assert(rtype == type, "get and set must return the expected type");
2618 assert(sig->count() == 3, "get and set has 3 arguments");
2619 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2620 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2621 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2622 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2623 #endif // ASSERT
2624 break;
2625 }
2626 case LS_cmp_swap:
2627 case LS_cmp_swap_weak: {
2628 // Check the signatures.
2629 #ifdef ASSERT
2630 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2631 assert(sig->count() == 4, "CAS has 4 arguments");
2632 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2633 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2634 #endif // ASSERT
2635 break;
2636 }
2637 case LS_cmp_exchange: {
2638 // Check the signatures.
2639 #ifdef ASSERT
2640 assert(rtype == type, "CAS must return the expected type");
2641 assert(sig->count() == 4, "CAS has 4 arguments");
2642 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2643 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2644 #endif // ASSERT
2645 break;
2646 }
2647 default:
2648 ShouldNotReachHere();
2649 }
2650 }
2651 #endif //PRODUCT
2652
2653 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2654
2655 // Get arguments:
2656 Node* receiver = nullptr;
2657 Node* base = nullptr;
2658 Node* offset = nullptr;
2659 Node* oldval = nullptr;
2660 Node* newval = nullptr;
2661 switch(kind) {
2662 case LS_cmp_swap:
2663 case LS_cmp_swap_weak:
2664 case LS_cmp_exchange: {
2665 const bool two_slot_type = type2size[type] == 2;
2666 receiver = argument(0); // type: oop
2667 base = argument(1); // type: oop
2668 offset = argument(2); // type: long
2669 oldval = argument(4); // type: oop, int, or long
2670 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2671 break;
2672 }
2673 case LS_get_add:
2674 case LS_get_set: {
2675 receiver = argument(0); // type: oop
2676 base = argument(1); // type: oop
2677 offset = argument(2); // type: long
2678 oldval = nullptr;
2679 newval = argument(4); // type: oop, int, or long
2680 break;
2681 }
2682 default:
2683 ShouldNotReachHere();
2684 }
2685
2686 // Build field offset expression.
2687 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2688 // to be plain byte offsets, which are also the same as those accepted
2689 // by oopDesc::field_addr.
2690 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2691 // 32-bit machines ignore the high half of long offsets
2692 offset = ConvL2X(offset);
2693 // Save state and restore on bailout
2694 SavedState old_state(this);
2695 Node* adr = make_unsafe_address(base, offset,type, false);
2696 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2697
2698 Compile::AliasType* alias_type = C->alias_type(adr_type);
2699 BasicType bt = alias_type->basic_type();
2700 if (bt != T_ILLEGAL &&
2701 (is_reference_type(bt) != (type == T_OBJECT))) {
2702 // Don't intrinsify mismatched object accesses.
2703 return false;
2704 }
2705
2706 old_state.discard();
2707
2708 // For CAS, unlike inline_unsafe_access, there seems no point in
2709 // trying to refine types. Just use the coarse types here.
2710 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2711 const Type *value_type = Type::get_const_basic_type(type);
2712
2713 switch (kind) {
2714 case LS_get_set:
2715 case LS_cmp_exchange: {
2716 if (type == T_OBJECT) {
2717 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2718 if (tjp != nullptr) {
2719 value_type = tjp;
2720 }
2721 }
2722 break;
2723 }
2724 case LS_cmp_swap:
2725 case LS_cmp_swap_weak:
2726 case LS_get_add:
2727 break;
2728 default:
2729 ShouldNotReachHere();
2730 }
2731
2732 // Null check receiver.
2733 receiver = null_check(receiver);
2734 if (stopped()) {
2735 return true;
2736 }
2737
2738 int alias_idx = C->get_alias_index(adr_type);
2739
2740 if (is_reference_type(type)) {
2741 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2742
2743 // Transformation of a value which could be null pointer (CastPP #null)
2744 // could be delayed during Parse (for example, in adjust_map_after_if()).
2745 // Execute transformation here to avoid barrier generation in such case.
2746 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2747 newval = _gvn.makecon(TypePtr::NULL_PTR);
2748
2749 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2750 // Refine the value to a null constant, when it is known to be null
2751 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2752 }
2753 }
2754
2755 Node* result = nullptr;
2756 switch (kind) {
2757 case LS_cmp_exchange: {
2758 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2759 oldval, newval, value_type, type, decorators);
2760 break;
2761 }
2762 case LS_cmp_swap_weak:
2763 decorators |= C2_WEAK_CMPXCHG;
2764 case LS_cmp_swap: {
2765 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2766 oldval, newval, value_type, type, decorators);
2767 break;
2768 }
2769 case LS_get_set: {
2770 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2771 newval, value_type, type, decorators);
2772 break;
2773 }
2774 case LS_get_add: {
2775 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2776 newval, value_type, type, decorators);
2777 break;
2778 }
2779 default:
2780 ShouldNotReachHere();
2781 }
2782
2783 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2784 set_result(result);
2785 return true;
2786 }
2787
2788 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2789 // Regardless of form, don't allow previous ld/st to move down,
2790 // then issue acquire, release, or volatile mem_bar.
2791 insert_mem_bar(Op_MemBarCPUOrder);
2792 switch(id) {
2793 case vmIntrinsics::_loadFence:
2794 insert_mem_bar(Op_LoadFence);
2795 return true;
2796 case vmIntrinsics::_storeFence:
2797 insert_mem_bar(Op_StoreFence);
2798 return true;
2799 case vmIntrinsics::_storeStoreFence:
2800 insert_mem_bar(Op_StoreStoreFence);
2801 return true;
2802 case vmIntrinsics::_fullFence:
2803 insert_mem_bar(Op_MemBarFull);
2804 return true;
2805 default:
2806 fatal_unexpected_iid(id);
2807 return false;
2808 }
2809 }
2810
2811 bool LibraryCallKit::inline_onspinwait() {
2812 insert_mem_bar(Op_OnSpinWait);
2813 return true;
2814 }
2815
2816 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2817 if (!kls->is_Con()) {
2818 return true;
2819 }
2820 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2821 if (klsptr == nullptr) {
2822 return true;
2823 }
2824 ciInstanceKlass* ik = klsptr->instance_klass();
2825 // don't need a guard for a klass that is already initialized
2826 return !ik->is_initialized();
2827 }
2828
2829 //----------------------------inline_unsafe_writeback0-------------------------
2830 // public native void Unsafe.writeback0(long address)
2831 bool LibraryCallKit::inline_unsafe_writeback0() {
2832 if (!Matcher::has_match_rule(Op_CacheWB)) {
2833 return false;
2834 }
2835 #ifndef PRODUCT
2836 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2837 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2838 ciSignature* sig = callee()->signature();
2839 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2840 #endif
2841 null_check_receiver(); // null-check, then ignore
2842 Node *addr = argument(1);
2843 addr = new CastX2PNode(addr);
2844 addr = _gvn.transform(addr);
2845 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2846 flush = _gvn.transform(flush);
2847 set_memory(flush, TypeRawPtr::BOTTOM);
2848 return true;
2849 }
2850
2851 //----------------------------inline_unsafe_writeback0-------------------------
2852 // public native void Unsafe.writeback0(long address)
2853 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2854 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2855 return false;
2856 }
2857 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2858 return false;
2859 }
2860 #ifndef PRODUCT
2861 assert(Matcher::has_match_rule(Op_CacheWB),
2862 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2863 : "found match rule for CacheWBPostSync but not CacheWB"));
2864
2865 #endif
2866 null_check_receiver(); // null-check, then ignore
2867 Node *sync;
2868 if (is_pre) {
2869 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2870 } else {
2871 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2872 }
2873 sync = _gvn.transform(sync);
2874 set_memory(sync, TypeRawPtr::BOTTOM);
2875 return true;
2876 }
2877
2878 //----------------------------inline_unsafe_allocate---------------------------
2879 // public native Object Unsafe.allocateInstance(Class<?> cls);
2880 bool LibraryCallKit::inline_unsafe_allocate() {
2881
2882 #if INCLUDE_JVMTI
2883 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2884 return false;
2885 }
2886 #endif //INCLUDE_JVMTI
2887
2888 if (callee()->is_static()) return false; // caller must have the capability!
2889
2890 null_check_receiver(); // null-check, then ignore
2891 Node* cls = null_check(argument(1));
2892 if (stopped()) return true;
2893
2894 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2895 kls = null_check(kls);
2896 if (stopped()) return true; // argument was like int.class
2897
2898 #if INCLUDE_JVMTI
2899 // Don't try to access new allocated obj in the intrinsic.
2900 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2901 // Deoptimize and allocate in interpreter instead.
2902 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2903 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2904 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2905 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2906 {
2907 BuildCutout unless(this, tst, PROB_MAX);
2908 uncommon_trap(Deoptimization::Reason_intrinsic,
2909 Deoptimization::Action_make_not_entrant);
2910 }
2911 if (stopped()) {
2912 return true;
2913 }
2914 #endif //INCLUDE_JVMTI
2915
2916 Node* test = nullptr;
2917 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2918 // Note: The argument might still be an illegal value like
2919 // Serializable.class or Object[].class. The runtime will handle it.
2920 // But we must make an explicit check for initialization.
2921 Node* insp = off_heap_plus_addr(kls, in_bytes(InstanceKlass::init_state_offset()));
2922 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2923 // can generate code to load it as unsigned byte.
2924 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2925 Node* bits = intcon(InstanceKlass::fully_initialized);
2926 test = _gvn.transform(new SubINode(inst, bits));
2927 // The 'test' is non-zero if we need to take a slow path.
2928 }
2929
2930 Node* obj = new_instance(kls, test);
2931 set_result(obj);
2932 return true;
2933 }
2934
2935 //------------------------inline_native_time_funcs--------------
2936 // inline code for System.currentTimeMillis() and System.nanoTime()
2937 // these have the same type and signature
2938 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2939 const TypeFunc* tf = OptoRuntime::void_long_Type();
2940 const TypePtr* no_memory_effects = nullptr;
2941 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2942 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2943 #ifdef ASSERT
2944 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2945 assert(value_top == top(), "second value must be top");
2946 #endif
2947 set_result(value);
2948 return true;
2949 }
2950
2951 //--------------------inline_native_vthread_start_transition--------------------
2952 // inline void startTransition(boolean is_mount);
2953 // inline void startFinalTransition();
2954 // Pseudocode of implementation:
2955 //
2956 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
2957 // carrier->set_is_in_vthread_transition(true);
2958 // OrderAccess::storeload();
2959 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
2960 // + global_vthread_transition_disable_count();
2961 // if (disable_requests > 0) {
2962 // slow path: runtime call
2963 // }
2964 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
2965 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
2966 IdealKit ideal(this);
2967
2968 Node* thread = ideal.thread();
2969 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
2970 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
2971 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2972 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2973 insert_mem_bar(Op_MemBarStoreLoad);
2974 ideal.sync_kit(this);
2975
2976 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
2977 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
2978 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
2979 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
2980 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, C->get_alias_index(vt_disable_addr_t), true /*require_atomic_access*/);
2981 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
2982
2983 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
2984 sync_kit(ideal);
2985 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
2986 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
2987 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
2988 ideal.sync_kit(this);
2989 }
2990 ideal.end_if();
2991
2992 final_sync(ideal);
2993 return true;
2994 }
2995
2996 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
2997 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
2998 IdealKit ideal(this);
2999
3000 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3001 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3002
3003 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3004 sync_kit(ideal);
3005 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3006 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3007 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3008 ideal.sync_kit(this);
3009 } ideal.else_(); {
3010 Node* thread = ideal.thread();
3011 Node* jt_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3012 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3013
3014 sync_kit(ideal);
3015 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3016 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3017 ideal.sync_kit(this);
3018 } ideal.end_if();
3019
3020 final_sync(ideal);
3021 return true;
3022 }
3023
3024 #if INCLUDE_JVMTI
3025
3026 // Always update the is_disable_suspend bit.
3027 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3028 if (!DoJVMTIVirtualThreadTransitions) {
3029 return true;
3030 }
3031 IdealKit ideal(this);
3032
3033 {
3034 // unconditionally update the is_disable_suspend bit in current JavaThread
3035 Node* thread = ideal.thread();
3036 Node* arg = argument(0); // argument for notification
3037 Node* addr = off_heap_plus_addr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3038 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3039
3040 sync_kit(ideal);
3041 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3042 ideal.sync_kit(this);
3043 }
3044 final_sync(ideal);
3045
3046 return true;
3047 }
3048
3049 #endif // INCLUDE_JVMTI
3050
3051 #ifdef JFR_HAVE_INTRINSICS
3052
3053 /**
3054 * if oop->klass != null
3055 * // normal class
3056 * epoch = _epoch_state ? 2 : 1
3057 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3058 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3059 * }
3060 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3061 * else
3062 * // primitive class
3063 * if oop->array_klass != null
3064 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3065 * else
3066 * id = LAST_TYPE_ID + 1 // void class path
3067 * if (!signaled)
3068 * signaled = true
3069 */
3070 bool LibraryCallKit::inline_native_classID() {
3071 Node* cls = argument(0);
3072
3073 IdealKit ideal(this);
3074 #define __ ideal.
3075 IdealVariable result(ideal); __ declarations_done();
3076 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3077 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3078 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3079
3080
3081 __ if_then(kls, BoolTest::ne, null()); {
3082 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3083 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3084
3085 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3086 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3087 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3088 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3089 mask = _gvn.transform(new OrLNode(mask, epoch));
3090 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3091
3092 float unlikely = PROB_UNLIKELY(0.999);
3093 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3094 sync_kit(ideal);
3095 make_runtime_call(RC_LEAF,
3096 OptoRuntime::class_id_load_barrier_Type(),
3097 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3098 "class id load barrier",
3099 TypePtr::BOTTOM,
3100 kls);
3101 ideal.sync_kit(this);
3102 } __ end_if();
3103
3104 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3105 } __ else_(); {
3106 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3107 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3108 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3109 __ if_then(array_kls, BoolTest::ne, null()); {
3110 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3111 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3112 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3113 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3114 } __ else_(); {
3115 // void class case
3116 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3117 } __ end_if();
3118
3119 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3120 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3121 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3122 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3123 } __ end_if();
3124 } __ end_if();
3125
3126 final_sync(ideal);
3127 set_result(ideal.value(result));
3128 #undef __
3129 return true;
3130 }
3131
3132 //------------------------inline_native_jvm_commit------------------
3133 bool LibraryCallKit::inline_native_jvm_commit() {
3134 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3135
3136 // Save input memory and i_o state.
3137 Node* input_memory_state = reset_memory();
3138 set_all_memory(input_memory_state);
3139 Node* input_io_state = i_o();
3140
3141 // TLS.
3142 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3143 // Jfr java buffer.
3144 Node* java_buffer_offset = _gvn.transform(AddPNode::make_off_heap(tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3145 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3146 Node* java_buffer_pos_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3147
3148 // Load the current value of the notified field in the JfrThreadLocal.
3149 Node* notified_offset = off_heap_plus_addr(tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3150 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3151
3152 // Test for notification.
3153 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3154 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3155 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3156
3157 // True branch, is notified.
3158 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3159 set_control(is_notified);
3160
3161 // Reset notified state.
3162 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3163 Node* notified_reset_memory = reset_memory();
3164
3165 // 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.
3166 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3167 // Convert the machine-word to a long.
3168 Node* current_pos = ConvX2L(current_pos_X);
3169
3170 // False branch, not notified.
3171 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3172 set_control(not_notified);
3173 set_all_memory(input_memory_state);
3174
3175 // Arg is the next position as a long.
3176 Node* arg = argument(0);
3177 // Convert long to machine-word.
3178 Node* next_pos_X = ConvL2X(arg);
3179
3180 // Store the next_position to the underlying jfr java buffer.
3181 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3182
3183 Node* commit_memory = reset_memory();
3184 set_all_memory(commit_memory);
3185
3186 // 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.
3187 Node* java_buffer_flags_offset = _gvn.transform(AddPNode::make_off_heap(java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3188 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3189 Node* lease_constant = _gvn.intcon(4);
3190
3191 // And flags with lease constant.
3192 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3193
3194 // Branch on lease to conditionalize returning the leased java buffer.
3195 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3196 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3197 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3198
3199 // False branch, not a lease.
3200 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3201
3202 // True branch, is lease.
3203 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3204 set_control(is_lease);
3205
3206 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3207 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3208 OptoRuntime::void_void_Type(),
3209 SharedRuntime::jfr_return_lease(),
3210 "return_lease", TypePtr::BOTTOM);
3211 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3212
3213 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3214 record_for_igvn(lease_compare_rgn);
3215 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3216 record_for_igvn(lease_compare_mem);
3217 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3218 record_for_igvn(lease_compare_io);
3219 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3220 record_for_igvn(lease_result_value);
3221
3222 // Update control and phi nodes.
3223 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3224 lease_compare_rgn->init_req(_false_path, not_lease);
3225
3226 lease_compare_mem->init_req(_true_path, reset_memory());
3227 lease_compare_mem->init_req(_false_path, commit_memory);
3228
3229 lease_compare_io->init_req(_true_path, i_o());
3230 lease_compare_io->init_req(_false_path, input_io_state);
3231
3232 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3233 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3234
3235 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3236 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3237 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3238 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3239
3240 // Update control and phi nodes.
3241 result_rgn->init_req(_true_path, is_notified);
3242 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3243
3244 result_mem->init_req(_true_path, notified_reset_memory);
3245 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3246
3247 result_io->init_req(_true_path, input_io_state);
3248 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3249
3250 result_value->init_req(_true_path, current_pos);
3251 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3252
3253 // Set output state.
3254 set_control(_gvn.transform(result_rgn));
3255 set_all_memory(_gvn.transform(result_mem));
3256 set_i_o(_gvn.transform(result_io));
3257 set_result(result_rgn, result_value);
3258 return true;
3259 }
3260
3261 /*
3262 * The intrinsic is a model of this pseudo-code:
3263 *
3264 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3265 * jobject h_event_writer = tl->java_event_writer();
3266 * if (h_event_writer == nullptr) {
3267 * return nullptr;
3268 * }
3269 * oop threadObj = Thread::threadObj();
3270 * oop vthread = java_lang_Thread::vthread(threadObj);
3271 * traceid tid;
3272 * bool pinVirtualThread;
3273 * bool excluded;
3274 * if (vthread != threadObj) { // i.e. current thread is virtual
3275 * tid = java_lang_Thread::tid(vthread);
3276 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3277 * pinVirtualThread = VMContinuations;
3278 * excluded = vthread_epoch_raw & excluded_mask;
3279 * if (!excluded) {
3280 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3281 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3282 * if (vthread_epoch != current_epoch) {
3283 * write_checkpoint();
3284 * }
3285 * }
3286 * } else {
3287 * tid = java_lang_Thread::tid(threadObj);
3288 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3289 * pinVirtualThread = false;
3290 * excluded = thread_epoch_raw & excluded_mask;
3291 * }
3292 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3293 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3294 * if (tid_in_event_writer != tid) {
3295 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3296 * setField(event_writer, "excluded", excluded);
3297 * setField(event_writer, "threadID", tid);
3298 * }
3299 * return event_writer
3300 */
3301 bool LibraryCallKit::inline_native_getEventWriter() {
3302 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3303
3304 // Save input memory and i_o state.
3305 Node* input_memory_state = reset_memory();
3306 set_all_memory(input_memory_state);
3307 Node* input_io_state = i_o();
3308
3309 // The most significant bit of the u2 is used to denote thread exclusion
3310 Node* excluded_shift = _gvn.intcon(15);
3311 Node* excluded_mask = _gvn.intcon(1 << 15);
3312 // The epoch generation is the range [1-32767]
3313 Node* epoch_mask = _gvn.intcon(32767);
3314
3315 // TLS
3316 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3317
3318 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3319 Node* jobj_ptr = off_heap_plus_addr(tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3320
3321 // Load the eventwriter jobject handle.
3322 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3323
3324 // Null check the jobject handle.
3325 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3326 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3327 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3328
3329 // False path, jobj is null.
3330 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3331
3332 // True path, jobj is not null.
3333 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3334
3335 set_control(jobj_is_not_null);
3336
3337 // Load the threadObj for the CarrierThread.
3338 Node* threadObj = generate_current_thread(tls_ptr);
3339
3340 // Load the vthread.
3341 Node* vthread = generate_virtual_thread(tls_ptr);
3342
3343 // If vthread != threadObj, this is a virtual thread.
3344 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3345 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3346 IfNode* iff_vthread_not_equal_threadObj =
3347 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3348
3349 // False branch, fallback to threadObj.
3350 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3351 set_control(vthread_equal_threadObj);
3352
3353 // Load the tid field from the vthread object.
3354 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3355
3356 // Load the raw epoch value from the threadObj.
3357 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3358 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3359 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3360 TypeInt::CHAR, T_CHAR,
3361 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3362
3363 // Mask off the excluded information from the epoch.
3364 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3365
3366 // True branch, this is a virtual thread.
3367 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3368 set_control(vthread_not_equal_threadObj);
3369
3370 // Load the tid field from the vthread object.
3371 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3372
3373 // Continuation support determines if a virtual thread should be pinned.
3374 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3375 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3376
3377 // Load the raw epoch value from the vthread.
3378 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3379 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3380 TypeInt::CHAR, T_CHAR,
3381 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3382
3383 // Mask off the excluded information from the epoch.
3384 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3385
3386 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3387 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3388 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3389 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3390
3391 // False branch, vthread is excluded, no need to write epoch info.
3392 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3393
3394 // True branch, vthread is included, update epoch info.
3395 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3396 set_control(included);
3397
3398 // Get epoch value.
3399 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3400
3401 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3402 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3403 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3404
3405 // Compare the epoch in the vthread to the current epoch generation.
3406 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3407 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3408 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3409
3410 // False path, epoch is equal, checkpoint information is valid.
3411 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3412
3413 // True path, epoch is not equal, write a checkpoint for the vthread.
3414 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3415
3416 set_control(epoch_is_not_equal);
3417
3418 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3419 // The call also updates the native thread local thread id and the vthread with the current epoch.
3420 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3421 OptoRuntime::jfr_write_checkpoint_Type(),
3422 SharedRuntime::jfr_write_checkpoint(),
3423 "write_checkpoint", TypePtr::BOTTOM);
3424 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3425
3426 // vthread epoch != current epoch
3427 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3428 record_for_igvn(epoch_compare_rgn);
3429 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3430 record_for_igvn(epoch_compare_mem);
3431 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3432 record_for_igvn(epoch_compare_io);
3433
3434 // Update control and phi nodes.
3435 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3436 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3437 epoch_compare_mem->init_req(_true_path, reset_memory());
3438 epoch_compare_mem->init_req(_false_path, input_memory_state);
3439 epoch_compare_io->init_req(_true_path, i_o());
3440 epoch_compare_io->init_req(_false_path, input_io_state);
3441
3442 // excluded != true
3443 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3444 record_for_igvn(exclude_compare_rgn);
3445 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3446 record_for_igvn(exclude_compare_mem);
3447 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3448 record_for_igvn(exclude_compare_io);
3449
3450 // Update control and phi nodes.
3451 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3452 exclude_compare_rgn->init_req(_false_path, excluded);
3453 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3454 exclude_compare_mem->init_req(_false_path, input_memory_state);
3455 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3456 exclude_compare_io->init_req(_false_path, input_io_state);
3457
3458 // vthread != threadObj
3459 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3460 record_for_igvn(vthread_compare_rgn);
3461 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3462 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3463 record_for_igvn(vthread_compare_io);
3464 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3465 record_for_igvn(tid);
3466 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3467 record_for_igvn(exclusion);
3468 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3469 record_for_igvn(pinVirtualThread);
3470
3471 // Update control and phi nodes.
3472 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3473 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3474 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3475 vthread_compare_mem->init_req(_false_path, input_memory_state);
3476 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3477 vthread_compare_io->init_req(_false_path, input_io_state);
3478 tid->init_req(_true_path, vthread_tid);
3479 tid->init_req(_false_path, thread_obj_tid);
3480 exclusion->init_req(_true_path, vthread_is_excluded);
3481 exclusion->init_req(_false_path, threadObj_is_excluded);
3482 pinVirtualThread->init_req(_true_path, continuation_support);
3483 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3484
3485 // Update branch state.
3486 set_control(_gvn.transform(vthread_compare_rgn));
3487 set_all_memory(_gvn.transform(vthread_compare_mem));
3488 set_i_o(_gvn.transform(vthread_compare_io));
3489
3490 // Load the event writer oop by dereferencing the jobject handle.
3491 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3492 assert(klass_EventWriter->is_loaded(), "invariant");
3493 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3494 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3495 const TypeOopPtr* const xtype = aklass->as_instance_type();
3496 Node* jobj_untagged = _gvn.transform(AddPNode::make_off_heap(jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3497 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3498
3499 // Load the current thread id from the event writer object.
3500 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3501 // Get the field offset to, conditionally, store an updated tid value later.
3502 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3503 // Get the field offset to, conditionally, store an updated exclusion value later.
3504 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3505 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3506 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3507
3508 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3509 record_for_igvn(event_writer_tid_compare_rgn);
3510 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3511 record_for_igvn(event_writer_tid_compare_mem);
3512 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3513 record_for_igvn(event_writer_tid_compare_io);
3514
3515 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3516 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3517 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3518 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3519
3520 // False path, tids are the same.
3521 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3522
3523 // True path, tid is not equal, need to update the tid in the event writer.
3524 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3525 record_for_igvn(tid_is_not_equal);
3526
3527 // Store the pin state to the event writer.
3528 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3529
3530 // Store the exclusion state to the event writer.
3531 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3532 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3533
3534 // Store the tid to the event writer.
3535 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3536
3537 // Update control and phi nodes.
3538 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3539 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3540 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3541 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3542 event_writer_tid_compare_io->init_req(_true_path, i_o());
3543 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3544
3545 // Result of top level CFG, Memory, IO and Value.
3546 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3547 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3548 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3549 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3550
3551 // Result control.
3552 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3553 result_rgn->init_req(_false_path, jobj_is_null);
3554
3555 // Result memory.
3556 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3557 result_mem->init_req(_false_path, input_memory_state);
3558
3559 // Result IO.
3560 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3561 result_io->init_req(_false_path, input_io_state);
3562
3563 // Result value.
3564 result_value->init_req(_true_path, event_writer); // return event writer oop
3565 result_value->init_req(_false_path, null()); // return null
3566
3567 // Set output state.
3568 set_control(_gvn.transform(result_rgn));
3569 set_all_memory(_gvn.transform(result_mem));
3570 set_i_o(_gvn.transform(result_io));
3571 set_result(result_rgn, result_value);
3572 return true;
3573 }
3574
3575 /*
3576 * The intrinsic is a model of this pseudo-code:
3577 *
3578 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3579 * if (carrierThread != thread) { // is virtual thread
3580 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3581 * bool excluded = vthread_epoch_raw & excluded_mask;
3582 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3583 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3584 * if (!excluded) {
3585 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3586 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3587 * }
3588 * AtomicAccess::release_store(&tl->_vthread, true);
3589 * return;
3590 * }
3591 * AtomicAccess::release_store(&tl->_vthread, false);
3592 */
3593 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3594 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3595
3596 Node* input_memory_state = reset_memory();
3597 set_all_memory(input_memory_state);
3598
3599 // The most significant bit of the u2 is used to denote thread exclusion
3600 Node* excluded_mask = _gvn.intcon(1 << 15);
3601 // The epoch generation is the range [1-32767]
3602 Node* epoch_mask = _gvn.intcon(32767);
3603
3604 Node* const carrierThread = generate_current_thread(jt);
3605 // If thread != carrierThread, this is a virtual thread.
3606 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3607 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3608 IfNode* iff_thread_not_equal_carrierThread =
3609 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3610
3611 Node* vthread_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3612
3613 // False branch, is carrierThread.
3614 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3615 // Store release
3616 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3617
3618 set_all_memory(input_memory_state);
3619
3620 // True branch, is virtual thread.
3621 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3622 set_control(thread_not_equal_carrierThread);
3623
3624 // Load the raw epoch value from the vthread.
3625 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3626 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3627 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3628
3629 // Mask off the excluded information from the epoch.
3630 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
3631
3632 // Load the tid field from the thread.
3633 Node* tid = load_field_from_object(thread, "tid", "J");
3634
3635 // Store the vthread tid to the jfr thread local.
3636 Node* thread_id_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3637 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3638
3639 // Branch is_excluded to conditionalize updating the epoch .
3640 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
3641 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3642 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3643
3644 // True branch, vthread is excluded, no need to write epoch info.
3645 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3646 set_control(excluded);
3647 Node* vthread_is_excluded = _gvn.intcon(1);
3648
3649 // False branch, vthread is included, update epoch info.
3650 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3651 set_control(included);
3652 Node* vthread_is_included = _gvn.intcon(0);
3653
3654 // Get epoch value.
3655 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
3656
3657 // Store the vthread epoch to the jfr thread local.
3658 Node* vthread_epoch_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3659 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3660
3661 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3662 record_for_igvn(excluded_rgn);
3663 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3664 record_for_igvn(excluded_mem);
3665 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3666 record_for_igvn(exclusion);
3667
3668 // Merge the excluded control and memory.
3669 excluded_rgn->init_req(_true_path, excluded);
3670 excluded_rgn->init_req(_false_path, included);
3671 excluded_mem->init_req(_true_path, tid_memory);
3672 excluded_mem->init_req(_false_path, included_memory);
3673 exclusion->init_req(_true_path, vthread_is_excluded);
3674 exclusion->init_req(_false_path, vthread_is_included);
3675
3676 // Set intermediate state.
3677 set_control(_gvn.transform(excluded_rgn));
3678 set_all_memory(excluded_mem);
3679
3680 // Store the vthread exclusion state to the jfr thread local.
3681 Node* thread_local_excluded_offset = off_heap_plus_addr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3682 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3683
3684 // Store release
3685 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3686
3687 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3688 record_for_igvn(thread_compare_rgn);
3689 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3690 record_for_igvn(thread_compare_mem);
3691 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3692 record_for_igvn(vthread);
3693
3694 // Merge the thread_compare control and memory.
3695 thread_compare_rgn->init_req(_true_path, control());
3696 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3697 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3698 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3699
3700 // Set output state.
3701 set_control(_gvn.transform(thread_compare_rgn));
3702 set_all_memory(_gvn.transform(thread_compare_mem));
3703 }
3704
3705 #endif // JFR_HAVE_INTRINSICS
3706
3707 //------------------------inline_native_currentCarrierThread------------------
3708 bool LibraryCallKit::inline_native_currentCarrierThread() {
3709 Node* junk = nullptr;
3710 set_result(generate_current_thread(junk));
3711 return true;
3712 }
3713
3714 //------------------------inline_native_currentThread------------------
3715 bool LibraryCallKit::inline_native_currentThread() {
3716 Node* junk = nullptr;
3717 set_result(generate_virtual_thread(junk));
3718 return true;
3719 }
3720
3721 //------------------------inline_native_setVthread------------------
3722 bool LibraryCallKit::inline_native_setCurrentThread() {
3723 assert(C->method()->changes_current_thread(),
3724 "method changes current Thread but is not annotated ChangesCurrentThread");
3725 Node* arr = argument(1);
3726 Node* thread = _gvn.transform(new ThreadLocalNode());
3727 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::vthread_offset()));
3728 Node* thread_obj_handle
3729 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3730 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3731 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3732
3733 // Change the _monitor_owner_id of the JavaThread
3734 Node* tid = load_field_from_object(arr, "tid", "J");
3735 Node* monitor_owner_id_offset = off_heap_plus_addr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3736 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3737
3738 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3739 return true;
3740 }
3741
3742 const Type* LibraryCallKit::scopedValueCache_type() {
3743 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3744 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3745 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3746
3747 // Because we create the scopedValue cache lazily we have to make the
3748 // type of the result BotPTR.
3749 bool xk = etype->klass_is_exact();
3750 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3751 return objects_type;
3752 }
3753
3754 Node* LibraryCallKit::scopedValueCache_helper() {
3755 Node* thread = _gvn.transform(new ThreadLocalNode());
3756 Node* p = off_heap_plus_addr(thread, in_bytes(JavaThread::scopedValueCache_offset()));
3757 // We cannot use immutable_memory() because we might flip onto a
3758 // different carrier thread, at which point we'll need to use that
3759 // carrier thread's cache.
3760 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3761 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3762 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3763 }
3764
3765 //------------------------inline_native_scopedValueCache------------------
3766 bool LibraryCallKit::inline_native_scopedValueCache() {
3767 Node* cache_obj_handle = scopedValueCache_helper();
3768 const Type* objects_type = scopedValueCache_type();
3769 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3770
3771 return true;
3772 }
3773
3774 //------------------------inline_native_setScopedValueCache------------------
3775 bool LibraryCallKit::inline_native_setScopedValueCache() {
3776 Node* arr = argument(0);
3777 Node* cache_obj_handle = scopedValueCache_helper();
3778 const Type* objects_type = scopedValueCache_type();
3779
3780 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3781 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3782
3783 return true;
3784 }
3785
3786 //------------------------inline_native_Continuation_pin and unpin-----------
3787
3788 // Shared implementation routine for both pin and unpin.
3789 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3790 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3791
3792 // Save input memory.
3793 Node* input_memory_state = reset_memory();
3794 set_all_memory(input_memory_state);
3795
3796 // TLS
3797 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3798 Node* last_continuation_offset = off_heap_plus_addr(tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3799 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3800
3801 // Null check the last continuation object.
3802 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3803 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3804 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3805
3806 // False path, last continuation is null.
3807 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3808
3809 // True path, last continuation is not null.
3810 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3811
3812 set_control(continuation_is_not_null);
3813
3814 // Load the pin count from the last continuation.
3815 Node* pin_count_offset = off_heap_plus_addr(last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3816 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3817
3818 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3819 Node* pin_count_rhs;
3820 if (unpin) {
3821 pin_count_rhs = _gvn.intcon(0);
3822 } else {
3823 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3824 }
3825 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
3826 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3827 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3828
3829 // True branch, pin count over/underflow.
3830 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3831 {
3832 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3833 // which will throw IllegalStateException for pin count over/underflow.
3834 // No memory changed so far - we can use memory create by reset_memory()
3835 // at the beginning of this intrinsic. No need to call reset_memory() again.
3836 PreserveJVMState pjvms(this);
3837 set_control(pin_count_over_underflow);
3838 uncommon_trap(Deoptimization::Reason_intrinsic,
3839 Deoptimization::Action_none);
3840 assert(stopped(), "invariant");
3841 }
3842
3843 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3844 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3845 set_control(valid_pin_count);
3846
3847 Node* next_pin_count;
3848 if (unpin) {
3849 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3850 } else {
3851 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3852 }
3853
3854 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3855
3856 // Result of top level CFG and Memory.
3857 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3858 record_for_igvn(result_rgn);
3859 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3860 record_for_igvn(result_mem);
3861
3862 result_rgn->init_req(_true_path, valid_pin_count);
3863 result_rgn->init_req(_false_path, continuation_is_null);
3864 result_mem->init_req(_true_path, reset_memory());
3865 result_mem->init_req(_false_path, input_memory_state);
3866
3867 // Set output state.
3868 set_control(_gvn.transform(result_rgn));
3869 set_all_memory(_gvn.transform(result_mem));
3870
3871 return true;
3872 }
3873
3874 //---------------------------load_mirror_from_klass----------------------------
3875 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3876 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3877 Node* p = off_heap_plus_addr(klass, in_bytes(Klass::java_mirror_offset()));
3878 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3879 // mirror = ((OopHandle)mirror)->resolve();
3880 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3881 }
3882
3883 //-----------------------load_klass_from_mirror_common-------------------------
3884 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3885 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3886 // and branch to the given path on the region.
3887 // If never_see_null, take an uncommon trap on null, so we can optimistically
3888 // compile for the non-null case.
3889 // If the region is null, force never_see_null = true.
3890 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3891 bool never_see_null,
3892 RegionNode* region,
3893 int null_path,
3894 int offset) {
3895 if (region == nullptr) never_see_null = true;
3896 Node* p = basic_plus_adr(mirror, offset);
3897 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3898 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3899 Node* null_ctl = top();
3900 kls = null_check_oop(kls, &null_ctl, never_see_null);
3901 if (region != nullptr) {
3902 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3903 region->init_req(null_path, null_ctl);
3904 } else {
3905 assert(null_ctl == top(), "no loose ends");
3906 }
3907 return kls;
3908 }
3909
3910 //--------------------(inline_native_Class_query helpers)---------------------
3911 // Use this for JVM_ACC_INTERFACE.
3912 // Fall through if (mods & mask) == bits, take the guard otherwise.
3913 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3914 ByteSize offset, const Type* type, BasicType bt) {
3915 // Branch around if the given klass has the given modifier bit set.
3916 // Like generate_guard, adds a new path onto the region.
3917 Node* modp = off_heap_plus_addr(kls, in_bytes(offset));
3918 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3919 Node* mask = intcon(modifier_mask);
3920 Node* bits = intcon(modifier_bits);
3921 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3922 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3923 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3924 return generate_fair_guard(bol, region);
3925 }
3926 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3927 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3928 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3929 }
3930
3931 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3932 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3933 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3934 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3935 }
3936
3937 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3938 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3939 }
3940
3941 //-------------------------inline_native_Class_query-------------------
3942 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3943 const Type* return_type = TypeInt::BOOL;
3944 Node* prim_return_value = top(); // what happens if it's a primitive class?
3945 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3946 bool expect_prim = false; // most of these guys expect to work on refs
3947
3948 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3949
3950 Node* mirror = argument(0);
3951 Node* obj = top();
3952
3953 switch (id) {
3954 case vmIntrinsics::_isInstance:
3955 // nothing is an instance of a primitive type
3956 prim_return_value = intcon(0);
3957 obj = argument(1);
3958 break;
3959 case vmIntrinsics::_isHidden:
3960 prim_return_value = intcon(0);
3961 break;
3962 case vmIntrinsics::_getSuperclass:
3963 prim_return_value = null();
3964 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3965 break;
3966 default:
3967 fatal_unexpected_iid(id);
3968 break;
3969 }
3970
3971 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3972 if (mirror_con == nullptr) return false; // cannot happen?
3973
3974 #ifndef PRODUCT
3975 if (C->print_intrinsics() || C->print_inlining()) {
3976 ciType* k = mirror_con->java_mirror_type();
3977 if (k) {
3978 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3979 k->print_name();
3980 tty->cr();
3981 }
3982 }
3983 #endif
3984
3985 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3986 RegionNode* region = new RegionNode(PATH_LIMIT);
3987 record_for_igvn(region);
3988 PhiNode* phi = new PhiNode(region, return_type);
3989
3990 // The mirror will never be null of Reflection.getClassAccessFlags, however
3991 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3992 // if it is. See bug 4774291.
3993
3994 // For Reflection.getClassAccessFlags(), the null check occurs in
3995 // the wrong place; see inline_unsafe_access(), above, for a similar
3996 // situation.
3997 mirror = null_check(mirror);
3998 // If mirror or obj is dead, only null-path is taken.
3999 if (stopped()) return true;
4000
4001 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4002
4003 // Now load the mirror's klass metaobject, and null-check it.
4004 // Side-effects region with the control path if the klass is null.
4005 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4006 // If kls is null, we have a primitive mirror.
4007 phi->init_req(_prim_path, prim_return_value);
4008 if (stopped()) { set_result(region, phi); return true; }
4009 bool safe_for_replace = (region->in(_prim_path) == top());
4010
4011 Node* p; // handy temp
4012 Node* null_ctl;
4013
4014 // Now that we have the non-null klass, we can perform the real query.
4015 // For constant classes, the query will constant-fold in LoadNode::Value.
4016 Node* query_value = top();
4017 switch (id) {
4018 case vmIntrinsics::_isInstance:
4019 // nothing is an instance of a primitive type
4020 query_value = gen_instanceof(obj, kls, safe_for_replace);
4021 break;
4022
4023 case vmIntrinsics::_isHidden:
4024 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4025 if (generate_hidden_class_guard(kls, region) != nullptr)
4026 // A guard was added. If the guard is taken, it was an hidden class.
4027 phi->add_req(intcon(1));
4028 // If we fall through, it's a plain class.
4029 query_value = intcon(0);
4030 break;
4031
4032
4033 case vmIntrinsics::_getSuperclass:
4034 // The rules here are somewhat unfortunate, but we can still do better
4035 // with random logic than with a JNI call.
4036 // Interfaces store null or Object as _super, but must report null.
4037 // Arrays store an intermediate super as _super, but must report Object.
4038 // Other types can report the actual _super.
4039 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4040 if (generate_array_guard(kls, region) != nullptr) {
4041 // A guard was added. If the guard is taken, it was an array.
4042 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4043 }
4044 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4045 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4046 if (generate_interface_guard(kls, region) != nullptr) {
4047 // A guard was added. If the guard is taken, it was an interface.
4048 phi->add_req(null());
4049 }
4050 // If we fall through, it's a plain class. Get its _super.
4051 p = off_heap_plus_addr(kls, in_bytes(Klass::super_offset()));
4052 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4053 null_ctl = top();
4054 kls = null_check_oop(kls, &null_ctl);
4055 if (null_ctl != top()) {
4056 // If the guard is taken, Object.superClass is null (both klass and mirror).
4057 region->add_req(null_ctl);
4058 phi ->add_req(null());
4059 }
4060 if (!stopped()) {
4061 query_value = load_mirror_from_klass(kls);
4062 }
4063 break;
4064
4065 default:
4066 fatal_unexpected_iid(id);
4067 break;
4068 }
4069
4070 // Fall-through is the normal case of a query to a real class.
4071 phi->init_req(1, query_value);
4072 region->init_req(1, control());
4073
4074 C->set_has_split_ifs(true); // Has chance for split-if optimization
4075 set_result(region, phi);
4076 return true;
4077 }
4078
4079 //-------------------------inline_Class_cast-------------------
4080 bool LibraryCallKit::inline_Class_cast() {
4081 Node* mirror = argument(0); // Class
4082 Node* obj = argument(1);
4083 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4084 if (mirror_con == nullptr) {
4085 return false; // dead path (mirror->is_top()).
4086 }
4087 if (obj == nullptr || obj->is_top()) {
4088 return false; // dead path
4089 }
4090 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4091
4092 // First, see if Class.cast() can be folded statically.
4093 // java_mirror_type() returns non-null for compile-time Class constants.
4094 ciType* tm = mirror_con->java_mirror_type();
4095 if (tm != nullptr && tm->is_klass() &&
4096 tp != nullptr) {
4097 if (!tp->is_loaded()) {
4098 // Don't use intrinsic when class is not loaded.
4099 return false;
4100 } else {
4101 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4102 if (static_res == Compile::SSC_always_true) {
4103 // isInstance() is true - fold the code.
4104 set_result(obj);
4105 return true;
4106 } else if (static_res == Compile::SSC_always_false) {
4107 // Don't use intrinsic, have to throw ClassCastException.
4108 // If the reference is null, the non-intrinsic bytecode will
4109 // be optimized appropriately.
4110 return false;
4111 }
4112 }
4113 }
4114
4115 // Bailout intrinsic and do normal inlining if exception path is frequent.
4116 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4117 return false;
4118 }
4119
4120 // Generate dynamic checks.
4121 // Class.cast() is java implementation of _checkcast bytecode.
4122 // Do checkcast (Parse::do_checkcast()) optimizations here.
4123
4124 mirror = null_check(mirror);
4125 // If mirror is dead, only null-path is taken.
4126 if (stopped()) {
4127 return true;
4128 }
4129
4130 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4131 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4132 RegionNode* region = new RegionNode(PATH_LIMIT);
4133 record_for_igvn(region);
4134
4135 // Now load the mirror's klass metaobject, and null-check it.
4136 // If kls is null, we have a primitive mirror and
4137 // nothing is an instance of a primitive type.
4138 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4139
4140 Node* res = top();
4141 if (!stopped()) {
4142 Node* bad_type_ctrl = top();
4143 // Do checkcast optimizations.
4144 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4145 region->init_req(_bad_type_path, bad_type_ctrl);
4146 }
4147 if (region->in(_prim_path) != top() ||
4148 region->in(_bad_type_path) != top()) {
4149 // Let Interpreter throw ClassCastException.
4150 PreserveJVMState pjvms(this);
4151 set_control(_gvn.transform(region));
4152 uncommon_trap(Deoptimization::Reason_intrinsic,
4153 Deoptimization::Action_maybe_recompile);
4154 }
4155 if (!stopped()) {
4156 set_result(res);
4157 }
4158 return true;
4159 }
4160
4161
4162 //--------------------------inline_native_subtype_check------------------------
4163 // This intrinsic takes the JNI calls out of the heart of
4164 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4165 bool LibraryCallKit::inline_native_subtype_check() {
4166 // Pull both arguments off the stack.
4167 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4168 args[0] = argument(0);
4169 args[1] = argument(1);
4170 Node* klasses[2]; // corresponding Klasses: superk, subk
4171 klasses[0] = klasses[1] = top();
4172
4173 enum {
4174 // A full decision tree on {superc is prim, subc is prim}:
4175 _prim_0_path = 1, // {P,N} => false
4176 // {P,P} & superc!=subc => false
4177 _prim_same_path, // {P,P} & superc==subc => true
4178 _prim_1_path, // {N,P} => false
4179 _ref_subtype_path, // {N,N} & subtype check wins => true
4180 _both_ref_path, // {N,N} & subtype check loses => false
4181 PATH_LIMIT
4182 };
4183
4184 RegionNode* region = new RegionNode(PATH_LIMIT);
4185 Node* phi = new PhiNode(region, TypeInt::BOOL);
4186 record_for_igvn(region);
4187
4188 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4189 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4190 int class_klass_offset = java_lang_Class::klass_offset();
4191
4192 // First null-check both mirrors and load each mirror's klass metaobject.
4193 int which_arg;
4194 for (which_arg = 0; which_arg <= 1; which_arg++) {
4195 Node* arg = args[which_arg];
4196 arg = null_check(arg);
4197 if (stopped()) break;
4198 args[which_arg] = arg;
4199
4200 Node* p = basic_plus_adr(arg, class_klass_offset);
4201 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4202 klasses[which_arg] = _gvn.transform(kls);
4203 }
4204
4205 // Having loaded both klasses, test each for null.
4206 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4207 for (which_arg = 0; which_arg <= 1; which_arg++) {
4208 Node* kls = klasses[which_arg];
4209 Node* null_ctl = top();
4210 kls = null_check_oop(kls, &null_ctl, never_see_null);
4211 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4212 region->init_req(prim_path, null_ctl);
4213 if (stopped()) break;
4214 klasses[which_arg] = kls;
4215 }
4216
4217 if (!stopped()) {
4218 // now we have two reference types, in klasses[0..1]
4219 Node* subk = klasses[1]; // the argument to isAssignableFrom
4220 Node* superk = klasses[0]; // the receiver
4221 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4222 // now we have a successful reference subtype check
4223 region->set_req(_ref_subtype_path, control());
4224 }
4225
4226 // If both operands are primitive (both klasses null), then
4227 // we must return true when they are identical primitives.
4228 // It is convenient to test this after the first null klass check.
4229 set_control(region->in(_prim_0_path)); // go back to first null check
4230 if (!stopped()) {
4231 // Since superc is primitive, make a guard for the superc==subc case.
4232 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4233 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4234 generate_guard(bol_eq, region, PROB_FAIR);
4235 if (region->req() == PATH_LIMIT+1) {
4236 // A guard was added. If the added guard is taken, superc==subc.
4237 region->swap_edges(PATH_LIMIT, _prim_same_path);
4238 region->del_req(PATH_LIMIT);
4239 }
4240 region->set_req(_prim_0_path, control()); // Not equal after all.
4241 }
4242
4243 // these are the only paths that produce 'true':
4244 phi->set_req(_prim_same_path, intcon(1));
4245 phi->set_req(_ref_subtype_path, intcon(1));
4246
4247 // pull together the cases:
4248 assert(region->req() == PATH_LIMIT, "sane region");
4249 for (uint i = 1; i < region->req(); i++) {
4250 Node* ctl = region->in(i);
4251 if (ctl == nullptr || ctl == top()) {
4252 region->set_req(i, top());
4253 phi ->set_req(i, top());
4254 } else if (phi->in(i) == nullptr) {
4255 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4256 }
4257 }
4258
4259 set_control(_gvn.transform(region));
4260 set_result(_gvn.transform(phi));
4261 return true;
4262 }
4263
4264 //---------------------generate_array_guard_common------------------------
4265 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4266 bool obj_array, bool not_array, Node** obj) {
4267
4268 if (stopped()) {
4269 return nullptr;
4270 }
4271
4272 // If obj_array/non_array==false/false:
4273 // Branch around if the given klass is in fact an array (either obj or prim).
4274 // If obj_array/non_array==false/true:
4275 // Branch around if the given klass is not an array klass of any kind.
4276 // If obj_array/non_array==true/true:
4277 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4278 // If obj_array/non_array==true/false:
4279 // Branch around if the kls is an oop array (Object[] or subtype)
4280 //
4281 // Like generate_guard, adds a new path onto the region.
4282 jint layout_con = 0;
4283 Node* layout_val = get_layout_helper(kls, layout_con);
4284 if (layout_val == nullptr) {
4285 bool query = (obj_array
4286 ? Klass::layout_helper_is_objArray(layout_con)
4287 : Klass::layout_helper_is_array(layout_con));
4288 if (query == not_array) {
4289 return nullptr; // never a branch
4290 } else { // always a branch
4291 Node* always_branch = control();
4292 if (region != nullptr)
4293 region->add_req(always_branch);
4294 set_control(top());
4295 return always_branch;
4296 }
4297 }
4298 // Now test the correct condition.
4299 jint nval = (obj_array
4300 ? (jint)(Klass::_lh_array_tag_type_value
4301 << Klass::_lh_array_tag_shift)
4302 : Klass::_lh_neutral_value);
4303 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4304 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4305 // invert the test if we are looking for a non-array
4306 if (not_array) btest = BoolTest(btest).negate();
4307 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4308 Node* ctrl = generate_fair_guard(bol, region);
4309 Node* is_array_ctrl = not_array ? control() : ctrl;
4310 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4311 // Keep track of the fact that 'obj' is an array to prevent
4312 // array specific accesses from floating above the guard.
4313 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4314 }
4315 return ctrl;
4316 }
4317
4318
4319 //-----------------------inline_native_newArray--------------------------
4320 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4321 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4322 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4323 Node* mirror;
4324 Node* count_val;
4325 if (uninitialized) {
4326 null_check_receiver();
4327 mirror = argument(1);
4328 count_val = argument(2);
4329 } else {
4330 mirror = argument(0);
4331 count_val = argument(1);
4332 }
4333
4334 mirror = null_check(mirror);
4335 // If mirror or obj is dead, only null-path is taken.
4336 if (stopped()) return true;
4337
4338 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4339 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4340 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4341 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4342 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4343
4344 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4345 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4346 result_reg, _slow_path);
4347 Node* normal_ctl = control();
4348 Node* no_array_ctl = result_reg->in(_slow_path);
4349
4350 // Generate code for the slow case. We make a call to newArray().
4351 set_control(no_array_ctl);
4352 if (!stopped()) {
4353 // Either the input type is void.class, or else the
4354 // array klass has not yet been cached. Either the
4355 // ensuing call will throw an exception, or else it
4356 // will cache the array klass for next time.
4357 PreserveJVMState pjvms(this);
4358 CallJavaNode* slow_call = nullptr;
4359 if (uninitialized) {
4360 // Generate optimized virtual call (holder class 'Unsafe' is final)
4361 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4362 } else {
4363 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4364 }
4365 Node* slow_result = set_results_for_java_call(slow_call);
4366 // this->control() comes from set_results_for_java_call
4367 result_reg->set_req(_slow_path, control());
4368 result_val->set_req(_slow_path, slow_result);
4369 result_io ->set_req(_slow_path, i_o());
4370 result_mem->set_req(_slow_path, reset_memory());
4371 }
4372
4373 set_control(normal_ctl);
4374 if (!stopped()) {
4375 // Normal case: The array type has been cached in the java.lang.Class.
4376 // The following call works fine even if the array type is polymorphic.
4377 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4378 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4379 result_reg->init_req(_normal_path, control());
4380 result_val->init_req(_normal_path, obj);
4381 result_io ->init_req(_normal_path, i_o());
4382 result_mem->init_req(_normal_path, reset_memory());
4383
4384 if (uninitialized) {
4385 // Mark the allocation so that zeroing is skipped
4386 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4387 alloc->maybe_set_complete(&_gvn);
4388 }
4389 }
4390
4391 // Return the combined state.
4392 set_i_o( _gvn.transform(result_io) );
4393 set_all_memory( _gvn.transform(result_mem));
4394
4395 C->set_has_split_ifs(true); // Has chance for split-if optimization
4396 set_result(result_reg, result_val);
4397 return true;
4398 }
4399
4400 //----------------------inline_native_getLength--------------------------
4401 // public static native int java.lang.reflect.Array.getLength(Object array);
4402 bool LibraryCallKit::inline_native_getLength() {
4403 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4404
4405 Node* array = null_check(argument(0));
4406 // If array is dead, only null-path is taken.
4407 if (stopped()) return true;
4408
4409 // Deoptimize if it is a non-array.
4410 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4411
4412 if (non_array != nullptr) {
4413 PreserveJVMState pjvms(this);
4414 set_control(non_array);
4415 uncommon_trap(Deoptimization::Reason_intrinsic,
4416 Deoptimization::Action_maybe_recompile);
4417 }
4418
4419 // If control is dead, only non-array-path is taken.
4420 if (stopped()) return true;
4421
4422 // The works fine even if the array type is polymorphic.
4423 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4424 Node* result = load_array_length(array);
4425
4426 C->set_has_split_ifs(true); // Has chance for split-if optimization
4427 set_result(result);
4428 return true;
4429 }
4430
4431 //------------------------inline_array_copyOf----------------------------
4432 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4433 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4434 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4435 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4436
4437 // Get the arguments.
4438 Node* original = argument(0);
4439 Node* start = is_copyOfRange? argument(1): intcon(0);
4440 Node* end = is_copyOfRange? argument(2): argument(1);
4441 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4442
4443 Node* newcopy = nullptr;
4444
4445 // Set the original stack and the reexecute bit for the interpreter to reexecute
4446 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4447 { PreserveReexecuteState preexecs(this);
4448 jvms()->set_should_reexecute(true);
4449
4450 array_type_mirror = null_check(array_type_mirror);
4451 original = null_check(original);
4452
4453 // Check if a null path was taken unconditionally.
4454 if (stopped()) return true;
4455
4456 Node* orig_length = load_array_length(original);
4457
4458 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4459 klass_node = null_check(klass_node);
4460
4461 RegionNode* bailout = new RegionNode(1);
4462 record_for_igvn(bailout);
4463
4464 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4465 // Bail out if that is so.
4466 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4467 if (not_objArray != nullptr) {
4468 // Improve the klass node's type from the new optimistic assumption:
4469 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4470 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4471 Node* cast = new CastPPNode(control(), klass_node, akls);
4472 klass_node = _gvn.transform(cast);
4473 }
4474
4475 // Bail out if either start or end is negative.
4476 generate_negative_guard(start, bailout, &start);
4477 generate_negative_guard(end, bailout, &end);
4478
4479 Node* length = end;
4480 if (_gvn.type(start) != TypeInt::ZERO) {
4481 length = _gvn.transform(new SubINode(end, start));
4482 }
4483
4484 // Bail out if length is negative (i.e., if start > end).
4485 // Without this the new_array would throw
4486 // NegativeArraySizeException but IllegalArgumentException is what
4487 // should be thrown
4488 generate_negative_guard(length, bailout, &length);
4489
4490 // Bail out if start is larger than the original length
4491 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4492 generate_negative_guard(orig_tail, bailout, &orig_tail);
4493
4494 if (bailout->req() > 1) {
4495 PreserveJVMState pjvms(this);
4496 set_control(_gvn.transform(bailout));
4497 uncommon_trap(Deoptimization::Reason_intrinsic,
4498 Deoptimization::Action_maybe_recompile);
4499 }
4500
4501 if (!stopped()) {
4502 // How many elements will we copy from the original?
4503 // The answer is MinI(orig_tail, length).
4504 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4505
4506 // Generate a direct call to the right arraycopy function(s).
4507 // We know the copy is disjoint but we might not know if the
4508 // oop stores need checking.
4509 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4510 // This will fail a store-check if x contains any non-nulls.
4511
4512 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4513 // loads/stores but it is legal only if we're sure the
4514 // Arrays.copyOf would succeed. So we need all input arguments
4515 // to the copyOf to be validated, including that the copy to the
4516 // new array won't trigger an ArrayStoreException. That subtype
4517 // check can be optimized if we know something on the type of
4518 // the input array from type speculation.
4519 if (_gvn.type(klass_node)->singleton()) {
4520 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4521 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4522
4523 int test = C->static_subtype_check(superk, subk);
4524 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4525 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4526 if (t_original->speculative_type() != nullptr) {
4527 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4528 }
4529 }
4530 }
4531
4532 bool validated = false;
4533 // Reason_class_check rather than Reason_intrinsic because we
4534 // want to intrinsify even if this traps.
4535 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4536 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4537
4538 if (not_subtype_ctrl != top()) {
4539 PreserveJVMState pjvms(this);
4540 set_control(not_subtype_ctrl);
4541 uncommon_trap(Deoptimization::Reason_class_check,
4542 Deoptimization::Action_make_not_entrant);
4543 assert(stopped(), "Should be stopped");
4544 }
4545 validated = true;
4546 }
4547
4548 if (!stopped()) {
4549 newcopy = new_array(klass_node, length, 0); // no arguments to push
4550
4551 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4552 load_object_klass(original), klass_node);
4553 if (!is_copyOfRange) {
4554 ac->set_copyof(validated);
4555 } else {
4556 ac->set_copyofrange(validated);
4557 }
4558 Node* n = _gvn.transform(ac);
4559 if (n == ac) {
4560 ac->connect_outputs(this);
4561 } else {
4562 assert(validated, "shouldn't transform if all arguments not validated");
4563 set_all_memory(n);
4564 }
4565 }
4566 }
4567 } // original reexecute is set back here
4568
4569 C->set_has_split_ifs(true); // Has chance for split-if optimization
4570 if (!stopped()) {
4571 set_result(newcopy);
4572 }
4573 return true;
4574 }
4575
4576
4577 //----------------------generate_virtual_guard---------------------------
4578 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4579 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4580 RegionNode* slow_region) {
4581 ciMethod* method = callee();
4582 int vtable_index = method->vtable_index();
4583 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4584 "bad index %d", vtable_index);
4585 // Get the Method* out of the appropriate vtable entry.
4586 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4587 vtable_index*vtableEntry::size_in_bytes() +
4588 in_bytes(vtableEntry::method_offset());
4589 Node* entry_addr = off_heap_plus_addr(obj_klass, entry_offset);
4590 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4591
4592 // Compare the target method with the expected method (e.g., Object.hashCode).
4593 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4594
4595 Node* native_call = makecon(native_call_addr);
4596 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4597 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4598
4599 return generate_slow_guard(test_native, slow_region);
4600 }
4601
4602 //-----------------------generate_method_call----------------------------
4603 // Use generate_method_call to make a slow-call to the real
4604 // method if the fast path fails. An alternative would be to
4605 // use a stub like OptoRuntime::slow_arraycopy_Java.
4606 // This only works for expanding the current library call,
4607 // not another intrinsic. (E.g., don't use this for making an
4608 // arraycopy call inside of the copyOf intrinsic.)
4609 CallJavaNode*
4610 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4611 // When compiling the intrinsic method itself, do not use this technique.
4612 guarantee(callee() != C->method(), "cannot make slow-call to self");
4613
4614 ciMethod* method = callee();
4615 // ensure the JVMS we have will be correct for this call
4616 guarantee(method_id == method->intrinsic_id(), "must match");
4617
4618 const TypeFunc* tf = TypeFunc::make(method);
4619 if (res_not_null) {
4620 assert(tf->return_type() == T_OBJECT, "");
4621 const TypeTuple* range = tf->range();
4622 const Type** fields = TypeTuple::fields(range->cnt());
4623 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4624 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4625 tf = TypeFunc::make(tf->domain(), new_range);
4626 }
4627 CallJavaNode* slow_call;
4628 if (is_static) {
4629 assert(!is_virtual, "");
4630 slow_call = new CallStaticJavaNode(C, tf,
4631 SharedRuntime::get_resolve_static_call_stub(), method);
4632 } else if (is_virtual) {
4633 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4634 int vtable_index = Method::invalid_vtable_index;
4635 if (UseInlineCaches) {
4636 // Suppress the vtable call
4637 } else {
4638 // hashCode and clone are not a miranda methods,
4639 // so the vtable index is fixed.
4640 // No need to use the linkResolver to get it.
4641 vtable_index = method->vtable_index();
4642 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4643 "bad index %d", vtable_index);
4644 }
4645 slow_call = new CallDynamicJavaNode(tf,
4646 SharedRuntime::get_resolve_virtual_call_stub(),
4647 method, vtable_index);
4648 } else { // neither virtual nor static: opt_virtual
4649 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4650 slow_call = new CallStaticJavaNode(C, tf,
4651 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4652 slow_call->set_optimized_virtual(true);
4653 }
4654 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4655 // To be able to issue a direct call (optimized virtual or virtual)
4656 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4657 // about the method being invoked should be attached to the call site to
4658 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4659 slow_call->set_override_symbolic_info(true);
4660 }
4661 set_arguments_for_java_call(slow_call);
4662 set_edges_for_java_call(slow_call);
4663 return slow_call;
4664 }
4665
4666
4667 /**
4668 * Build special case code for calls to hashCode on an object. This call may
4669 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4670 * slightly different code.
4671 */
4672 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4673 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4674 assert(!(is_virtual && is_static), "either virtual, special, or static");
4675
4676 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4677
4678 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4679 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4680 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4681 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4682 Node* obj = nullptr;
4683 if (!is_static) {
4684 // Check for hashing null object
4685 obj = null_check_receiver();
4686 if (stopped()) return true; // unconditionally null
4687 result_reg->init_req(_null_path, top());
4688 result_val->init_req(_null_path, top());
4689 } else {
4690 // Do a null check, and return zero if null.
4691 // System.identityHashCode(null) == 0
4692 obj = argument(0);
4693 Node* null_ctl = top();
4694 obj = null_check_oop(obj, &null_ctl);
4695 result_reg->init_req(_null_path, null_ctl);
4696 result_val->init_req(_null_path, _gvn.intcon(0));
4697 }
4698
4699 // Unconditionally null? Then return right away.
4700 if (stopped()) {
4701 set_control( result_reg->in(_null_path));
4702 if (!stopped())
4703 set_result(result_val->in(_null_path));
4704 return true;
4705 }
4706
4707 // We only go to the fast case code if we pass a number of guards. The
4708 // paths which do not pass are accumulated in the slow_region.
4709 RegionNode* slow_region = new RegionNode(1);
4710 record_for_igvn(slow_region);
4711
4712 // If this is a virtual call, we generate a funny guard. We pull out
4713 // the vtable entry corresponding to hashCode() from the target object.
4714 // If the target method which we are calling happens to be the native
4715 // Object hashCode() method, we pass the guard. We do not need this
4716 // guard for non-virtual calls -- the caller is known to be the native
4717 // Object hashCode().
4718 if (is_virtual) {
4719 // After null check, get the object's klass.
4720 Node* obj_klass = load_object_klass(obj);
4721 generate_virtual_guard(obj_klass, slow_region);
4722 }
4723
4724 // Get the header out of the object, use LoadMarkNode when available
4725 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4726 // The control of the load must be null. Otherwise, the load can move before
4727 // the null check after castPP removal.
4728 Node* no_ctrl = nullptr;
4729 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4730
4731 if (!UseObjectMonitorTable) {
4732 // Test the header to see if it is safe to read w.r.t. locking.
4733 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4734 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4735 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4736 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4737 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4738
4739 generate_slow_guard(test_monitor, slow_region);
4740 }
4741
4742 // Get the hash value and check to see that it has been properly assigned.
4743 // We depend on hash_mask being at most 32 bits and avoid the use of
4744 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4745 // vm: see markWord.hpp.
4746 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4747 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4748 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4749 // This hack lets the hash bits live anywhere in the mark object now, as long
4750 // as the shift drops the relevant bits into the low 32 bits. Note that
4751 // Java spec says that HashCode is an int so there's no point in capturing
4752 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4753 hshifted_header = ConvX2I(hshifted_header);
4754 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4755
4756 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4757 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4758 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4759
4760 generate_slow_guard(test_assigned, slow_region);
4761
4762 Node* init_mem = reset_memory();
4763 // fill in the rest of the null path:
4764 result_io ->init_req(_null_path, i_o());
4765 result_mem->init_req(_null_path, init_mem);
4766
4767 result_val->init_req(_fast_path, hash_val);
4768 result_reg->init_req(_fast_path, control());
4769 result_io ->init_req(_fast_path, i_o());
4770 result_mem->init_req(_fast_path, init_mem);
4771
4772 // Generate code for the slow case. We make a call to hashCode().
4773 set_control(_gvn.transform(slow_region));
4774 if (!stopped()) {
4775 // No need for PreserveJVMState, because we're using up the present state.
4776 set_all_memory(init_mem);
4777 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4778 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4779 Node* slow_result = set_results_for_java_call(slow_call);
4780 // this->control() comes from set_results_for_java_call
4781 result_reg->init_req(_slow_path, control());
4782 result_val->init_req(_slow_path, slow_result);
4783 result_io ->set_req(_slow_path, i_o());
4784 result_mem ->set_req(_slow_path, reset_memory());
4785 }
4786
4787 // Return the combined state.
4788 set_i_o( _gvn.transform(result_io) );
4789 set_all_memory( _gvn.transform(result_mem));
4790
4791 set_result(result_reg, result_val);
4792 return true;
4793 }
4794
4795 //---------------------------inline_native_getClass----------------------------
4796 // public final native Class<?> java.lang.Object.getClass();
4797 //
4798 // Build special case code for calls to getClass on an object.
4799 bool LibraryCallKit::inline_native_getClass() {
4800 Node* obj = null_check_receiver();
4801 if (stopped()) return true;
4802 set_result(load_mirror_from_klass(load_object_klass(obj)));
4803 return true;
4804 }
4805
4806 //-----------------inline_native_Reflection_getCallerClass---------------------
4807 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4808 //
4809 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4810 //
4811 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4812 // in that it must skip particular security frames and checks for
4813 // caller sensitive methods.
4814 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4815 #ifndef PRODUCT
4816 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4817 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4818 }
4819 #endif
4820
4821 if (!jvms()->has_method()) {
4822 #ifndef PRODUCT
4823 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4824 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4825 }
4826 #endif
4827 return false;
4828 }
4829
4830 // Walk back up the JVM state to find the caller at the required
4831 // depth.
4832 JVMState* caller_jvms = jvms();
4833
4834 // Cf. JVM_GetCallerClass
4835 // NOTE: Start the loop at depth 1 because the current JVM state does
4836 // not include the Reflection.getCallerClass() frame.
4837 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4838 ciMethod* m = caller_jvms->method();
4839 switch (n) {
4840 case 0:
4841 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4842 break;
4843 case 1:
4844 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4845 if (!m->caller_sensitive()) {
4846 #ifndef PRODUCT
4847 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4848 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4849 }
4850 #endif
4851 return false; // bail-out; let JVM_GetCallerClass do the work
4852 }
4853 break;
4854 default:
4855 if (!m->is_ignored_by_security_stack_walk()) {
4856 // We have reached the desired frame; return the holder class.
4857 // Acquire method holder as java.lang.Class and push as constant.
4858 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4859 ciInstance* caller_mirror = caller_klass->java_mirror();
4860 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4861
4862 #ifndef PRODUCT
4863 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4864 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());
4865 tty->print_cr(" JVM state at this point:");
4866 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4867 ciMethod* m = jvms()->of_depth(i)->method();
4868 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4869 }
4870 }
4871 #endif
4872 return true;
4873 }
4874 break;
4875 }
4876 }
4877
4878 #ifndef PRODUCT
4879 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4880 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4881 tty->print_cr(" JVM state at this point:");
4882 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4883 ciMethod* m = jvms()->of_depth(i)->method();
4884 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4885 }
4886 }
4887 #endif
4888
4889 return false; // bail-out; let JVM_GetCallerClass do the work
4890 }
4891
4892 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4893 Node* arg = argument(0);
4894 Node* result = nullptr;
4895
4896 switch (id) {
4897 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4898 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4899 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4900 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4901 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4902 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4903
4904 case vmIntrinsics::_doubleToLongBits: {
4905 // two paths (plus control) merge in a wood
4906 RegionNode *r = new RegionNode(3);
4907 Node *phi = new PhiNode(r, TypeLong::LONG);
4908
4909 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4910 // Build the boolean node
4911 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4912
4913 // Branch either way.
4914 // NaN case is less traveled, which makes all the difference.
4915 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4916 Node *opt_isnan = _gvn.transform(ifisnan);
4917 assert( opt_isnan->is_If(), "Expect an IfNode");
4918 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4919 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4920
4921 set_control(iftrue);
4922
4923 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4924 Node *slow_result = longcon(nan_bits); // return NaN
4925 phi->init_req(1, _gvn.transform( slow_result ));
4926 r->init_req(1, iftrue);
4927
4928 // Else fall through
4929 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4930 set_control(iffalse);
4931
4932 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4933 r->init_req(2, iffalse);
4934
4935 // Post merge
4936 set_control(_gvn.transform(r));
4937 record_for_igvn(r);
4938
4939 C->set_has_split_ifs(true); // Has chance for split-if optimization
4940 result = phi;
4941 assert(result->bottom_type()->isa_long(), "must be");
4942 break;
4943 }
4944
4945 case vmIntrinsics::_floatToIntBits: {
4946 // two paths (plus control) merge in a wood
4947 RegionNode *r = new RegionNode(3);
4948 Node *phi = new PhiNode(r, TypeInt::INT);
4949
4950 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4951 // Build the boolean node
4952 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4953
4954 // Branch either way.
4955 // NaN case is less traveled, which makes all the difference.
4956 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4957 Node *opt_isnan = _gvn.transform(ifisnan);
4958 assert( opt_isnan->is_If(), "Expect an IfNode");
4959 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4960 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4961
4962 set_control(iftrue);
4963
4964 static const jint nan_bits = 0x7fc00000;
4965 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4966 phi->init_req(1, _gvn.transform( slow_result ));
4967 r->init_req(1, iftrue);
4968
4969 // Else fall through
4970 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4971 set_control(iffalse);
4972
4973 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4974 r->init_req(2, iffalse);
4975
4976 // Post merge
4977 set_control(_gvn.transform(r));
4978 record_for_igvn(r);
4979
4980 C->set_has_split_ifs(true); // Has chance for split-if optimization
4981 result = phi;
4982 assert(result->bottom_type()->isa_int(), "must be");
4983 break;
4984 }
4985
4986 default:
4987 fatal_unexpected_iid(id);
4988 break;
4989 }
4990 set_result(_gvn.transform(result));
4991 return true;
4992 }
4993
4994 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4995 Node* arg = argument(0);
4996 Node* result = nullptr;
4997
4998 switch (id) {
4999 case vmIntrinsics::_floatIsInfinite:
5000 result = new IsInfiniteFNode(arg);
5001 break;
5002 case vmIntrinsics::_floatIsFinite:
5003 result = new IsFiniteFNode(arg);
5004 break;
5005 case vmIntrinsics::_doubleIsInfinite:
5006 result = new IsInfiniteDNode(arg);
5007 break;
5008 case vmIntrinsics::_doubleIsFinite:
5009 result = new IsFiniteDNode(arg);
5010 break;
5011 default:
5012 fatal_unexpected_iid(id);
5013 break;
5014 }
5015 set_result(_gvn.transform(result));
5016 return true;
5017 }
5018
5019 //----------------------inline_unsafe_copyMemory-------------------------
5020 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5021
5022 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5023 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5024 const Type* base_t = gvn.type(base);
5025
5026 bool in_native = (base_t == TypePtr::NULL_PTR);
5027 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5028 bool is_mixed = !in_heap && !in_native;
5029
5030 if (is_mixed) {
5031 return true; // mixed accesses can touch both on-heap and off-heap memory
5032 }
5033 if (in_heap) {
5034 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5035 if (!is_prim_array) {
5036 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5037 // there's not enough type information available to determine proper memory slice for it.
5038 return true;
5039 }
5040 }
5041 return false;
5042 }
5043
5044 bool LibraryCallKit::inline_unsafe_copyMemory() {
5045 if (callee()->is_static()) return false; // caller must have the capability!
5046 null_check_receiver(); // null-check receiver
5047 if (stopped()) return true;
5048
5049 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5050
5051 Node* src_base = argument(1); // type: oop
5052 Node* src_off = ConvL2X(argument(2)); // type: long
5053 Node* dst_base = argument(4); // type: oop
5054 Node* dst_off = ConvL2X(argument(5)); // type: long
5055 Node* size = ConvL2X(argument(7)); // type: long
5056
5057 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5058 "fieldOffset must be byte-scaled");
5059
5060 Node* src_addr = make_unsafe_address(src_base, src_off);
5061 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5062
5063 Node* thread = _gvn.transform(new ThreadLocalNode());
5064 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5065 BasicType doing_unsafe_access_bt = T_BYTE;
5066 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5067
5068 // update volatile field
5069 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5070
5071 int flags = RC_LEAF | RC_NO_FP;
5072
5073 const TypePtr* dst_type = TypePtr::BOTTOM;
5074
5075 // Adjust memory effects of the runtime call based on input values.
5076 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5077 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5078 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5079
5080 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5081 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5082 flags |= RC_NARROW_MEM; // narrow in memory
5083 }
5084 }
5085
5086 // Call it. Note that the length argument is not scaled.
5087 make_runtime_call(flags,
5088 OptoRuntime::fast_arraycopy_Type(),
5089 StubRoutines::unsafe_arraycopy(),
5090 "unsafe_arraycopy",
5091 dst_type,
5092 src_addr, dst_addr, size XTOP);
5093
5094 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5095
5096 return true;
5097 }
5098
5099 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5100 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5101 bool LibraryCallKit::inline_unsafe_setMemory() {
5102 if (callee()->is_static()) return false; // caller must have the capability!
5103 null_check_receiver(); // null-check receiver
5104 if (stopped()) return true;
5105
5106 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5107
5108 Node* dst_base = argument(1); // type: oop
5109 Node* dst_off = ConvL2X(argument(2)); // type: long
5110 Node* size = ConvL2X(argument(4)); // type: long
5111 Node* byte = argument(6); // type: byte
5112
5113 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5114 "fieldOffset must be byte-scaled");
5115
5116 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5117
5118 Node* thread = _gvn.transform(new ThreadLocalNode());
5119 Node* doing_unsafe_access_addr = off_heap_plus_addr(thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5120 BasicType doing_unsafe_access_bt = T_BYTE;
5121 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5122
5123 // update volatile field
5124 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5125
5126 int flags = RC_LEAF | RC_NO_FP;
5127
5128 const TypePtr* dst_type = TypePtr::BOTTOM;
5129
5130 // Adjust memory effects of the runtime call based on input values.
5131 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5132 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5133
5134 flags |= RC_NARROW_MEM; // narrow in memory
5135 }
5136
5137 // Call it. Note that the length argument is not scaled.
5138 make_runtime_call(flags,
5139 OptoRuntime::unsafe_setmemory_Type(),
5140 StubRoutines::unsafe_setmemory(),
5141 "unsafe_setmemory",
5142 dst_type,
5143 dst_addr, size XTOP, byte);
5144
5145 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5146
5147 return true;
5148 }
5149
5150 #undef XTOP
5151
5152 //------------------------clone_coping-----------------------------------
5153 // Helper function for inline_native_clone.
5154 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5155 assert(obj_size != nullptr, "");
5156 Node* raw_obj = alloc_obj->in(1);
5157 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5158
5159 AllocateNode* alloc = nullptr;
5160 if (ReduceBulkZeroing &&
5161 // If we are implementing an array clone without knowing its source type
5162 // (can happen when compiling the array-guarded branch of a reflective
5163 // Object.clone() invocation), initialize the array within the allocation.
5164 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5165 // to a runtime clone call that assumes fully initialized source arrays.
5166 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5167 // We will be completely responsible for initializing this object -
5168 // mark Initialize node as complete.
5169 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5170 // The object was just allocated - there should be no any stores!
5171 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5172 // Mark as complete_with_arraycopy so that on AllocateNode
5173 // expansion, we know this AllocateNode is initialized by an array
5174 // copy and a StoreStore barrier exists after the array copy.
5175 alloc->initialization()->set_complete_with_arraycopy();
5176 }
5177
5178 Node* size = _gvn.transform(obj_size);
5179 access_clone(obj, alloc_obj, size, is_array);
5180
5181 // Do not let reads from the cloned object float above the arraycopy.
5182 if (alloc != nullptr) {
5183 // Do not let stores that initialize this object be reordered with
5184 // a subsequent store that would make this object accessible by
5185 // other threads.
5186 // Record what AllocateNode this StoreStore protects so that
5187 // escape analysis can go from the MemBarStoreStoreNode to the
5188 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5189 // based on the escape status of the AllocateNode.
5190 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5191 } else {
5192 insert_mem_bar(Op_MemBarCPUOrder);
5193 }
5194 }
5195
5196 //------------------------inline_native_clone----------------------------
5197 // protected native Object java.lang.Object.clone();
5198 //
5199 // Here are the simple edge cases:
5200 // null receiver => normal trap
5201 // virtual and clone was overridden => slow path to out-of-line clone
5202 // not cloneable or finalizer => slow path to out-of-line Object.clone
5203 //
5204 // The general case has two steps, allocation and copying.
5205 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5206 //
5207 // Copying also has two cases, oop arrays and everything else.
5208 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5209 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5210 //
5211 // These steps fold up nicely if and when the cloned object's klass
5212 // can be sharply typed as an object array, a type array, or an instance.
5213 //
5214 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5215 PhiNode* result_val;
5216
5217 // Set the reexecute bit for the interpreter to reexecute
5218 // the bytecode that invokes Object.clone if deoptimization happens.
5219 { PreserveReexecuteState preexecs(this);
5220 jvms()->set_should_reexecute(true);
5221
5222 Node* obj = null_check_receiver();
5223 if (stopped()) return true;
5224
5225 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5226
5227 // If we are going to clone an instance, we need its exact type to
5228 // know the number and types of fields to convert the clone to
5229 // loads/stores. Maybe a speculative type can help us.
5230 if (!obj_type->klass_is_exact() &&
5231 obj_type->speculative_type() != nullptr &&
5232 obj_type->speculative_type()->is_instance_klass()) {
5233 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5234 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5235 !spec_ik->has_injected_fields()) {
5236 if (!obj_type->isa_instptr() ||
5237 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5238 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5239 }
5240 }
5241 }
5242
5243 // Conservatively insert a memory barrier on all memory slices.
5244 // Do not let writes into the original float below the clone.
5245 insert_mem_bar(Op_MemBarCPUOrder);
5246
5247 // paths into result_reg:
5248 enum {
5249 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5250 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5251 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5252 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5253 PATH_LIMIT
5254 };
5255 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5256 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5257 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5258 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5259 record_for_igvn(result_reg);
5260
5261 Node* obj_klass = load_object_klass(obj);
5262 Node* array_obj = obj;
5263 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5264 if (array_ctl != nullptr) {
5265 // It's an array.
5266 PreserveJVMState pjvms(this);
5267 set_control(array_ctl);
5268 Node* obj_length = load_array_length(array_obj);
5269 Node* array_size = nullptr; // Size of the array without object alignment padding.
5270 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5271
5272 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5273 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5274 // If it is an oop array, it requires very special treatment,
5275 // because gc barriers are required when accessing the array.
5276 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5277 if (is_obja != nullptr) {
5278 PreserveJVMState pjvms2(this);
5279 set_control(is_obja);
5280 // Generate a direct call to the right arraycopy function(s).
5281 // Clones are always tightly coupled.
5282 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5283 ac->set_clone_oop_array();
5284 Node* n = _gvn.transform(ac);
5285 assert(n == ac, "cannot disappear");
5286 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5287
5288 result_reg->init_req(_objArray_path, control());
5289 result_val->init_req(_objArray_path, alloc_obj);
5290 result_i_o ->set_req(_objArray_path, i_o());
5291 result_mem ->set_req(_objArray_path, reset_memory());
5292 }
5293 }
5294 // Otherwise, there are no barriers to worry about.
5295 // (We can dispense with card marks if we know the allocation
5296 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5297 // causes the non-eden paths to take compensating steps to
5298 // simulate a fresh allocation, so that no further
5299 // card marks are required in compiled code to initialize
5300 // the object.)
5301
5302 if (!stopped()) {
5303 copy_to_clone(array_obj, alloc_obj, array_size, true);
5304
5305 // Present the results of the copy.
5306 result_reg->init_req(_array_path, control());
5307 result_val->init_req(_array_path, alloc_obj);
5308 result_i_o ->set_req(_array_path, i_o());
5309 result_mem ->set_req(_array_path, reset_memory());
5310 }
5311 }
5312
5313 // We only go to the instance fast case code if we pass a number of guards.
5314 // The paths which do not pass are accumulated in the slow_region.
5315 RegionNode* slow_region = new RegionNode(1);
5316 record_for_igvn(slow_region);
5317 if (!stopped()) {
5318 // It's an instance (we did array above). Make the slow-path tests.
5319 // If this is a virtual call, we generate a funny guard. We grab
5320 // the vtable entry corresponding to clone() from the target object.
5321 // If the target method which we are calling happens to be the
5322 // Object clone() method, we pass the guard. We do not need this
5323 // guard for non-virtual calls; the caller is known to be the native
5324 // Object clone().
5325 if (is_virtual) {
5326 generate_virtual_guard(obj_klass, slow_region);
5327 }
5328
5329 // The object must be easily cloneable and must not have a finalizer.
5330 // Both of these conditions may be checked in a single test.
5331 // We could optimize the test further, but we don't care.
5332 generate_misc_flags_guard(obj_klass,
5333 // Test both conditions:
5334 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5335 // Must be cloneable but not finalizer:
5336 KlassFlags::_misc_is_cloneable_fast,
5337 slow_region);
5338 }
5339
5340 if (!stopped()) {
5341 // It's an instance, and it passed the slow-path tests.
5342 PreserveJVMState pjvms(this);
5343 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5344 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5345 // is reexecuted if deoptimization occurs and there could be problems when merging
5346 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5347 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5348
5349 copy_to_clone(obj, alloc_obj, obj_size, false);
5350
5351 // Present the results of the slow call.
5352 result_reg->init_req(_instance_path, control());
5353 result_val->init_req(_instance_path, alloc_obj);
5354 result_i_o ->set_req(_instance_path, i_o());
5355 result_mem ->set_req(_instance_path, reset_memory());
5356 }
5357
5358 // Generate code for the slow case. We make a call to clone().
5359 set_control(_gvn.transform(slow_region));
5360 if (!stopped()) {
5361 PreserveJVMState pjvms(this);
5362 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5363 // We need to deoptimize on exception (see comment above)
5364 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5365 // this->control() comes from set_results_for_java_call
5366 result_reg->init_req(_slow_path, control());
5367 result_val->init_req(_slow_path, slow_result);
5368 result_i_o ->set_req(_slow_path, i_o());
5369 result_mem ->set_req(_slow_path, reset_memory());
5370 }
5371
5372 // Return the combined state.
5373 set_control( _gvn.transform(result_reg));
5374 set_i_o( _gvn.transform(result_i_o));
5375 set_all_memory( _gvn.transform(result_mem));
5376 } // original reexecute is set back here
5377
5378 set_result(_gvn.transform(result_val));
5379 return true;
5380 }
5381
5382 // If we have a tightly coupled allocation, the arraycopy may take care
5383 // of the array initialization. If one of the guards we insert between
5384 // the allocation and the arraycopy causes a deoptimization, an
5385 // uninitialized array will escape the compiled method. To prevent that
5386 // we set the JVM state for uncommon traps between the allocation and
5387 // the arraycopy to the state before the allocation so, in case of
5388 // deoptimization, we'll reexecute the allocation and the
5389 // initialization.
5390 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5391 if (alloc != nullptr) {
5392 ciMethod* trap_method = alloc->jvms()->method();
5393 int trap_bci = alloc->jvms()->bci();
5394
5395 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5396 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5397 // Make sure there's no store between the allocation and the
5398 // arraycopy otherwise visible side effects could be rexecuted
5399 // in case of deoptimization and cause incorrect execution.
5400 bool no_interfering_store = true;
5401 Node* mem = alloc->in(TypeFunc::Memory);
5402 if (mem->is_MergeMem()) {
5403 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5404 Node* n = mms.memory();
5405 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5406 assert(n->is_Store(), "what else?");
5407 no_interfering_store = false;
5408 break;
5409 }
5410 }
5411 } else {
5412 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5413 Node* n = mms.memory();
5414 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5415 assert(n->is_Store(), "what else?");
5416 no_interfering_store = false;
5417 break;
5418 }
5419 }
5420 }
5421
5422 if (no_interfering_store) {
5423 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5424
5425 JVMState* saved_jvms = jvms();
5426 saved_reexecute_sp = _reexecute_sp;
5427
5428 set_jvms(sfpt->jvms());
5429 _reexecute_sp = jvms()->sp();
5430
5431 return saved_jvms;
5432 }
5433 }
5434 }
5435 return nullptr;
5436 }
5437
5438 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5439 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5440 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5441 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5442 uint size = alloc->req();
5443 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5444 old_jvms->set_map(sfpt);
5445 for (uint i = 0; i < size; i++) {
5446 sfpt->init_req(i, alloc->in(i));
5447 }
5448 // re-push array length for deoptimization
5449 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5450 old_jvms->set_sp(old_jvms->sp()+1);
5451 old_jvms->set_monoff(old_jvms->monoff()+1);
5452 old_jvms->set_scloff(old_jvms->scloff()+1);
5453 old_jvms->set_endoff(old_jvms->endoff()+1);
5454 old_jvms->set_should_reexecute(true);
5455
5456 sfpt->set_i_o(map()->i_o());
5457 sfpt->set_memory(map()->memory());
5458 sfpt->set_control(map()->control());
5459 return sfpt;
5460 }
5461
5462 // In case of a deoptimization, we restart execution at the
5463 // allocation, allocating a new array. We would leave an uninitialized
5464 // array in the heap that GCs wouldn't expect. Move the allocation
5465 // after the traps so we don't allocate the array if we
5466 // deoptimize. This is possible because tightly_coupled_allocation()
5467 // guarantees there's no observer of the allocated array at this point
5468 // and the control flow is simple enough.
5469 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5470 int saved_reexecute_sp, uint new_idx) {
5471 if (saved_jvms_before_guards != nullptr && !stopped()) {
5472 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5473
5474 assert(alloc != nullptr, "only with a tightly coupled allocation");
5475 // restore JVM state to the state at the arraycopy
5476 saved_jvms_before_guards->map()->set_control(map()->control());
5477 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5478 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5479 // If we've improved the types of some nodes (null check) while
5480 // emitting the guards, propagate them to the current state
5481 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5482 set_jvms(saved_jvms_before_guards);
5483 _reexecute_sp = saved_reexecute_sp;
5484
5485 // Remove the allocation from above the guards
5486 CallProjections callprojs;
5487 alloc->extract_projections(&callprojs, true);
5488 InitializeNode* init = alloc->initialization();
5489 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5490 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5491 init->replace_mem_projs_by(alloc_mem, C);
5492
5493 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5494 // the allocation (i.e. is only valid if the allocation succeeds):
5495 // 1) replace CastIINode with AllocateArrayNode's length here
5496 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5497 //
5498 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5499 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5500 Node* init_control = init->proj_out(TypeFunc::Control);
5501 Node* alloc_length = alloc->Ideal_length();
5502 #ifdef ASSERT
5503 Node* prev_cast = nullptr;
5504 #endif
5505 for (uint i = 0; i < init_control->outcnt(); i++) {
5506 Node* init_out = init_control->raw_out(i);
5507 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5508 #ifdef ASSERT
5509 if (prev_cast == nullptr) {
5510 prev_cast = init_out;
5511 } else {
5512 if (prev_cast->cmp(*init_out) == false) {
5513 prev_cast->dump();
5514 init_out->dump();
5515 assert(false, "not equal CastIINode");
5516 }
5517 }
5518 #endif
5519 C->gvn_replace_by(init_out, alloc_length);
5520 }
5521 }
5522 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5523
5524 // move the allocation here (after the guards)
5525 _gvn.hash_delete(alloc);
5526 alloc->set_req(TypeFunc::Control, control());
5527 alloc->set_req(TypeFunc::I_O, i_o());
5528 Node *mem = reset_memory();
5529 set_all_memory(mem);
5530 alloc->set_req(TypeFunc::Memory, mem);
5531 set_control(init->proj_out_or_null(TypeFunc::Control));
5532 set_i_o(callprojs.fallthrough_ioproj);
5533
5534 // Update memory as done in GraphKit::set_output_for_allocation()
5535 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5536 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5537 if (ary_type->isa_aryptr() && length_type != nullptr) {
5538 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5539 }
5540 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5541 int elemidx = C->get_alias_index(telemref);
5542 // Need to properly move every memory projection for the Initialize
5543 #ifdef ASSERT
5544 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5545 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5546 #endif
5547 auto move_proj = [&](ProjNode* proj) {
5548 int alias_idx = C->get_alias_index(proj->adr_type());
5549 assert(alias_idx == Compile::AliasIdxRaw ||
5550 alias_idx == elemidx ||
5551 alias_idx == mark_idx ||
5552 alias_idx == klass_idx, "should be raw memory or array element type");
5553 set_memory(proj, alias_idx);
5554 };
5555 init->for_each_proj(move_proj, TypeFunc::Memory);
5556
5557 Node* allocx = _gvn.transform(alloc);
5558 assert(allocx == alloc, "where has the allocation gone?");
5559 assert(dest->is_CheckCastPP(), "not an allocation result?");
5560
5561 _gvn.hash_delete(dest);
5562 dest->set_req(0, control());
5563 Node* destx = _gvn.transform(dest);
5564 assert(destx == dest, "where has the allocation result gone?");
5565
5566 array_ideal_length(alloc, ary_type, true);
5567 }
5568 }
5569
5570 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5571 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5572 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5573 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5574 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5575 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5576 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5577 JVMState* saved_jvms_before_guards) {
5578 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5579 // There is at least one unrelated uncommon trap which needs to be replaced.
5580 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5581
5582 JVMState* saved_jvms = jvms();
5583 const int saved_reexecute_sp = _reexecute_sp;
5584 set_jvms(sfpt->jvms());
5585 _reexecute_sp = jvms()->sp();
5586
5587 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5588
5589 // Restore state
5590 set_jvms(saved_jvms);
5591 _reexecute_sp = saved_reexecute_sp;
5592 }
5593 }
5594
5595 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5596 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5597 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5598 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5599 while (if_proj->is_IfProj()) {
5600 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5601 if (uncommon_trap != nullptr) {
5602 create_new_uncommon_trap(uncommon_trap);
5603 }
5604 assert(if_proj->in(0)->is_If(), "must be If");
5605 if_proj = if_proj->in(0)->in(0);
5606 }
5607 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5608 "must have reached control projection of init node");
5609 }
5610
5611 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5612 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5613 assert(trap_request != 0, "no valid UCT trap request");
5614 PreserveJVMState pjvms(this);
5615 set_control(uncommon_trap_call->in(0));
5616 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5617 Deoptimization::trap_request_action(trap_request));
5618 assert(stopped(), "Should be stopped");
5619 _gvn.hash_delete(uncommon_trap_call);
5620 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5621 }
5622
5623 // Common checks for array sorting intrinsics arguments.
5624 // Returns `true` if checks passed.
5625 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5626 // check address of the class
5627 if (elementType == nullptr || elementType->is_top()) {
5628 return false; // dead path
5629 }
5630 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5631 if (elem_klass == nullptr) {
5632 return false; // dead path
5633 }
5634 // java_mirror_type() returns non-null for compile-time Class constants only
5635 ciType* elem_type = elem_klass->java_mirror_type();
5636 if (elem_type == nullptr) {
5637 return false;
5638 }
5639 bt = elem_type->basic_type();
5640 // Disable the intrinsic if the CPU does not support SIMD sort
5641 if (!Matcher::supports_simd_sort(bt)) {
5642 return false;
5643 }
5644 // check address of the array
5645 if (obj == nullptr || obj->is_top()) {
5646 return false; // dead path
5647 }
5648 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5649 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5650 return false; // failed input validation
5651 }
5652 return true;
5653 }
5654
5655 //------------------------------inline_array_partition-----------------------
5656 bool LibraryCallKit::inline_array_partition() {
5657 address stubAddr = StubRoutines::select_array_partition_function();
5658 if (stubAddr == nullptr) {
5659 return false; // Intrinsic's stub is not implemented on this platform
5660 }
5661 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5662
5663 // no receiver because it is a static method
5664 Node* elementType = argument(0);
5665 Node* obj = argument(1);
5666 Node* offset = argument(2); // long
5667 Node* fromIndex = argument(4);
5668 Node* toIndex = argument(5);
5669 Node* indexPivot1 = argument(6);
5670 Node* indexPivot2 = argument(7);
5671 // PartitionOperation: argument(8) is ignored
5672
5673 Node* pivotIndices = nullptr;
5674 BasicType bt = T_ILLEGAL;
5675
5676 if (!check_array_sort_arguments(elementType, obj, bt)) {
5677 return false;
5678 }
5679 null_check(obj);
5680 // If obj is dead, only null-path is taken.
5681 if (stopped()) {
5682 return true;
5683 }
5684 // Set the original stack and the reexecute bit for the interpreter to reexecute
5685 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5686 { PreserveReexecuteState preexecs(this);
5687 jvms()->set_should_reexecute(true);
5688
5689 Node* obj_adr = make_unsafe_address(obj, offset);
5690
5691 // create the pivotIndices array of type int and size = 2
5692 Node* size = intcon(2);
5693 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5694 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5695 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5696 guarantee(alloc != nullptr, "created above");
5697 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5698
5699 // pass the basic type enum to the stub
5700 Node* elemType = intcon(bt);
5701
5702 // Call the stub
5703 const char *stubName = "array_partition_stub";
5704 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5705 stubAddr, stubName, TypePtr::BOTTOM,
5706 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5707 indexPivot1, indexPivot2);
5708
5709 } // original reexecute is set back here
5710
5711 if (!stopped()) {
5712 set_result(pivotIndices);
5713 }
5714
5715 return true;
5716 }
5717
5718
5719 //------------------------------inline_array_sort-----------------------
5720 bool LibraryCallKit::inline_array_sort() {
5721 address stubAddr = StubRoutines::select_arraysort_function();
5722 if (stubAddr == nullptr) {
5723 return false; // Intrinsic's stub is not implemented on this platform
5724 }
5725 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5726
5727 // no receiver because it is a static method
5728 Node* elementType = argument(0);
5729 Node* obj = argument(1);
5730 Node* offset = argument(2); // long
5731 Node* fromIndex = argument(4);
5732 Node* toIndex = argument(5);
5733 // SortOperation: argument(6) is ignored
5734
5735 BasicType bt = T_ILLEGAL;
5736
5737 if (!check_array_sort_arguments(elementType, obj, bt)) {
5738 return false;
5739 }
5740 null_check(obj);
5741 // If obj is dead, only null-path is taken.
5742 if (stopped()) {
5743 return true;
5744 }
5745 Node* obj_adr = make_unsafe_address(obj, offset);
5746
5747 // pass the basic type enum to the stub
5748 Node* elemType = intcon(bt);
5749
5750 // Call the stub.
5751 const char *stubName = "arraysort_stub";
5752 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5753 stubAddr, stubName, TypePtr::BOTTOM,
5754 obj_adr, elemType, fromIndex, toIndex);
5755
5756 return true;
5757 }
5758
5759
5760 //------------------------------inline_arraycopy-----------------------
5761 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5762 // Object dest, int destPos,
5763 // int length);
5764 bool LibraryCallKit::inline_arraycopy() {
5765 // Get the arguments.
5766 Node* src = argument(0); // type: oop
5767 Node* src_offset = argument(1); // type: int
5768 Node* dest = argument(2); // type: oop
5769 Node* dest_offset = argument(3); // type: int
5770 Node* length = argument(4); // type: int
5771
5772 uint new_idx = C->unique();
5773
5774 // Check for allocation before we add nodes that would confuse
5775 // tightly_coupled_allocation()
5776 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5777
5778 int saved_reexecute_sp = -1;
5779 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5780 // See arraycopy_restore_alloc_state() comment
5781 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5782 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5783 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5784 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5785
5786 // The following tests must be performed
5787 // (1) src and dest are arrays.
5788 // (2) src and dest arrays must have elements of the same BasicType
5789 // (3) src and dest must not be null.
5790 // (4) src_offset must not be negative.
5791 // (5) dest_offset must not be negative.
5792 // (6) length must not be negative.
5793 // (7) src_offset + length must not exceed length of src.
5794 // (8) dest_offset + length must not exceed length of dest.
5795 // (9) each element of an oop array must be assignable
5796
5797 // (3) src and dest must not be null.
5798 // always do this here because we need the JVM state for uncommon traps
5799 Node* null_ctl = top();
5800 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5801 assert(null_ctl->is_top(), "no null control here");
5802 dest = null_check(dest, T_ARRAY);
5803
5804 if (!can_emit_guards) {
5805 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5806 // guards but the arraycopy node could still take advantage of a
5807 // tightly allocated allocation. tightly_coupled_allocation() is
5808 // called again to make sure it takes the null check above into
5809 // account: the null check is mandatory and if it caused an
5810 // uncommon trap to be emitted then the allocation can't be
5811 // considered tightly coupled in this context.
5812 alloc = tightly_coupled_allocation(dest);
5813 }
5814
5815 bool validated = false;
5816
5817 const Type* src_type = _gvn.type(src);
5818 const Type* dest_type = _gvn.type(dest);
5819 const TypeAryPtr* top_src = src_type->isa_aryptr();
5820 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5821
5822 // Do we have the type of src?
5823 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5824 // Do we have the type of dest?
5825 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5826 // Is the type for src from speculation?
5827 bool src_spec = false;
5828 // Is the type for dest from speculation?
5829 bool dest_spec = false;
5830
5831 if ((!has_src || !has_dest) && can_emit_guards) {
5832 // We don't have sufficient type information, let's see if
5833 // speculative types can help. We need to have types for both src
5834 // and dest so that it pays off.
5835
5836 // Do we already have or could we have type information for src
5837 bool could_have_src = has_src;
5838 // Do we already have or could we have type information for dest
5839 bool could_have_dest = has_dest;
5840
5841 ciKlass* src_k = nullptr;
5842 if (!has_src) {
5843 src_k = src_type->speculative_type_not_null();
5844 if (src_k != nullptr && src_k->is_array_klass()) {
5845 could_have_src = true;
5846 }
5847 }
5848
5849 ciKlass* dest_k = nullptr;
5850 if (!has_dest) {
5851 dest_k = dest_type->speculative_type_not_null();
5852 if (dest_k != nullptr && dest_k->is_array_klass()) {
5853 could_have_dest = true;
5854 }
5855 }
5856
5857 if (could_have_src && could_have_dest) {
5858 // This is going to pay off so emit the required guards
5859 if (!has_src) {
5860 src = maybe_cast_profiled_obj(src, src_k, true);
5861 src_type = _gvn.type(src);
5862 top_src = src_type->isa_aryptr();
5863 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5864 src_spec = true;
5865 }
5866 if (!has_dest) {
5867 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5868 dest_type = _gvn.type(dest);
5869 top_dest = dest_type->isa_aryptr();
5870 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5871 dest_spec = true;
5872 }
5873 }
5874 }
5875
5876 if (has_src && has_dest && can_emit_guards) {
5877 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5878 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5879 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5880 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5881
5882 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5883 // If both arrays are object arrays then having the exact types
5884 // for both will remove the need for a subtype check at runtime
5885 // before the call and may make it possible to pick a faster copy
5886 // routine (without a subtype check on every element)
5887 // Do we have the exact type of src?
5888 bool could_have_src = src_spec;
5889 // Do we have the exact type of dest?
5890 bool could_have_dest = dest_spec;
5891 ciKlass* src_k = nullptr;
5892 ciKlass* dest_k = nullptr;
5893 if (!src_spec) {
5894 src_k = src_type->speculative_type_not_null();
5895 if (src_k != nullptr && src_k->is_array_klass()) {
5896 could_have_src = true;
5897 }
5898 }
5899 if (!dest_spec) {
5900 dest_k = dest_type->speculative_type_not_null();
5901 if (dest_k != nullptr && dest_k->is_array_klass()) {
5902 could_have_dest = true;
5903 }
5904 }
5905 if (could_have_src && could_have_dest) {
5906 // If we can have both exact types, emit the missing guards
5907 if (could_have_src && !src_spec) {
5908 src = maybe_cast_profiled_obj(src, src_k, true);
5909 }
5910 if (could_have_dest && !dest_spec) {
5911 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5912 }
5913 }
5914 }
5915 }
5916
5917 ciMethod* trap_method = method();
5918 int trap_bci = bci();
5919 if (saved_jvms_before_guards != nullptr) {
5920 trap_method = alloc->jvms()->method();
5921 trap_bci = alloc->jvms()->bci();
5922 }
5923
5924 bool negative_length_guard_generated = false;
5925
5926 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5927 can_emit_guards &&
5928 !src->is_top() && !dest->is_top()) {
5929 // validate arguments: enables transformation the ArrayCopyNode
5930 validated = true;
5931
5932 RegionNode* slow_region = new RegionNode(1);
5933 record_for_igvn(slow_region);
5934
5935 // (1) src and dest are arrays.
5936 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5937 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5938
5939 // (2) src and dest arrays must have elements of the same BasicType
5940 // done at macro expansion or at Ideal transformation time
5941
5942 // (4) src_offset must not be negative.
5943 generate_negative_guard(src_offset, slow_region);
5944
5945 // (5) dest_offset must not be negative.
5946 generate_negative_guard(dest_offset, slow_region);
5947
5948 // (7) src_offset + length must not exceed length of src.
5949 generate_limit_guard(src_offset, length,
5950 load_array_length(src),
5951 slow_region);
5952
5953 // (8) dest_offset + length must not exceed length of dest.
5954 generate_limit_guard(dest_offset, length,
5955 load_array_length(dest),
5956 slow_region);
5957
5958 // (6) length must not be negative.
5959 // This is also checked in generate_arraycopy() during macro expansion, but
5960 // we also have to check it here for the case where the ArrayCopyNode will
5961 // be eliminated by Escape Analysis.
5962 if (EliminateAllocations) {
5963 generate_negative_guard(length, slow_region);
5964 negative_length_guard_generated = true;
5965 }
5966
5967 // (9) each element of an oop array must be assignable
5968 Node* dest_klass = load_object_klass(dest);
5969 if (src != dest) {
5970 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5971
5972 if (not_subtype_ctrl != top()) {
5973 PreserveJVMState pjvms(this);
5974 set_control(not_subtype_ctrl);
5975 uncommon_trap(Deoptimization::Reason_intrinsic,
5976 Deoptimization::Action_make_not_entrant);
5977 assert(stopped(), "Should be stopped");
5978 }
5979 }
5980 {
5981 PreserveJVMState pjvms(this);
5982 set_control(_gvn.transform(slow_region));
5983 uncommon_trap(Deoptimization::Reason_intrinsic,
5984 Deoptimization::Action_make_not_entrant);
5985 assert(stopped(), "Should be stopped");
5986 }
5987
5988 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5989 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5990 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5991 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5992 }
5993
5994 if (stopped()) {
5995 return true;
5996 }
5997
5998 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5999 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6000 // so the compiler has a chance to eliminate them: during macro expansion,
6001 // we have to set their control (CastPP nodes are eliminated).
6002 load_object_klass(src), load_object_klass(dest),
6003 load_array_length(src), load_array_length(dest));
6004
6005 ac->set_arraycopy(validated);
6006
6007 Node* n = _gvn.transform(ac);
6008 if (n == ac) {
6009 ac->connect_outputs(this);
6010 } else {
6011 assert(validated, "shouldn't transform if all arguments not validated");
6012 set_all_memory(n);
6013 }
6014 clear_upper_avx();
6015
6016
6017 return true;
6018 }
6019
6020
6021 // Helper function which determines if an arraycopy immediately follows
6022 // an allocation, with no intervening tests or other escapes for the object.
6023 AllocateArrayNode*
6024 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6025 if (stopped()) return nullptr; // no fast path
6026 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6027
6028 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6029 if (alloc == nullptr) return nullptr;
6030
6031 Node* rawmem = memory(Compile::AliasIdxRaw);
6032 // Is the allocation's memory state untouched?
6033 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6034 // Bail out if there have been raw-memory effects since the allocation.
6035 // (Example: There might have been a call or safepoint.)
6036 return nullptr;
6037 }
6038 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6039 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6040 return nullptr;
6041 }
6042
6043 // There must be no unexpected observers of this allocation.
6044 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6045 Node* obs = ptr->fast_out(i);
6046 if (obs != this->map()) {
6047 return nullptr;
6048 }
6049 }
6050
6051 // This arraycopy must unconditionally follow the allocation of the ptr.
6052 Node* alloc_ctl = ptr->in(0);
6053 Node* ctl = control();
6054 while (ctl != alloc_ctl) {
6055 // There may be guards which feed into the slow_region.
6056 // Any other control flow means that we might not get a chance
6057 // to finish initializing the allocated object.
6058 // Various low-level checks bottom out in uncommon traps. These
6059 // are considered safe since we've already checked above that
6060 // there is no unexpected observer of this allocation.
6061 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6062 assert(ctl->in(0)->is_If(), "must be If");
6063 ctl = ctl->in(0)->in(0);
6064 } else {
6065 return nullptr;
6066 }
6067 }
6068
6069 // If we get this far, we have an allocation which immediately
6070 // precedes the arraycopy, and we can take over zeroing the new object.
6071 // The arraycopy will finish the initialization, and provide
6072 // a new control state to which we will anchor the destination pointer.
6073
6074 return alloc;
6075 }
6076
6077 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6078 if (node->is_IfProj()) {
6079 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6080 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6081 Node* obs = other_proj->fast_out(j);
6082 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6083 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6084 return obs->as_CallStaticJava();
6085 }
6086 }
6087 }
6088 return nullptr;
6089 }
6090
6091 //-------------inline_encodeISOArray-----------------------------------
6092 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6093 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6094 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6095 // encode char[] to byte[] in ISO_8859_1 or ASCII
6096 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6097 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6098 // no receiver since it is static method
6099 Node *src = argument(0);
6100 Node *src_offset = argument(1);
6101 Node *dst = argument(2);
6102 Node *dst_offset = argument(3);
6103 Node *length = argument(4);
6104
6105 // Cast source & target arrays to not-null
6106 src = must_be_not_null(src, true);
6107 dst = must_be_not_null(dst, true);
6108 if (stopped()) {
6109 return true;
6110 }
6111
6112 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6113 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6114 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6115 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6116 // failed array check
6117 return false;
6118 }
6119
6120 // Figure out the size and type of the elements we will be copying.
6121 BasicType src_elem = src_type->elem()->array_element_basic_type();
6122 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6123 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6124 return false;
6125 }
6126
6127 // Check source & target bounds
6128 RegionNode* bailout = create_bailout();
6129 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6130 generate_string_range_check(dst, dst_offset, length, false, bailout);
6131 if (check_bailout(bailout)) {
6132 return true;
6133 }
6134
6135 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6136 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6137 // 'src_start' points to src array + scaled offset
6138 // 'dst_start' points to dst array + scaled offset
6139
6140 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6141 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6142 enc = _gvn.transform(enc);
6143 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6144 set_memory(res_mem, mtype);
6145 set_result(enc);
6146 clear_upper_avx();
6147
6148 return true;
6149 }
6150
6151 //-------------inline_multiplyToLen-----------------------------------
6152 bool LibraryCallKit::inline_multiplyToLen() {
6153 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6154
6155 address stubAddr = StubRoutines::multiplyToLen();
6156 if (stubAddr == nullptr) {
6157 return false; // Intrinsic's stub is not implemented on this platform
6158 }
6159 const char* stubName = "multiplyToLen";
6160
6161 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6162
6163 // no receiver because it is a static method
6164 Node* x = argument(0);
6165 Node* xlen = argument(1);
6166 Node* y = argument(2);
6167 Node* ylen = argument(3);
6168 Node* z = argument(4);
6169
6170 x = must_be_not_null(x, true);
6171 y = must_be_not_null(y, true);
6172
6173 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6174 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6175 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6176 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6177 // failed array check
6178 return false;
6179 }
6180
6181 BasicType x_elem = x_type->elem()->array_element_basic_type();
6182 BasicType y_elem = y_type->elem()->array_element_basic_type();
6183 if (x_elem != T_INT || y_elem != T_INT) {
6184 return false;
6185 }
6186
6187 Node* x_start = array_element_address(x, intcon(0), x_elem);
6188 Node* y_start = array_element_address(y, intcon(0), y_elem);
6189 // 'x_start' points to x array + scaled xlen
6190 // 'y_start' points to y array + scaled ylen
6191
6192 Node* z_start = array_element_address(z, intcon(0), T_INT);
6193
6194 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6195 OptoRuntime::multiplyToLen_Type(),
6196 stubAddr, stubName, TypePtr::BOTTOM,
6197 x_start, xlen, y_start, ylen, z_start);
6198
6199 C->set_has_split_ifs(true); // Has chance for split-if optimization
6200 set_result(z);
6201 return true;
6202 }
6203
6204 //-------------inline_squareToLen------------------------------------
6205 bool LibraryCallKit::inline_squareToLen() {
6206 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6207
6208 address stubAddr = StubRoutines::squareToLen();
6209 if (stubAddr == nullptr) {
6210 return false; // Intrinsic's stub is not implemented on this platform
6211 }
6212 const char* stubName = "squareToLen";
6213
6214 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6215
6216 Node* x = argument(0);
6217 Node* len = argument(1);
6218 Node* z = argument(2);
6219 Node* zlen = argument(3);
6220
6221 x = must_be_not_null(x, true);
6222 z = must_be_not_null(z, true);
6223
6224 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6225 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6226 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6227 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6228 // failed array check
6229 return false;
6230 }
6231
6232 BasicType x_elem = x_type->elem()->array_element_basic_type();
6233 BasicType z_elem = z_type->elem()->array_element_basic_type();
6234 if (x_elem != T_INT || z_elem != T_INT) {
6235 return false;
6236 }
6237
6238
6239 Node* x_start = array_element_address(x, intcon(0), x_elem);
6240 Node* z_start = array_element_address(z, intcon(0), z_elem);
6241
6242 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6243 OptoRuntime::squareToLen_Type(),
6244 stubAddr, stubName, TypePtr::BOTTOM,
6245 x_start, len, z_start, zlen);
6246
6247 set_result(z);
6248 return true;
6249 }
6250
6251 //-------------inline_mulAdd------------------------------------------
6252 bool LibraryCallKit::inline_mulAdd() {
6253 assert(UseMulAddIntrinsic, "not implemented on this platform");
6254
6255 address stubAddr = StubRoutines::mulAdd();
6256 if (stubAddr == nullptr) {
6257 return false; // Intrinsic's stub is not implemented on this platform
6258 }
6259 const char* stubName = "mulAdd";
6260
6261 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6262
6263 Node* out = argument(0);
6264 Node* in = argument(1);
6265 Node* offset = argument(2);
6266 Node* len = argument(3);
6267 Node* k = argument(4);
6268
6269 in = must_be_not_null(in, true);
6270 out = must_be_not_null(out, true);
6271
6272 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6273 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6274 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6275 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6276 // failed array check
6277 return false;
6278 }
6279
6280 BasicType out_elem = out_type->elem()->array_element_basic_type();
6281 BasicType in_elem = in_type->elem()->array_element_basic_type();
6282 if (out_elem != T_INT || in_elem != T_INT) {
6283 return false;
6284 }
6285
6286 Node* outlen = load_array_length(out);
6287 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6288 Node* out_start = array_element_address(out, intcon(0), out_elem);
6289 Node* in_start = array_element_address(in, intcon(0), in_elem);
6290
6291 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6292 OptoRuntime::mulAdd_Type(),
6293 stubAddr, stubName, TypePtr::BOTTOM,
6294 out_start,in_start, new_offset, len, k);
6295 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6296 set_result(result);
6297 return true;
6298 }
6299
6300 //-------------inline_montgomeryMultiply-----------------------------------
6301 bool LibraryCallKit::inline_montgomeryMultiply() {
6302 address stubAddr = StubRoutines::montgomeryMultiply();
6303 if (stubAddr == nullptr) {
6304 return false; // Intrinsic's stub is not implemented on this platform
6305 }
6306
6307 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6308 const char* stubName = "montgomery_multiply";
6309
6310 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6311
6312 Node* a = argument(0);
6313 Node* b = argument(1);
6314 Node* n = argument(2);
6315 Node* len = argument(3);
6316 Node* inv = argument(4);
6317 Node* m = argument(6);
6318
6319 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6320 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6321 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6322 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6323 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6324 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6325 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6326 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6327 // failed array check
6328 return false;
6329 }
6330
6331 BasicType a_elem = a_type->elem()->array_element_basic_type();
6332 BasicType b_elem = b_type->elem()->array_element_basic_type();
6333 BasicType n_elem = n_type->elem()->array_element_basic_type();
6334 BasicType m_elem = m_type->elem()->array_element_basic_type();
6335 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6336 return false;
6337 }
6338
6339 // Make the call
6340 {
6341 Node* a_start = array_element_address(a, intcon(0), a_elem);
6342 Node* b_start = array_element_address(b, intcon(0), b_elem);
6343 Node* n_start = array_element_address(n, intcon(0), n_elem);
6344 Node* m_start = array_element_address(m, intcon(0), m_elem);
6345
6346 Node* call = make_runtime_call(RC_LEAF,
6347 OptoRuntime::montgomeryMultiply_Type(),
6348 stubAddr, stubName, TypePtr::BOTTOM,
6349 a_start, b_start, n_start, len, inv, top(),
6350 m_start);
6351 set_result(m);
6352 }
6353
6354 return true;
6355 }
6356
6357 bool LibraryCallKit::inline_montgomerySquare() {
6358 address stubAddr = StubRoutines::montgomerySquare();
6359 if (stubAddr == nullptr) {
6360 return false; // Intrinsic's stub is not implemented on this platform
6361 }
6362
6363 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6364 const char* stubName = "montgomery_square";
6365
6366 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6367
6368 Node* a = argument(0);
6369 Node* n = argument(1);
6370 Node* len = argument(2);
6371 Node* inv = argument(3);
6372 Node* m = argument(5);
6373
6374 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6375 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6376 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6377 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6378 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6379 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6380 // failed array check
6381 return false;
6382 }
6383
6384 BasicType a_elem = a_type->elem()->array_element_basic_type();
6385 BasicType n_elem = n_type->elem()->array_element_basic_type();
6386 BasicType m_elem = m_type->elem()->array_element_basic_type();
6387 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6388 return false;
6389 }
6390
6391 // Make the call
6392 {
6393 Node* a_start = array_element_address(a, intcon(0), a_elem);
6394 Node* n_start = array_element_address(n, intcon(0), n_elem);
6395 Node* m_start = array_element_address(m, intcon(0), m_elem);
6396
6397 Node* call = make_runtime_call(RC_LEAF,
6398 OptoRuntime::montgomerySquare_Type(),
6399 stubAddr, stubName, TypePtr::BOTTOM,
6400 a_start, n_start, len, inv, top(),
6401 m_start);
6402 set_result(m);
6403 }
6404
6405 return true;
6406 }
6407
6408 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6409 address stubAddr = nullptr;
6410 const char* stubName = nullptr;
6411
6412 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6413 if (stubAddr == nullptr) {
6414 return false; // Intrinsic's stub is not implemented on this platform
6415 }
6416
6417 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6418
6419 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6420
6421 Node* newArr = argument(0);
6422 Node* oldArr = argument(1);
6423 Node* newIdx = argument(2);
6424 Node* shiftCount = argument(3);
6425 Node* numIter = argument(4);
6426
6427 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6428 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6429 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6430 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6431 return false;
6432 }
6433
6434 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6435 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6436 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6437 return false;
6438 }
6439
6440 // Make the call
6441 {
6442 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6443 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6444
6445 Node* call = make_runtime_call(RC_LEAF,
6446 OptoRuntime::bigIntegerShift_Type(),
6447 stubAddr,
6448 stubName,
6449 TypePtr::BOTTOM,
6450 newArr_start,
6451 oldArr_start,
6452 newIdx,
6453 shiftCount,
6454 numIter);
6455 }
6456
6457 return true;
6458 }
6459
6460 //-------------inline_vectorizedMismatch------------------------------
6461 bool LibraryCallKit::inline_vectorizedMismatch() {
6462 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6463
6464 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6465 Node* obja = argument(0); // Object
6466 Node* aoffset = argument(1); // long
6467 Node* objb = argument(3); // Object
6468 Node* boffset = argument(4); // long
6469 Node* length = argument(6); // int
6470 Node* scale = argument(7); // int
6471
6472 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6473 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6474 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6475 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6476 scale == top()) {
6477 return false; // failed input validation
6478 }
6479
6480 Node* obja_adr = make_unsafe_address(obja, aoffset);
6481 Node* objb_adr = make_unsafe_address(objb, boffset);
6482
6483 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6484 //
6485 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6486 // if (length <= inline_limit) {
6487 // inline_path:
6488 // vmask = VectorMaskGen length
6489 // vload1 = LoadVectorMasked obja, vmask
6490 // vload2 = LoadVectorMasked objb, vmask
6491 // result1 = VectorCmpMasked vload1, vload2, vmask
6492 // } else {
6493 // call_stub_path:
6494 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6495 // }
6496 // exit_block:
6497 // return Phi(result1, result2);
6498 //
6499 enum { inline_path = 1, // input is small enough to process it all at once
6500 stub_path = 2, // input is too large; call into the VM
6501 PATH_LIMIT = 3
6502 };
6503
6504 Node* exit_block = new RegionNode(PATH_LIMIT);
6505 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6506 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6507
6508 Node* call_stub_path = control();
6509
6510 BasicType elem_bt = T_ILLEGAL;
6511
6512 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6513 if (scale_t->is_con()) {
6514 switch (scale_t->get_con()) {
6515 case 0: elem_bt = T_BYTE; break;
6516 case 1: elem_bt = T_SHORT; break;
6517 case 2: elem_bt = T_INT; break;
6518 case 3: elem_bt = T_LONG; break;
6519
6520 default: elem_bt = T_ILLEGAL; break; // not supported
6521 }
6522 }
6523
6524 int inline_limit = 0;
6525 bool do_partial_inline = false;
6526
6527 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6528 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6529 do_partial_inline = inline_limit >= 16;
6530 }
6531
6532 if (do_partial_inline) {
6533 assert(elem_bt != T_ILLEGAL, "sanity");
6534
6535 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6536 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6537 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6538
6539 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6540 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6541 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6542
6543 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6544
6545 if (!stopped()) {
6546 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6547
6548 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6549 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6550 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6551 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6552
6553 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6554 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6555 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6556 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6557
6558 exit_block->init_req(inline_path, control());
6559 memory_phi->init_req(inline_path, map()->memory());
6560 result_phi->init_req(inline_path, result);
6561
6562 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6563 clear_upper_avx();
6564 }
6565 }
6566 }
6567
6568 if (call_stub_path != nullptr) {
6569 set_control(call_stub_path);
6570
6571 Node* call = make_runtime_call(RC_LEAF,
6572 OptoRuntime::vectorizedMismatch_Type(),
6573 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6574 obja_adr, objb_adr, length, scale);
6575
6576 exit_block->init_req(stub_path, control());
6577 memory_phi->init_req(stub_path, map()->memory());
6578 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6579 }
6580
6581 exit_block = _gvn.transform(exit_block);
6582 memory_phi = _gvn.transform(memory_phi);
6583 result_phi = _gvn.transform(result_phi);
6584
6585 record_for_igvn(exit_block);
6586 record_for_igvn(memory_phi);
6587 record_for_igvn(result_phi);
6588
6589 set_control(exit_block);
6590 set_all_memory(memory_phi);
6591 set_result(result_phi);
6592
6593 return true;
6594 }
6595
6596 //------------------------------inline_vectorizedHashcode----------------------------
6597 bool LibraryCallKit::inline_vectorizedHashCode() {
6598 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6599
6600 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6601 Node* array = argument(0);
6602 Node* offset = argument(1);
6603 Node* length = argument(2);
6604 Node* initialValue = argument(3);
6605 Node* basic_type = argument(4);
6606
6607 if (basic_type == top()) {
6608 return false; // failed input validation
6609 }
6610
6611 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6612 if (!basic_type_t->is_con()) {
6613 return false; // Only intrinsify if mode argument is constant
6614 }
6615
6616 array = must_be_not_null(array, true);
6617
6618 BasicType bt = (BasicType)basic_type_t->get_con();
6619
6620 // Resolve address of first element
6621 Node* array_start = array_element_address(array, offset, bt);
6622
6623 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6624 array_start, length, initialValue, basic_type)));
6625 clear_upper_avx();
6626
6627 return true;
6628 }
6629
6630 /**
6631 * Calculate CRC32 for byte.
6632 * int java.util.zip.CRC32.update(int crc, int b)
6633 */
6634 bool LibraryCallKit::inline_updateCRC32() {
6635 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6636 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6637 // no receiver since it is static method
6638 Node* crc = argument(0); // type: int
6639 Node* b = argument(1); // type: int
6640
6641 /*
6642 * int c = ~ crc;
6643 * b = timesXtoThe32[(b ^ c) & 0xFF];
6644 * b = b ^ (c >>> 8);
6645 * crc = ~b;
6646 */
6647
6648 Node* M1 = intcon(-1);
6649 crc = _gvn.transform(new XorINode(crc, M1));
6650 Node* result = _gvn.transform(new XorINode(crc, b));
6651 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6652
6653 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6654 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6655 Node* adr = off_heap_plus_addr(base, ConvI2X(offset));
6656 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6657
6658 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6659 result = _gvn.transform(new XorINode(crc, result));
6660 result = _gvn.transform(new XorINode(result, M1));
6661 set_result(result);
6662 return true;
6663 }
6664
6665 /**
6666 * Calculate CRC32 for byte[] array.
6667 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6668 */
6669 bool LibraryCallKit::inline_updateBytesCRC32() {
6670 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6671 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6672 // no receiver since it is static method
6673 Node* crc = argument(0); // type: int
6674 Node* src = argument(1); // type: oop
6675 Node* offset = argument(2); // type: int
6676 Node* length = argument(3); // type: int
6677
6678 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6679 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6680 // failed array check
6681 return false;
6682 }
6683
6684 // Figure out the size and type of the elements we will be copying.
6685 BasicType src_elem = src_type->elem()->array_element_basic_type();
6686 if (src_elem != T_BYTE) {
6687 return false;
6688 }
6689
6690 // 'src_start' points to src array + scaled offset
6691 src = must_be_not_null(src, true);
6692 Node* src_start = array_element_address(src, offset, src_elem);
6693
6694 // We assume that range check is done by caller.
6695 // TODO: generate range check (offset+length < src.length) in debug VM.
6696
6697 // Call the stub.
6698 address stubAddr = StubRoutines::updateBytesCRC32();
6699 const char *stubName = "updateBytesCRC32";
6700
6701 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6702 stubAddr, stubName, TypePtr::BOTTOM,
6703 crc, src_start, length);
6704 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6705 set_result(result);
6706 return true;
6707 }
6708
6709 /**
6710 * Calculate CRC32 for ByteBuffer.
6711 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6712 */
6713 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6714 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6715 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6716 // no receiver since it is static method
6717 Node* crc = argument(0); // type: int
6718 Node* src = argument(1); // type: long
6719 Node* offset = argument(3); // type: int
6720 Node* length = argument(4); // type: int
6721
6722 src = ConvL2X(src); // adjust Java long to machine word
6723 Node* base = _gvn.transform(new CastX2PNode(src));
6724 offset = ConvI2X(offset);
6725
6726 // 'src_start' points to src array + scaled offset
6727 Node* src_start = off_heap_plus_addr(base, offset);
6728
6729 // Call the stub.
6730 address stubAddr = StubRoutines::updateBytesCRC32();
6731 const char *stubName = "updateBytesCRC32";
6732
6733 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6734 stubAddr, stubName, TypePtr::BOTTOM,
6735 crc, src_start, length);
6736 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6737 set_result(result);
6738 return true;
6739 }
6740
6741 //------------------------------get_table_from_crc32c_class-----------------------
6742 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6743 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6744 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6745
6746 return table;
6747 }
6748
6749 //------------------------------inline_updateBytesCRC32C-----------------------
6750 //
6751 // Calculate CRC32C for byte[] array.
6752 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6753 //
6754 bool LibraryCallKit::inline_updateBytesCRC32C() {
6755 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6756 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6757 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6758 // no receiver since it is a static method
6759 Node* crc = argument(0); // type: int
6760 Node* src = argument(1); // type: oop
6761 Node* offset = argument(2); // type: int
6762 Node* end = argument(3); // type: int
6763
6764 Node* length = _gvn.transform(new SubINode(end, offset));
6765
6766 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6767 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6768 // failed array check
6769 return false;
6770 }
6771
6772 // Figure out the size and type of the elements we will be copying.
6773 BasicType src_elem = src_type->elem()->array_element_basic_type();
6774 if (src_elem != T_BYTE) {
6775 return false;
6776 }
6777
6778 // 'src_start' points to src array + scaled offset
6779 src = must_be_not_null(src, true);
6780 Node* src_start = array_element_address(src, offset, src_elem);
6781
6782 // static final int[] byteTable in class CRC32C
6783 Node* table = get_table_from_crc32c_class(callee()->holder());
6784 table = must_be_not_null(table, true);
6785 Node* table_start = array_element_address(table, intcon(0), T_INT);
6786
6787 // We assume that range check is done by caller.
6788 // TODO: generate range check (offset+length < src.length) in debug VM.
6789
6790 // Call the stub.
6791 address stubAddr = StubRoutines::updateBytesCRC32C();
6792 const char *stubName = "updateBytesCRC32C";
6793
6794 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6795 stubAddr, stubName, TypePtr::BOTTOM,
6796 crc, src_start, length, table_start);
6797 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6798 set_result(result);
6799 return true;
6800 }
6801
6802 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6803 //
6804 // Calculate CRC32C for DirectByteBuffer.
6805 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6806 //
6807 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6808 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6809 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6810 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6811 // no receiver since it is a static method
6812 Node* crc = argument(0); // type: int
6813 Node* src = argument(1); // type: long
6814 Node* offset = argument(3); // type: int
6815 Node* end = argument(4); // type: int
6816
6817 Node* length = _gvn.transform(new SubINode(end, offset));
6818
6819 src = ConvL2X(src); // adjust Java long to machine word
6820 Node* base = _gvn.transform(new CastX2PNode(src));
6821 offset = ConvI2X(offset);
6822
6823 // 'src_start' points to src array + scaled offset
6824 Node* src_start = off_heap_plus_addr(base, offset);
6825
6826 // static final int[] byteTable in class CRC32C
6827 Node* table = get_table_from_crc32c_class(callee()->holder());
6828 table = must_be_not_null(table, true);
6829 Node* table_start = array_element_address(table, intcon(0), T_INT);
6830
6831 // Call the stub.
6832 address stubAddr = StubRoutines::updateBytesCRC32C();
6833 const char *stubName = "updateBytesCRC32C";
6834
6835 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6836 stubAddr, stubName, TypePtr::BOTTOM,
6837 crc, src_start, length, table_start);
6838 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6839 set_result(result);
6840 return true;
6841 }
6842
6843 //------------------------------inline_updateBytesAdler32----------------------
6844 //
6845 // Calculate Adler32 checksum for byte[] array.
6846 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6847 //
6848 bool LibraryCallKit::inline_updateBytesAdler32() {
6849 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6850 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6851 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6852 // no receiver since it is static method
6853 Node* crc = argument(0); // type: int
6854 Node* src = argument(1); // type: oop
6855 Node* offset = argument(2); // type: int
6856 Node* length = argument(3); // type: int
6857
6858 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6859 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6860 // failed array check
6861 return false;
6862 }
6863
6864 // Figure out the size and type of the elements we will be copying.
6865 BasicType src_elem = src_type->elem()->array_element_basic_type();
6866 if (src_elem != T_BYTE) {
6867 return false;
6868 }
6869
6870 // 'src_start' points to src array + scaled offset
6871 Node* src_start = array_element_address(src, offset, src_elem);
6872
6873 // We assume that range check is done by caller.
6874 // TODO: generate range check (offset+length < src.length) in debug VM.
6875
6876 // Call the stub.
6877 address stubAddr = StubRoutines::updateBytesAdler32();
6878 const char *stubName = "updateBytesAdler32";
6879
6880 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6881 stubAddr, stubName, TypePtr::BOTTOM,
6882 crc, src_start, length);
6883 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6884 set_result(result);
6885 return true;
6886 }
6887
6888 //------------------------------inline_updateByteBufferAdler32---------------
6889 //
6890 // Calculate Adler32 checksum for DirectByteBuffer.
6891 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6892 //
6893 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6894 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6895 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6896 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6897 // no receiver since it is static method
6898 Node* crc = argument(0); // type: int
6899 Node* src = argument(1); // type: long
6900 Node* offset = argument(3); // type: int
6901 Node* length = argument(4); // type: int
6902
6903 src = ConvL2X(src); // adjust Java long to machine word
6904 Node* base = _gvn.transform(new CastX2PNode(src));
6905 offset = ConvI2X(offset);
6906
6907 // 'src_start' points to src array + scaled offset
6908 Node* src_start = off_heap_plus_addr(base, offset);
6909
6910 // Call the stub.
6911 address stubAddr = StubRoutines::updateBytesAdler32();
6912 const char *stubName = "updateBytesAdler32";
6913
6914 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6915 stubAddr, stubName, TypePtr::BOTTOM,
6916 crc, src_start, length);
6917
6918 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6919 set_result(result);
6920 return true;
6921 }
6922
6923 //----------------------------inline_reference_get0----------------------------
6924 // public T java.lang.ref.Reference.get();
6925 bool LibraryCallKit::inline_reference_get0() {
6926 const int referent_offset = java_lang_ref_Reference::referent_offset();
6927
6928 // Get the argument:
6929 Node* reference_obj = null_check_receiver();
6930 if (stopped()) return true;
6931
6932 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6933 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6934 decorators, /*is_static*/ false, nullptr);
6935 if (result == nullptr) return false;
6936
6937 // Add memory barrier to prevent commoning reads from this field
6938 // across safepoint since GC can change its value.
6939 insert_mem_bar(Op_MemBarCPUOrder);
6940
6941 set_result(result);
6942 return true;
6943 }
6944
6945 //----------------------------inline_reference_refersTo0----------------------------
6946 // bool java.lang.ref.Reference.refersTo0();
6947 // bool java.lang.ref.PhantomReference.refersTo0();
6948 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6949 // Get arguments:
6950 Node* reference_obj = null_check_receiver();
6951 Node* other_obj = argument(1);
6952 if (stopped()) return true;
6953
6954 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6955 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6956 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6957 decorators, /*is_static*/ false, nullptr);
6958 if (referent == nullptr) return false;
6959
6960 // Add memory barrier to prevent commoning reads from this field
6961 // across safepoint since GC can change its value.
6962 insert_mem_bar(Op_MemBarCPUOrder);
6963
6964 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6965 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6966 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6967
6968 RegionNode* region = new RegionNode(3);
6969 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6970
6971 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6972 region->init_req(1, if_true);
6973 phi->init_req(1, intcon(1));
6974
6975 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6976 region->init_req(2, if_false);
6977 phi->init_req(2, intcon(0));
6978
6979 set_control(_gvn.transform(region));
6980 record_for_igvn(region);
6981 set_result(_gvn.transform(phi));
6982 return true;
6983 }
6984
6985 //----------------------------inline_reference_clear0----------------------------
6986 // void java.lang.ref.Reference.clear0();
6987 // void java.lang.ref.PhantomReference.clear0();
6988 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6989 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6990
6991 // Get arguments
6992 Node* reference_obj = null_check_receiver();
6993 if (stopped()) return true;
6994
6995 // Common access parameters
6996 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6997 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6998 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6999 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7000 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7001
7002 Node* referent = access_load_at(reference_obj,
7003 referent_field_addr,
7004 referent_field_addr_type,
7005 val_type,
7006 T_OBJECT,
7007 decorators);
7008
7009 IdealKit ideal(this);
7010 #define __ ideal.
7011 __ if_then(referent, BoolTest::ne, null());
7012 sync_kit(ideal);
7013 access_store_at(reference_obj,
7014 referent_field_addr,
7015 referent_field_addr_type,
7016 null(),
7017 val_type,
7018 T_OBJECT,
7019 decorators);
7020 __ sync_kit(this);
7021 __ end_if();
7022 final_sync(ideal);
7023 #undef __
7024
7025 return true;
7026 }
7027
7028 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7029 DecoratorSet decorators, bool is_static,
7030 ciInstanceKlass* fromKls) {
7031 if (fromKls == nullptr) {
7032 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7033 assert(tinst != nullptr, "obj is null");
7034 assert(tinst->is_loaded(), "obj is not loaded");
7035 fromKls = tinst->instance_klass();
7036 } else {
7037 assert(is_static, "only for static field access");
7038 }
7039 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7040 ciSymbol::make(fieldTypeString),
7041 is_static);
7042
7043 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7044 if (field == nullptr) return (Node *) nullptr;
7045
7046 if (is_static) {
7047 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7048 fromObj = makecon(tip);
7049 }
7050
7051 // Next code copied from Parse::do_get_xxx():
7052
7053 // Compute address and memory type.
7054 int offset = field->offset_in_bytes();
7055 bool is_vol = field->is_volatile();
7056 ciType* field_klass = field->type();
7057 assert(field_klass->is_loaded(), "should be loaded");
7058 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7059 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7060 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7061 "slice of address and input slice don't match");
7062 BasicType bt = field->layout_type();
7063
7064 // Build the resultant type of the load
7065 const Type *type;
7066 if (bt == T_OBJECT) {
7067 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7068 } else {
7069 type = Type::get_const_basic_type(bt);
7070 }
7071
7072 if (is_vol) {
7073 decorators |= MO_SEQ_CST;
7074 }
7075
7076 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7077 }
7078
7079 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7080 bool is_exact /* true */, bool is_static /* false */,
7081 ciInstanceKlass * fromKls /* nullptr */) {
7082 if (fromKls == nullptr) {
7083 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7084 assert(tinst != nullptr, "obj is null");
7085 assert(tinst->is_loaded(), "obj is not loaded");
7086 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7087 fromKls = tinst->instance_klass();
7088 }
7089 else {
7090 assert(is_static, "only for static field access");
7091 }
7092 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7093 ciSymbol::make(fieldTypeString),
7094 is_static);
7095
7096 assert(field != nullptr, "undefined field");
7097 assert(!field->is_volatile(), "not defined for volatile fields");
7098
7099 if (is_static) {
7100 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7101 fromObj = makecon(tip);
7102 }
7103
7104 // Next code copied from Parse::do_get_xxx():
7105
7106 // Compute address and memory type.
7107 int offset = field->offset_in_bytes();
7108 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7109
7110 return adr;
7111 }
7112
7113 //------------------------------inline_aescrypt_Block-----------------------
7114 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7115 address stubAddr = nullptr;
7116 const char *stubName;
7117 bool is_decrypt = false;
7118 assert(UseAES, "need AES instruction support");
7119
7120 switch(id) {
7121 case vmIntrinsics::_aescrypt_encryptBlock:
7122 stubAddr = StubRoutines::aescrypt_encryptBlock();
7123 stubName = "aescrypt_encryptBlock";
7124 break;
7125 case vmIntrinsics::_aescrypt_decryptBlock:
7126 stubAddr = StubRoutines::aescrypt_decryptBlock();
7127 stubName = "aescrypt_decryptBlock";
7128 is_decrypt = true;
7129 break;
7130 default:
7131 break;
7132 }
7133 if (stubAddr == nullptr) return false;
7134
7135 Node* aescrypt_object = argument(0);
7136 Node* src = argument(1);
7137 Node* src_offset = argument(2);
7138 Node* dest = argument(3);
7139 Node* dest_offset = argument(4);
7140
7141 src = must_be_not_null(src, true);
7142 dest = must_be_not_null(dest, true);
7143
7144 // (1) src and dest are arrays.
7145 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7146 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7147 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7148 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7149
7150 // for the quick and dirty code we will skip all the checks.
7151 // we are just trying to get the call to be generated.
7152 Node* src_start = src;
7153 Node* dest_start = dest;
7154 if (src_offset != nullptr || dest_offset != nullptr) {
7155 assert(src_offset != nullptr && dest_offset != nullptr, "");
7156 src_start = array_element_address(src, src_offset, T_BYTE);
7157 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7158 }
7159
7160 // now need to get the start of its expanded key array
7161 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7162 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7163 if (k_start == nullptr) return false;
7164
7165 // Call the stub.
7166 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7167 stubAddr, stubName, TypePtr::BOTTOM,
7168 src_start, dest_start, k_start);
7169
7170 return true;
7171 }
7172
7173 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7174 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7175 address stubAddr = nullptr;
7176 const char *stubName = nullptr;
7177 bool is_decrypt = false;
7178 assert(UseAES, "need AES instruction support");
7179
7180 switch(id) {
7181 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7182 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7183 stubName = "cipherBlockChaining_encryptAESCrypt";
7184 break;
7185 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7186 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7187 stubName = "cipherBlockChaining_decryptAESCrypt";
7188 is_decrypt = true;
7189 break;
7190 default:
7191 break;
7192 }
7193 if (stubAddr == nullptr) return false;
7194
7195 Node* cipherBlockChaining_object = argument(0);
7196 Node* src = argument(1);
7197 Node* src_offset = argument(2);
7198 Node* len = argument(3);
7199 Node* dest = argument(4);
7200 Node* dest_offset = argument(5);
7201
7202 src = must_be_not_null(src, false);
7203 dest = must_be_not_null(dest, false);
7204
7205 // (1) src and dest are arrays.
7206 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7207 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7208 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7209 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7210
7211 // checks are the responsibility of the caller
7212 Node* src_start = src;
7213 Node* dest_start = dest;
7214 if (src_offset != nullptr || dest_offset != nullptr) {
7215 assert(src_offset != nullptr && dest_offset != nullptr, "");
7216 src_start = array_element_address(src, src_offset, T_BYTE);
7217 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7218 }
7219
7220 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7221 // (because of the predicated logic executed earlier).
7222 // so we cast it here safely.
7223 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7224
7225 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7226 if (embeddedCipherObj == nullptr) return false;
7227
7228 // cast it to what we know it will be at runtime
7229 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7230 assert(tinst != nullptr, "CBC obj is null");
7231 assert(tinst->is_loaded(), "CBC obj is not loaded");
7232 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7233 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7234
7235 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7236 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7237 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7238 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7239 aescrypt_object = _gvn.transform(aescrypt_object);
7240
7241 // we need to get the start of the aescrypt_object's expanded key array
7242 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7243 if (k_start == nullptr) return false;
7244
7245 // similarly, get the start address of the r vector
7246 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7247 if (objRvec == nullptr) return false;
7248 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7249
7250 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7251 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7252 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7253 stubAddr, stubName, TypePtr::BOTTOM,
7254 src_start, dest_start, k_start, r_start, len);
7255
7256 // return cipher length (int)
7257 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7258 set_result(retvalue);
7259 return true;
7260 }
7261
7262 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7263 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7264 address stubAddr = nullptr;
7265 const char *stubName = nullptr;
7266 bool is_decrypt = false;
7267 assert(UseAES, "need AES instruction support");
7268
7269 switch (id) {
7270 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7271 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7272 stubName = "electronicCodeBook_encryptAESCrypt";
7273 break;
7274 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7275 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7276 stubName = "electronicCodeBook_decryptAESCrypt";
7277 is_decrypt = true;
7278 break;
7279 default:
7280 break;
7281 }
7282
7283 if (stubAddr == nullptr) return false;
7284
7285 Node* electronicCodeBook_object = argument(0);
7286 Node* src = argument(1);
7287 Node* src_offset = argument(2);
7288 Node* len = argument(3);
7289 Node* dest = argument(4);
7290 Node* dest_offset = argument(5);
7291
7292 // (1) src and dest are arrays.
7293 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7294 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7295 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7296 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7297
7298 // checks are the responsibility of the caller
7299 Node* src_start = src;
7300 Node* dest_start = dest;
7301 if (src_offset != nullptr || dest_offset != nullptr) {
7302 assert(src_offset != nullptr && dest_offset != nullptr, "");
7303 src_start = array_element_address(src, src_offset, T_BYTE);
7304 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7305 }
7306
7307 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7308 // (because of the predicated logic executed earlier).
7309 // so we cast it here safely.
7310 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7311
7312 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7313 if (embeddedCipherObj == nullptr) return false;
7314
7315 // cast it to what we know it will be at runtime
7316 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7317 assert(tinst != nullptr, "ECB obj is null");
7318 assert(tinst->is_loaded(), "ECB obj is not loaded");
7319 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7320 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7321
7322 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7323 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7324 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7325 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7326 aescrypt_object = _gvn.transform(aescrypt_object);
7327
7328 // we need to get the start of the aescrypt_object's expanded key array
7329 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7330 if (k_start == nullptr) return false;
7331
7332 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7333 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7334 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7335 stubAddr, stubName, TypePtr::BOTTOM,
7336 src_start, dest_start, k_start, len);
7337
7338 // return cipher length (int)
7339 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7340 set_result(retvalue);
7341 return true;
7342 }
7343
7344 //------------------------------inline_counterMode_AESCrypt-----------------------
7345 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7346 assert(UseAES, "need AES instruction support");
7347 if (!UseAESCTRIntrinsics) return false;
7348
7349 address stubAddr = nullptr;
7350 const char *stubName = nullptr;
7351 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7352 stubAddr = StubRoutines::counterMode_AESCrypt();
7353 stubName = "counterMode_AESCrypt";
7354 }
7355 if (stubAddr == nullptr) return false;
7356
7357 Node* counterMode_object = argument(0);
7358 Node* src = argument(1);
7359 Node* src_offset = argument(2);
7360 Node* len = argument(3);
7361 Node* dest = argument(4);
7362 Node* dest_offset = argument(5);
7363
7364 // (1) src and dest are arrays.
7365 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7366 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7367 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7368 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7369
7370 // checks are the responsibility of the caller
7371 Node* src_start = src;
7372 Node* dest_start = dest;
7373 if (src_offset != nullptr || dest_offset != nullptr) {
7374 assert(src_offset != nullptr && dest_offset != nullptr, "");
7375 src_start = array_element_address(src, src_offset, T_BYTE);
7376 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7377 }
7378
7379 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7380 // (because of the predicated logic executed earlier).
7381 // so we cast it here safely.
7382 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7383 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7384 if (embeddedCipherObj == nullptr) return false;
7385 // cast it to what we know it will be at runtime
7386 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7387 assert(tinst != nullptr, "CTR obj is null");
7388 assert(tinst->is_loaded(), "CTR obj is not loaded");
7389 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7390 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7391 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7392 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7393 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7394 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7395 aescrypt_object = _gvn.transform(aescrypt_object);
7396 // we need to get the start of the aescrypt_object's expanded key array
7397 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7398 if (k_start == nullptr) return false;
7399 // similarly, get the start address of the r vector
7400 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7401 if (obj_counter == nullptr) return false;
7402 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7403
7404 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7405 if (saved_encCounter == nullptr) return false;
7406 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7407 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7408
7409 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7410 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7411 OptoRuntime::counterMode_aescrypt_Type(),
7412 stubAddr, stubName, TypePtr::BOTTOM,
7413 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7414
7415 // return cipher length (int)
7416 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7417 set_result(retvalue);
7418 return true;
7419 }
7420
7421 //------------------------------get_key_start_from_aescrypt_object-----------------------
7422 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7423 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7424 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7425 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7426 // The following platform specific stubs of encryption and decryption use the same round keys.
7427 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7428 bool use_decryption_key = false;
7429 #else
7430 bool use_decryption_key = is_decrypt;
7431 #endif
7432 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7433 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7434 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7435
7436 // now have the array, need to get the start address of the selected key array
7437 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7438 return k_start;
7439 }
7440
7441 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7442 // Return node representing slow path of predicate check.
7443 // the pseudo code we want to emulate with this predicate is:
7444 // for encryption:
7445 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7446 // for decryption:
7447 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7448 // note cipher==plain is more conservative than the original java code but that's OK
7449 //
7450 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7451 // The receiver was checked for null already.
7452 Node* objCBC = argument(0);
7453
7454 Node* src = argument(1);
7455 Node* dest = argument(4);
7456
7457 // Load embeddedCipher field of CipherBlockChaining object.
7458 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7459
7460 // get AESCrypt klass for instanceOf check
7461 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7462 // will have same classloader as CipherBlockChaining object
7463 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7464 assert(tinst != nullptr, "CBCobj is null");
7465 assert(tinst->is_loaded(), "CBCobj is not loaded");
7466
7467 // we want to do an instanceof comparison against the AESCrypt class
7468 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7469 if (!klass_AESCrypt->is_loaded()) {
7470 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7471 Node* ctrl = control();
7472 set_control(top()); // no regular fast path
7473 return ctrl;
7474 }
7475
7476 src = must_be_not_null(src, true);
7477 dest = must_be_not_null(dest, true);
7478
7479 // Resolve oops to stable for CmpP below.
7480 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7481
7482 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7483 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7484 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7485
7486 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7487
7488 // for encryption, we are done
7489 if (!decrypting)
7490 return instof_false; // even if it is null
7491
7492 // for decryption, we need to add a further check to avoid
7493 // taking the intrinsic path when cipher and plain are the same
7494 // see the original java code for why.
7495 RegionNode* region = new RegionNode(3);
7496 region->init_req(1, instof_false);
7497
7498 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7499 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7500 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7501 region->init_req(2, src_dest_conjoint);
7502
7503 record_for_igvn(region);
7504 return _gvn.transform(region);
7505 }
7506
7507 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7508 // Return node representing slow path of predicate check.
7509 // the pseudo code we want to emulate with this predicate is:
7510 // for encryption:
7511 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7512 // for decryption:
7513 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7514 // note cipher==plain is more conservative than the original java code but that's OK
7515 //
7516 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7517 // The receiver was checked for null already.
7518 Node* objECB = argument(0);
7519
7520 // Load embeddedCipher field of ElectronicCodeBook object.
7521 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7522
7523 // get AESCrypt klass for instanceOf check
7524 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7525 // will have same classloader as ElectronicCodeBook object
7526 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7527 assert(tinst != nullptr, "ECBobj is null");
7528 assert(tinst->is_loaded(), "ECBobj is not loaded");
7529
7530 // we want to do an instanceof comparison against the AESCrypt class
7531 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7532 if (!klass_AESCrypt->is_loaded()) {
7533 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7534 Node* ctrl = control();
7535 set_control(top()); // no regular fast path
7536 return ctrl;
7537 }
7538 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7539
7540 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7541 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7542 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7543
7544 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7545
7546 // for encryption, we are done
7547 if (!decrypting)
7548 return instof_false; // even if it is null
7549
7550 // for decryption, we need to add a further check to avoid
7551 // taking the intrinsic path when cipher and plain are the same
7552 // see the original java code for why.
7553 RegionNode* region = new RegionNode(3);
7554 region->init_req(1, instof_false);
7555 Node* src = argument(1);
7556 Node* dest = argument(4);
7557 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7558 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7559 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7560 region->init_req(2, src_dest_conjoint);
7561
7562 record_for_igvn(region);
7563 return _gvn.transform(region);
7564 }
7565
7566 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7567 // Return node representing slow path of predicate check.
7568 // the pseudo code we want to emulate with this predicate is:
7569 // for encryption:
7570 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7571 // for decryption:
7572 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7573 // note cipher==plain is more conservative than the original java code but that's OK
7574 //
7575
7576 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7577 // The receiver was checked for null already.
7578 Node* objCTR = argument(0);
7579
7580 // Load embeddedCipher field of CipherBlockChaining object.
7581 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7582
7583 // get AESCrypt klass for instanceOf check
7584 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7585 // will have same classloader as CipherBlockChaining object
7586 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7587 assert(tinst != nullptr, "CTRobj is null");
7588 assert(tinst->is_loaded(), "CTRobj is not loaded");
7589
7590 // we want to do an instanceof comparison against the AESCrypt class
7591 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7592 if (!klass_AESCrypt->is_loaded()) {
7593 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7594 Node* ctrl = control();
7595 set_control(top()); // no regular fast path
7596 return ctrl;
7597 }
7598
7599 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7600 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7601 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7602 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7603 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7604
7605 return instof_false; // even if it is null
7606 }
7607
7608 //------------------------------inline_ghash_processBlocks
7609 bool LibraryCallKit::inline_ghash_processBlocks() {
7610 address stubAddr;
7611 const char *stubName;
7612 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7613
7614 stubAddr = StubRoutines::ghash_processBlocks();
7615 stubName = "ghash_processBlocks";
7616
7617 Node* data = argument(0);
7618 Node* offset = argument(1);
7619 Node* len = argument(2);
7620 Node* state = argument(3);
7621 Node* subkeyH = argument(4);
7622
7623 state = must_be_not_null(state, true);
7624 subkeyH = must_be_not_null(subkeyH, true);
7625 data = must_be_not_null(data, true);
7626
7627 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7628 assert(state_start, "state is null");
7629 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7630 assert(subkeyH_start, "subkeyH is null");
7631 Node* data_start = array_element_address(data, offset, T_BYTE);
7632 assert(data_start, "data is null");
7633
7634 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7635 OptoRuntime::ghash_processBlocks_Type(),
7636 stubAddr, stubName, TypePtr::BOTTOM,
7637 state_start, subkeyH_start, data_start, len);
7638 return true;
7639 }
7640
7641 //------------------------------inline_chacha20Block
7642 bool LibraryCallKit::inline_chacha20Block() {
7643 address stubAddr;
7644 const char *stubName;
7645 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7646
7647 stubAddr = StubRoutines::chacha20Block();
7648 stubName = "chacha20Block";
7649
7650 Node* state = argument(0);
7651 Node* result = argument(1);
7652
7653 state = must_be_not_null(state, true);
7654 result = must_be_not_null(result, true);
7655
7656 Node* state_start = array_element_address(state, intcon(0), T_INT);
7657 assert(state_start, "state is null");
7658 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7659 assert(result_start, "result is null");
7660
7661 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7662 OptoRuntime::chacha20Block_Type(),
7663 stubAddr, stubName, TypePtr::BOTTOM,
7664 state_start, result_start);
7665 // return key stream length (int)
7666 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7667 set_result(retvalue);
7668 return true;
7669 }
7670
7671 //------------------------------inline_kyberNtt
7672 bool LibraryCallKit::inline_kyberNtt() {
7673 address stubAddr;
7674 const char *stubName;
7675 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7676 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7677
7678 stubAddr = StubRoutines::kyberNtt();
7679 stubName = "kyberNtt";
7680 if (!stubAddr) return false;
7681
7682 Node* coeffs = argument(0);
7683 Node* ntt_zetas = argument(1);
7684
7685 coeffs = must_be_not_null(coeffs, true);
7686 ntt_zetas = must_be_not_null(ntt_zetas, true);
7687
7688 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7689 assert(coeffs_start, "coeffs is null");
7690 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7691 assert(ntt_zetas_start, "ntt_zetas is null");
7692 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7693 OptoRuntime::kyberNtt_Type(),
7694 stubAddr, stubName, TypePtr::BOTTOM,
7695 coeffs_start, ntt_zetas_start);
7696 // return an int
7697 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7698 set_result(retvalue);
7699 return true;
7700 }
7701
7702 //------------------------------inline_kyberInverseNtt
7703 bool LibraryCallKit::inline_kyberInverseNtt() {
7704 address stubAddr;
7705 const char *stubName;
7706 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7707 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7708
7709 stubAddr = StubRoutines::kyberInverseNtt();
7710 stubName = "kyberInverseNtt";
7711 if (!stubAddr) return false;
7712
7713 Node* coeffs = argument(0);
7714 Node* zetas = argument(1);
7715
7716 coeffs = must_be_not_null(coeffs, true);
7717 zetas = must_be_not_null(zetas, true);
7718
7719 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7720 assert(coeffs_start, "coeffs is null");
7721 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7722 assert(zetas_start, "inverseNtt_zetas is null");
7723 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7724 OptoRuntime::kyberInverseNtt_Type(),
7725 stubAddr, stubName, TypePtr::BOTTOM,
7726 coeffs_start, zetas_start);
7727
7728 // return an int
7729 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7730 set_result(retvalue);
7731 return true;
7732 }
7733
7734 //------------------------------inline_kyberNttMult
7735 bool LibraryCallKit::inline_kyberNttMult() {
7736 address stubAddr;
7737 const char *stubName;
7738 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7739 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7740
7741 stubAddr = StubRoutines::kyberNttMult();
7742 stubName = "kyberNttMult";
7743 if (!stubAddr) return false;
7744
7745 Node* result = argument(0);
7746 Node* ntta = argument(1);
7747 Node* nttb = argument(2);
7748 Node* zetas = argument(3);
7749
7750 result = must_be_not_null(result, true);
7751 ntta = must_be_not_null(ntta, true);
7752 nttb = must_be_not_null(nttb, true);
7753 zetas = must_be_not_null(zetas, true);
7754
7755 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7756 assert(result_start, "result is null");
7757 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7758 assert(ntta_start, "ntta is null");
7759 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7760 assert(nttb_start, "nttb is null");
7761 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7762 assert(zetas_start, "nttMult_zetas is null");
7763 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7764 OptoRuntime::kyberNttMult_Type(),
7765 stubAddr, stubName, TypePtr::BOTTOM,
7766 result_start, ntta_start, nttb_start,
7767 zetas_start);
7768
7769 // return an int
7770 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7771 set_result(retvalue);
7772
7773 return true;
7774 }
7775
7776 //------------------------------inline_kyberAddPoly_2
7777 bool LibraryCallKit::inline_kyberAddPoly_2() {
7778 address stubAddr;
7779 const char *stubName;
7780 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7781 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7782
7783 stubAddr = StubRoutines::kyberAddPoly_2();
7784 stubName = "kyberAddPoly_2";
7785 if (!stubAddr) return false;
7786
7787 Node* result = argument(0);
7788 Node* a = argument(1);
7789 Node* b = argument(2);
7790
7791 result = must_be_not_null(result, true);
7792 a = must_be_not_null(a, true);
7793 b = must_be_not_null(b, true);
7794
7795 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7796 assert(result_start, "result is null");
7797 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7798 assert(a_start, "a is null");
7799 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7800 assert(b_start, "b is null");
7801 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7802 OptoRuntime::kyberAddPoly_2_Type(),
7803 stubAddr, stubName, TypePtr::BOTTOM,
7804 result_start, a_start, b_start);
7805 // return an int
7806 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7807 set_result(retvalue);
7808 return true;
7809 }
7810
7811 //------------------------------inline_kyberAddPoly_3
7812 bool LibraryCallKit::inline_kyberAddPoly_3() {
7813 address stubAddr;
7814 const char *stubName;
7815 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7816 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7817
7818 stubAddr = StubRoutines::kyberAddPoly_3();
7819 stubName = "kyberAddPoly_3";
7820 if (!stubAddr) return false;
7821
7822 Node* result = argument(0);
7823 Node* a = argument(1);
7824 Node* b = argument(2);
7825 Node* c = argument(3);
7826
7827 result = must_be_not_null(result, true);
7828 a = must_be_not_null(a, true);
7829 b = must_be_not_null(b, true);
7830 c = must_be_not_null(c, true);
7831
7832 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7833 assert(result_start, "result is null");
7834 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7835 assert(a_start, "a is null");
7836 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7837 assert(b_start, "b is null");
7838 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7839 assert(c_start, "c is null");
7840 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7841 OptoRuntime::kyberAddPoly_3_Type(),
7842 stubAddr, stubName, TypePtr::BOTTOM,
7843 result_start, a_start, b_start, c_start);
7844 // return an int
7845 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7846 set_result(retvalue);
7847 return true;
7848 }
7849
7850 //------------------------------inline_kyber12To16
7851 bool LibraryCallKit::inline_kyber12To16() {
7852 address stubAddr;
7853 const char *stubName;
7854 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7855 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7856
7857 stubAddr = StubRoutines::kyber12To16();
7858 stubName = "kyber12To16";
7859 if (!stubAddr) return false;
7860
7861 Node* condensed = argument(0);
7862 Node* condensedOffs = argument(1);
7863 Node* parsed = argument(2);
7864 Node* parsedLength = argument(3);
7865
7866 condensed = must_be_not_null(condensed, true);
7867 parsed = must_be_not_null(parsed, true);
7868
7869 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7870 assert(condensed_start, "condensed is null");
7871 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7872 assert(parsed_start, "parsed is null");
7873 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7874 OptoRuntime::kyber12To16_Type(),
7875 stubAddr, stubName, TypePtr::BOTTOM,
7876 condensed_start, condensedOffs, parsed_start, parsedLength);
7877 // return an int
7878 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7879 set_result(retvalue);
7880 return true;
7881
7882 }
7883
7884 //------------------------------inline_kyberBarrettReduce
7885 bool LibraryCallKit::inline_kyberBarrettReduce() {
7886 address stubAddr;
7887 const char *stubName;
7888 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7889 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7890
7891 stubAddr = StubRoutines::kyberBarrettReduce();
7892 stubName = "kyberBarrettReduce";
7893 if (!stubAddr) return false;
7894
7895 Node* coeffs = argument(0);
7896
7897 coeffs = must_be_not_null(coeffs, true);
7898
7899 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7900 assert(coeffs_start, "coeffs is null");
7901 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7902 OptoRuntime::kyberBarrettReduce_Type(),
7903 stubAddr, stubName, TypePtr::BOTTOM,
7904 coeffs_start);
7905 // return an int
7906 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7907 set_result(retvalue);
7908 return true;
7909 }
7910
7911 //------------------------------inline_dilithiumAlmostNtt
7912 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7913 address stubAddr;
7914 const char *stubName;
7915 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7916 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7917
7918 stubAddr = StubRoutines::dilithiumAlmostNtt();
7919 stubName = "dilithiumAlmostNtt";
7920 if (!stubAddr) return false;
7921
7922 Node* coeffs = argument(0);
7923 Node* ntt_zetas = argument(1);
7924
7925 coeffs = must_be_not_null(coeffs, true);
7926 ntt_zetas = must_be_not_null(ntt_zetas, true);
7927
7928 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7929 assert(coeffs_start, "coeffs is null");
7930 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7931 assert(ntt_zetas_start, "ntt_zetas is null");
7932 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7933 OptoRuntime::dilithiumAlmostNtt_Type(),
7934 stubAddr, stubName, TypePtr::BOTTOM,
7935 coeffs_start, ntt_zetas_start);
7936 // return an int
7937 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7938 set_result(retvalue);
7939 return true;
7940 }
7941
7942 //------------------------------inline_dilithiumAlmostInverseNtt
7943 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7944 address stubAddr;
7945 const char *stubName;
7946 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7947 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7948
7949 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7950 stubName = "dilithiumAlmostInverseNtt";
7951 if (!stubAddr) return false;
7952
7953 Node* coeffs = argument(0);
7954 Node* zetas = argument(1);
7955
7956 coeffs = must_be_not_null(coeffs, true);
7957 zetas = must_be_not_null(zetas, true);
7958
7959 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7960 assert(coeffs_start, "coeffs is null");
7961 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7962 assert(zetas_start, "inverseNtt_zetas is null");
7963 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7964 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7965 stubAddr, stubName, TypePtr::BOTTOM,
7966 coeffs_start, zetas_start);
7967 // return an int
7968 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7969 set_result(retvalue);
7970 return true;
7971 }
7972
7973 //------------------------------inline_dilithiumNttMult
7974 bool LibraryCallKit::inline_dilithiumNttMult() {
7975 address stubAddr;
7976 const char *stubName;
7977 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7978 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7979
7980 stubAddr = StubRoutines::dilithiumNttMult();
7981 stubName = "dilithiumNttMult";
7982 if (!stubAddr) return false;
7983
7984 Node* result = argument(0);
7985 Node* ntta = argument(1);
7986 Node* nttb = argument(2);
7987 Node* zetas = argument(3);
7988
7989 result = must_be_not_null(result, true);
7990 ntta = must_be_not_null(ntta, true);
7991 nttb = must_be_not_null(nttb, true);
7992 zetas = must_be_not_null(zetas, true);
7993
7994 Node* result_start = array_element_address(result, intcon(0), T_INT);
7995 assert(result_start, "result is null");
7996 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
7997 assert(ntta_start, "ntta is null");
7998 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
7999 assert(nttb_start, "nttb is null");
8000 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8001 OptoRuntime::dilithiumNttMult_Type(),
8002 stubAddr, stubName, TypePtr::BOTTOM,
8003 result_start, ntta_start, nttb_start);
8004
8005 // return an int
8006 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8007 set_result(retvalue);
8008
8009 return true;
8010 }
8011
8012 //------------------------------inline_dilithiumMontMulByConstant
8013 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8014 address stubAddr;
8015 const char *stubName;
8016 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8017 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8018
8019 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8020 stubName = "dilithiumMontMulByConstant";
8021 if (!stubAddr) return false;
8022
8023 Node* coeffs = argument(0);
8024 Node* constant = argument(1);
8025
8026 coeffs = must_be_not_null(coeffs, true);
8027
8028 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8029 assert(coeffs_start, "coeffs is null");
8030 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8031 OptoRuntime::dilithiumMontMulByConstant_Type(),
8032 stubAddr, stubName, TypePtr::BOTTOM,
8033 coeffs_start, constant);
8034
8035 // return an int
8036 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8037 set_result(retvalue);
8038 return true;
8039 }
8040
8041
8042 //------------------------------inline_dilithiumDecomposePoly
8043 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8044 address stubAddr;
8045 const char *stubName;
8046 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8047 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8048
8049 stubAddr = StubRoutines::dilithiumDecomposePoly();
8050 stubName = "dilithiumDecomposePoly";
8051 if (!stubAddr) return false;
8052
8053 Node* input = argument(0);
8054 Node* lowPart = argument(1);
8055 Node* highPart = argument(2);
8056 Node* twoGamma2 = argument(3);
8057 Node* multiplier = argument(4);
8058
8059 input = must_be_not_null(input, true);
8060 lowPart = must_be_not_null(lowPart, true);
8061 highPart = must_be_not_null(highPart, true);
8062
8063 Node* input_start = array_element_address(input, intcon(0), T_INT);
8064 assert(input_start, "input is null");
8065 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8066 assert(lowPart_start, "lowPart is null");
8067 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8068 assert(highPart_start, "highPart is null");
8069
8070 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8071 OptoRuntime::dilithiumDecomposePoly_Type(),
8072 stubAddr, stubName, TypePtr::BOTTOM,
8073 input_start, lowPart_start, highPart_start,
8074 twoGamma2, multiplier);
8075
8076 // return an int
8077 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8078 set_result(retvalue);
8079 return true;
8080 }
8081
8082 bool LibraryCallKit::inline_base64_encodeBlock() {
8083 address stubAddr;
8084 const char *stubName;
8085 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8086 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8087 stubAddr = StubRoutines::base64_encodeBlock();
8088 stubName = "encodeBlock";
8089
8090 if (!stubAddr) return false;
8091 Node* base64obj = argument(0);
8092 Node* src = argument(1);
8093 Node* offset = argument(2);
8094 Node* len = argument(3);
8095 Node* dest = argument(4);
8096 Node* dp = argument(5);
8097 Node* isURL = argument(6);
8098
8099 src = must_be_not_null(src, true);
8100 dest = must_be_not_null(dest, true);
8101
8102 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8103 assert(src_start, "source array is null");
8104 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8105 assert(dest_start, "destination array is null");
8106
8107 Node* base64 = make_runtime_call(RC_LEAF,
8108 OptoRuntime::base64_encodeBlock_Type(),
8109 stubAddr, stubName, TypePtr::BOTTOM,
8110 src_start, offset, len, dest_start, dp, isURL);
8111 return true;
8112 }
8113
8114 bool LibraryCallKit::inline_base64_decodeBlock() {
8115 address stubAddr;
8116 const char *stubName;
8117 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8118 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8119 stubAddr = StubRoutines::base64_decodeBlock();
8120 stubName = "decodeBlock";
8121
8122 if (!stubAddr) return false;
8123 Node* base64obj = argument(0);
8124 Node* src = argument(1);
8125 Node* src_offset = argument(2);
8126 Node* len = argument(3);
8127 Node* dest = argument(4);
8128 Node* dest_offset = argument(5);
8129 Node* isURL = argument(6);
8130 Node* isMIME = argument(7);
8131
8132 src = must_be_not_null(src, true);
8133 dest = must_be_not_null(dest, true);
8134
8135 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8136 assert(src_start, "source array is null");
8137 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8138 assert(dest_start, "destination array is null");
8139
8140 Node* call = make_runtime_call(RC_LEAF,
8141 OptoRuntime::base64_decodeBlock_Type(),
8142 stubAddr, stubName, TypePtr::BOTTOM,
8143 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8144 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8145 set_result(result);
8146 return true;
8147 }
8148
8149 bool LibraryCallKit::inline_poly1305_processBlocks() {
8150 address stubAddr;
8151 const char *stubName;
8152 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8153 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8154 stubAddr = StubRoutines::poly1305_processBlocks();
8155 stubName = "poly1305_processBlocks";
8156
8157 if (!stubAddr) return false;
8158 null_check_receiver(); // null-check receiver
8159 if (stopped()) return true;
8160
8161 Node* input = argument(1);
8162 Node* input_offset = argument(2);
8163 Node* len = argument(3);
8164 Node* alimbs = argument(4);
8165 Node* rlimbs = argument(5);
8166
8167 input = must_be_not_null(input, true);
8168 alimbs = must_be_not_null(alimbs, true);
8169 rlimbs = must_be_not_null(rlimbs, true);
8170
8171 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8172 assert(input_start, "input array is null");
8173 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8174 assert(acc_start, "acc array is null");
8175 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8176 assert(r_start, "r array is null");
8177
8178 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8179 OptoRuntime::poly1305_processBlocks_Type(),
8180 stubAddr, stubName, TypePtr::BOTTOM,
8181 input_start, len, acc_start, r_start);
8182 return true;
8183 }
8184
8185 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8186 address stubAddr;
8187 const char *stubName;
8188 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8189 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8190 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8191 stubName = "intpoly_montgomeryMult_P256";
8192
8193 if (!stubAddr) return false;
8194 null_check_receiver(); // null-check receiver
8195 if (stopped()) return true;
8196
8197 Node* a = argument(1);
8198 Node* b = argument(2);
8199 Node* r = argument(3);
8200
8201 a = must_be_not_null(a, true);
8202 b = must_be_not_null(b, true);
8203 r = must_be_not_null(r, true);
8204
8205 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8206 assert(a_start, "a array is null");
8207 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8208 assert(b_start, "b array is null");
8209 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8210 assert(r_start, "r array is null");
8211
8212 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8213 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8214 stubAddr, stubName, TypePtr::BOTTOM,
8215 a_start, b_start, r_start);
8216 return true;
8217 }
8218
8219 bool LibraryCallKit::inline_intpoly_assign() {
8220 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8221 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8222 const char *stubName = "intpoly_assign";
8223 address stubAddr = StubRoutines::intpoly_assign();
8224 if (!stubAddr) return false;
8225
8226 Node* set = argument(0);
8227 Node* a = argument(1);
8228 Node* b = argument(2);
8229 Node* arr_length = load_array_length(a);
8230
8231 a = must_be_not_null(a, true);
8232 b = must_be_not_null(b, true);
8233
8234 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8235 assert(a_start, "a array is null");
8236 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8237 assert(b_start, "b array is null");
8238
8239 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8240 OptoRuntime::intpoly_assign_Type(),
8241 stubAddr, stubName, TypePtr::BOTTOM,
8242 set, a_start, b_start, arr_length);
8243 return true;
8244 }
8245
8246 //------------------------------inline_digestBase_implCompress-----------------------
8247 //
8248 // Calculate MD5 for single-block byte[] array.
8249 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8250 //
8251 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8252 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8253 //
8254 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8255 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8256 //
8257 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8258 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8259 //
8260 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8261 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8262 //
8263 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8264 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8265
8266 Node* digestBase_obj = argument(0);
8267 Node* src = argument(1); // type oop
8268 Node* ofs = argument(2); // type int
8269
8270 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8271 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8272 // failed array check
8273 return false;
8274 }
8275 // Figure out the size and type of the elements we will be copying.
8276 BasicType src_elem = src_type->elem()->array_element_basic_type();
8277 if (src_elem != T_BYTE) {
8278 return false;
8279 }
8280 // 'src_start' points to src array + offset
8281 src = must_be_not_null(src, true);
8282 Node* src_start = array_element_address(src, ofs, src_elem);
8283 Node* state = nullptr;
8284 Node* block_size = nullptr;
8285 address stubAddr;
8286 const char *stubName;
8287
8288 switch(id) {
8289 case vmIntrinsics::_md5_implCompress:
8290 assert(UseMD5Intrinsics, "need MD5 instruction support");
8291 state = get_state_from_digest_object(digestBase_obj, T_INT);
8292 stubAddr = StubRoutines::md5_implCompress();
8293 stubName = "md5_implCompress";
8294 break;
8295 case vmIntrinsics::_sha_implCompress:
8296 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8297 state = get_state_from_digest_object(digestBase_obj, T_INT);
8298 stubAddr = StubRoutines::sha1_implCompress();
8299 stubName = "sha1_implCompress";
8300 break;
8301 case vmIntrinsics::_sha2_implCompress:
8302 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8303 state = get_state_from_digest_object(digestBase_obj, T_INT);
8304 stubAddr = StubRoutines::sha256_implCompress();
8305 stubName = "sha256_implCompress";
8306 break;
8307 case vmIntrinsics::_sha5_implCompress:
8308 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8309 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8310 stubAddr = StubRoutines::sha512_implCompress();
8311 stubName = "sha512_implCompress";
8312 break;
8313 case vmIntrinsics::_sha3_implCompress:
8314 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8315 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8316 stubAddr = StubRoutines::sha3_implCompress();
8317 stubName = "sha3_implCompress";
8318 block_size = get_block_size_from_digest_object(digestBase_obj);
8319 if (block_size == nullptr) return false;
8320 break;
8321 default:
8322 fatal_unexpected_iid(id);
8323 return false;
8324 }
8325 if (state == nullptr) return false;
8326
8327 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8328 if (stubAddr == nullptr) return false;
8329
8330 // Call the stub.
8331 Node* call;
8332 if (block_size == nullptr) {
8333 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8334 stubAddr, stubName, TypePtr::BOTTOM,
8335 src_start, state);
8336 } else {
8337 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8338 stubAddr, stubName, TypePtr::BOTTOM,
8339 src_start, state, block_size);
8340 }
8341
8342 return true;
8343 }
8344
8345 //------------------------------inline_double_keccak
8346 bool LibraryCallKit::inline_double_keccak() {
8347 address stubAddr;
8348 const char *stubName;
8349 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8350 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8351
8352 stubAddr = StubRoutines::double_keccak();
8353 stubName = "double_keccak";
8354 if (!stubAddr) return false;
8355
8356 Node* status0 = argument(0);
8357 Node* status1 = argument(1);
8358
8359 status0 = must_be_not_null(status0, true);
8360 status1 = must_be_not_null(status1, true);
8361
8362 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8363 assert(status0_start, "status0 is null");
8364 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8365 assert(status1_start, "status1 is null");
8366 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8367 OptoRuntime::double_keccak_Type(),
8368 stubAddr, stubName, TypePtr::BOTTOM,
8369 status0_start, status1_start);
8370 // return an int
8371 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8372 set_result(retvalue);
8373 return true;
8374 }
8375
8376
8377 //------------------------------inline_digestBase_implCompressMB-----------------------
8378 //
8379 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8380 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8381 //
8382 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8383 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8384 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8385 assert((uint)predicate < 5, "sanity");
8386 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8387
8388 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8389 Node* src = argument(1); // byte[] array
8390 Node* ofs = argument(2); // type int
8391 Node* limit = argument(3); // type int
8392
8393 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8394 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8395 // failed array check
8396 return false;
8397 }
8398 // Figure out the size and type of the elements we will be copying.
8399 BasicType src_elem = src_type->elem()->array_element_basic_type();
8400 if (src_elem != T_BYTE) {
8401 return false;
8402 }
8403 // 'src_start' points to src array + offset
8404 src = must_be_not_null(src, false);
8405 Node* src_start = array_element_address(src, ofs, src_elem);
8406
8407 const char* klass_digestBase_name = nullptr;
8408 const char* stub_name = nullptr;
8409 address stub_addr = nullptr;
8410 BasicType elem_type = T_INT;
8411
8412 switch (predicate) {
8413 case 0:
8414 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8415 klass_digestBase_name = "sun/security/provider/MD5";
8416 stub_name = "md5_implCompressMB";
8417 stub_addr = StubRoutines::md5_implCompressMB();
8418 }
8419 break;
8420 case 1:
8421 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8422 klass_digestBase_name = "sun/security/provider/SHA";
8423 stub_name = "sha1_implCompressMB";
8424 stub_addr = StubRoutines::sha1_implCompressMB();
8425 }
8426 break;
8427 case 2:
8428 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8429 klass_digestBase_name = "sun/security/provider/SHA2";
8430 stub_name = "sha256_implCompressMB";
8431 stub_addr = StubRoutines::sha256_implCompressMB();
8432 }
8433 break;
8434 case 3:
8435 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8436 klass_digestBase_name = "sun/security/provider/SHA5";
8437 stub_name = "sha512_implCompressMB";
8438 stub_addr = StubRoutines::sha512_implCompressMB();
8439 elem_type = T_LONG;
8440 }
8441 break;
8442 case 4:
8443 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8444 klass_digestBase_name = "sun/security/provider/SHA3";
8445 stub_name = "sha3_implCompressMB";
8446 stub_addr = StubRoutines::sha3_implCompressMB();
8447 elem_type = T_LONG;
8448 }
8449 break;
8450 default:
8451 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8452 }
8453 if (klass_digestBase_name != nullptr) {
8454 assert(stub_addr != nullptr, "Stub is generated");
8455 if (stub_addr == nullptr) return false;
8456
8457 // get DigestBase klass to lookup for SHA klass
8458 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8459 assert(tinst != nullptr, "digestBase_obj is not instance???");
8460 assert(tinst->is_loaded(), "DigestBase is not loaded");
8461
8462 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8463 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8464 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8465 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8466 }
8467 return false;
8468 }
8469
8470 //------------------------------inline_digestBase_implCompressMB-----------------------
8471 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8472 BasicType elem_type, address stubAddr, const char *stubName,
8473 Node* src_start, Node* ofs, Node* limit) {
8474 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8475 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8476 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8477 digest_obj = _gvn.transform(digest_obj);
8478
8479 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8480 if (state == nullptr) return false;
8481
8482 Node* block_size = nullptr;
8483 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8484 block_size = get_block_size_from_digest_object(digest_obj);
8485 if (block_size == nullptr) return false;
8486 }
8487
8488 // Call the stub.
8489 Node* call;
8490 if (block_size == nullptr) {
8491 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8492 OptoRuntime::digestBase_implCompressMB_Type(false),
8493 stubAddr, stubName, TypePtr::BOTTOM,
8494 src_start, state, ofs, limit);
8495 } else {
8496 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8497 OptoRuntime::digestBase_implCompressMB_Type(true),
8498 stubAddr, stubName, TypePtr::BOTTOM,
8499 src_start, state, block_size, ofs, limit);
8500 }
8501
8502 // return ofs (int)
8503 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8504 set_result(result);
8505
8506 return true;
8507 }
8508
8509 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8510 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8511 assert(UseAES, "need AES instruction support");
8512 address stubAddr = nullptr;
8513 const char *stubName = nullptr;
8514 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8515 stubName = "galoisCounterMode_AESCrypt";
8516
8517 if (stubAddr == nullptr) return false;
8518
8519 Node* in = argument(0);
8520 Node* inOfs = argument(1);
8521 Node* len = argument(2);
8522 Node* ct = argument(3);
8523 Node* ctOfs = argument(4);
8524 Node* out = argument(5);
8525 Node* outOfs = argument(6);
8526 Node* gctr_object = argument(7);
8527 Node* ghash_object = argument(8);
8528
8529 // (1) in, ct and out are arrays.
8530 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8531 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8532 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8533 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8534 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8535 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8536
8537 // checks are the responsibility of the caller
8538 Node* in_start = in;
8539 Node* ct_start = ct;
8540 Node* out_start = out;
8541 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8542 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8543 in_start = array_element_address(in, inOfs, T_BYTE);
8544 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8545 out_start = array_element_address(out, outOfs, T_BYTE);
8546 }
8547
8548 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8549 // (because of the predicated logic executed earlier).
8550 // so we cast it here safely.
8551 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8552 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8553 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8554 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8555 Node* state = load_field_from_object(ghash_object, "state", "[J");
8556
8557 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8558 return false;
8559 }
8560 // cast it to what we know it will be at runtime
8561 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8562 assert(tinst != nullptr, "GCTR obj is null");
8563 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8564 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8565 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8566 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8567 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8568 const TypeOopPtr* xtype = aklass->as_instance_type();
8569 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8570 aescrypt_object = _gvn.transform(aescrypt_object);
8571 // we need to get the start of the aescrypt_object's expanded key array
8572 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8573 if (k_start == nullptr) return false;
8574 // similarly, get the start address of the r vector
8575 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8576 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8577 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8578
8579
8580 // Call the stub, passing params
8581 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8582 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8583 stubAddr, stubName, TypePtr::BOTTOM,
8584 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8585
8586 // return cipher length (int)
8587 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8588 set_result(retvalue);
8589
8590 return true;
8591 }
8592
8593 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8594 // Return node representing slow path of predicate check.
8595 // the pseudo code we want to emulate with this predicate is:
8596 // for encryption:
8597 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8598 // for decryption:
8599 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8600 // note cipher==plain is more conservative than the original java code but that's OK
8601 //
8602
8603 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8604 // The receiver was checked for null already.
8605 Node* objGCTR = argument(7);
8606 // Load embeddedCipher field of GCTR object.
8607 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8608 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8609
8610 // get AESCrypt klass for instanceOf check
8611 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8612 // will have same classloader as CipherBlockChaining object
8613 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8614 assert(tinst != nullptr, "GCTR obj is null");
8615 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8616
8617 // we want to do an instanceof comparison against the AESCrypt class
8618 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8619 if (!klass_AESCrypt->is_loaded()) {
8620 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8621 Node* ctrl = control();
8622 set_control(top()); // no regular fast path
8623 return ctrl;
8624 }
8625
8626 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8627 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8628 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8629 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8630 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8631
8632 return instof_false; // even if it is null
8633 }
8634
8635 //------------------------------get_state_from_digest_object-----------------------
8636 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8637 const char* state_type;
8638 switch (elem_type) {
8639 case T_BYTE: state_type = "[B"; break;
8640 case T_INT: state_type = "[I"; break;
8641 case T_LONG: state_type = "[J"; break;
8642 default: ShouldNotReachHere();
8643 }
8644 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8645 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8646 if (digest_state == nullptr) return (Node *) nullptr;
8647
8648 // now have the array, need to get the start address of the state array
8649 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8650 return state;
8651 }
8652
8653 //------------------------------get_block_size_from_sha3_object----------------------------------
8654 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8655 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8656 assert (block_size != nullptr, "sanity");
8657 return block_size;
8658 }
8659
8660 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8661 // Return node representing slow path of predicate check.
8662 // the pseudo code we want to emulate with this predicate is:
8663 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8664 //
8665 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8666 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8667 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8668 assert((uint)predicate < 5, "sanity");
8669
8670 // The receiver was checked for null already.
8671 Node* digestBaseObj = argument(0);
8672
8673 // get DigestBase klass for instanceOf check
8674 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8675 assert(tinst != nullptr, "digestBaseObj is null");
8676 assert(tinst->is_loaded(), "DigestBase is not loaded");
8677
8678 const char* klass_name = nullptr;
8679 switch (predicate) {
8680 case 0:
8681 if (UseMD5Intrinsics) {
8682 // we want to do an instanceof comparison against the MD5 class
8683 klass_name = "sun/security/provider/MD5";
8684 }
8685 break;
8686 case 1:
8687 if (UseSHA1Intrinsics) {
8688 // we want to do an instanceof comparison against the SHA class
8689 klass_name = "sun/security/provider/SHA";
8690 }
8691 break;
8692 case 2:
8693 if (UseSHA256Intrinsics) {
8694 // we want to do an instanceof comparison against the SHA2 class
8695 klass_name = "sun/security/provider/SHA2";
8696 }
8697 break;
8698 case 3:
8699 if (UseSHA512Intrinsics) {
8700 // we want to do an instanceof comparison against the SHA5 class
8701 klass_name = "sun/security/provider/SHA5";
8702 }
8703 break;
8704 case 4:
8705 if (UseSHA3Intrinsics) {
8706 // we want to do an instanceof comparison against the SHA3 class
8707 klass_name = "sun/security/provider/SHA3";
8708 }
8709 break;
8710 default:
8711 fatal("unknown SHA intrinsic predicate: %d", predicate);
8712 }
8713
8714 ciKlass* klass = nullptr;
8715 if (klass_name != nullptr) {
8716 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8717 }
8718 if ((klass == nullptr) || !klass->is_loaded()) {
8719 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8720 Node* ctrl = control();
8721 set_control(top()); // no intrinsic path
8722 return ctrl;
8723 }
8724 ciInstanceKlass* instklass = klass->as_instance_klass();
8725
8726 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8727 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8728 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8729 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8730
8731 return instof_false; // even if it is null
8732 }
8733
8734 //-------------inline_fma-----------------------------------
8735 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8736 Node *a = nullptr;
8737 Node *b = nullptr;
8738 Node *c = nullptr;
8739 Node* result = nullptr;
8740 switch (id) {
8741 case vmIntrinsics::_fmaD:
8742 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8743 // no receiver since it is static method
8744 a = argument(0);
8745 b = argument(2);
8746 c = argument(4);
8747 result = _gvn.transform(new FmaDNode(a, b, c));
8748 break;
8749 case vmIntrinsics::_fmaF:
8750 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8751 a = argument(0);
8752 b = argument(1);
8753 c = argument(2);
8754 result = _gvn.transform(new FmaFNode(a, b, c));
8755 break;
8756 default:
8757 fatal_unexpected_iid(id); break;
8758 }
8759 set_result(result);
8760 return true;
8761 }
8762
8763 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8764 // argument(0) is receiver
8765 Node* codePoint = argument(1);
8766 Node* n = nullptr;
8767
8768 switch (id) {
8769 case vmIntrinsics::_isDigit :
8770 n = new DigitNode(control(), codePoint);
8771 break;
8772 case vmIntrinsics::_isLowerCase :
8773 n = new LowerCaseNode(control(), codePoint);
8774 break;
8775 case vmIntrinsics::_isUpperCase :
8776 n = new UpperCaseNode(control(), codePoint);
8777 break;
8778 case vmIntrinsics::_isWhitespace :
8779 n = new WhitespaceNode(control(), codePoint);
8780 break;
8781 default:
8782 fatal_unexpected_iid(id);
8783 }
8784
8785 set_result(_gvn.transform(n));
8786 return true;
8787 }
8788
8789 bool LibraryCallKit::inline_profileBoolean() {
8790 Node* counts = argument(1);
8791 const TypeAryPtr* ary = nullptr;
8792 ciArray* aobj = nullptr;
8793 if (counts->is_Con()
8794 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8795 && (aobj = ary->const_oop()->as_array()) != nullptr
8796 && (aobj->length() == 2)) {
8797 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8798 jint false_cnt = aobj->element_value(0).as_int();
8799 jint true_cnt = aobj->element_value(1).as_int();
8800
8801 if (C->log() != nullptr) {
8802 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8803 false_cnt, true_cnt);
8804 }
8805
8806 if (false_cnt + true_cnt == 0) {
8807 // According to profile, never executed.
8808 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8809 Deoptimization::Action_reinterpret);
8810 return true;
8811 }
8812
8813 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8814 // is a number of each value occurrences.
8815 Node* result = argument(0);
8816 if (false_cnt == 0 || true_cnt == 0) {
8817 // According to profile, one value has been never seen.
8818 int expected_val = (false_cnt == 0) ? 1 : 0;
8819
8820 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8821 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8822
8823 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8824 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8825 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8826
8827 { // Slow path: uncommon trap for never seen value and then reexecute
8828 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8829 // the value has been seen at least once.
8830 PreserveJVMState pjvms(this);
8831 PreserveReexecuteState preexecs(this);
8832 jvms()->set_should_reexecute(true);
8833
8834 set_control(slow_path);
8835 set_i_o(i_o());
8836
8837 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8838 Deoptimization::Action_reinterpret);
8839 }
8840 // The guard for never seen value enables sharpening of the result and
8841 // returning a constant. It allows to eliminate branches on the same value
8842 // later on.
8843 set_control(fast_path);
8844 result = intcon(expected_val);
8845 }
8846 // Stop profiling.
8847 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8848 // By replacing method body with profile data (represented as ProfileBooleanNode
8849 // on IR level) we effectively disable profiling.
8850 // It enables full speed execution once optimized code is generated.
8851 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8852 C->record_for_igvn(profile);
8853 set_result(profile);
8854 return true;
8855 } else {
8856 // Continue profiling.
8857 // Profile data isn't available at the moment. So, execute method's bytecode version.
8858 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8859 // is compiled and counters aren't available since corresponding MethodHandle
8860 // isn't a compile-time constant.
8861 return false;
8862 }
8863 }
8864
8865 bool LibraryCallKit::inline_isCompileConstant() {
8866 Node* n = argument(0);
8867 set_result(n->is_Con() ? intcon(1) : intcon(0));
8868 return true;
8869 }
8870
8871 //------------------------------- inline_getObjectSize --------------------------------------
8872 //
8873 // Calculate the runtime size of the object/array.
8874 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8875 //
8876 bool LibraryCallKit::inline_getObjectSize() {
8877 Node* obj = argument(3);
8878 Node* klass_node = load_object_klass(obj);
8879
8880 jint layout_con = Klass::_lh_neutral_value;
8881 Node* layout_val = get_layout_helper(klass_node, layout_con);
8882 int layout_is_con = (layout_val == nullptr);
8883
8884 if (layout_is_con) {
8885 // Layout helper is constant, can figure out things at compile time.
8886
8887 if (Klass::layout_helper_is_instance(layout_con)) {
8888 // Instance case: layout_con contains the size itself.
8889 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8890 set_result(size);
8891 } else {
8892 // Array case: size is round(header + element_size*arraylength).
8893 // Since arraylength is different for every array instance, we have to
8894 // compute the whole thing at runtime.
8895
8896 Node* arr_length = load_array_length(obj);
8897
8898 int round_mask = MinObjAlignmentInBytes - 1;
8899 int hsize = Klass::layout_helper_header_size(layout_con);
8900 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8901
8902 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8903 round_mask = 0; // strength-reduce it if it goes away completely
8904 }
8905 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8906 Node* header_size = intcon(hsize + round_mask);
8907
8908 Node* lengthx = ConvI2X(arr_length);
8909 Node* headerx = ConvI2X(header_size);
8910
8911 Node* abody = lengthx;
8912 if (eshift != 0) {
8913 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8914 }
8915 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8916 if (round_mask != 0) {
8917 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8918 }
8919 size = ConvX2L(size);
8920 set_result(size);
8921 }
8922 } else {
8923 // Layout helper is not constant, need to test for array-ness at runtime.
8924
8925 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8926 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8927 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8928 record_for_igvn(result_reg);
8929
8930 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8931 if (array_ctl != nullptr) {
8932 // Array case: size is round(header + element_size*arraylength).
8933 // Since arraylength is different for every array instance, we have to
8934 // compute the whole thing at runtime.
8935
8936 PreserveJVMState pjvms(this);
8937 set_control(array_ctl);
8938 Node* arr_length = load_array_length(obj);
8939
8940 int round_mask = MinObjAlignmentInBytes - 1;
8941 Node* mask = intcon(round_mask);
8942
8943 Node* hss = intcon(Klass::_lh_header_size_shift);
8944 Node* hsm = intcon(Klass::_lh_header_size_mask);
8945 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8946 header_size = _gvn.transform(new AndINode(header_size, hsm));
8947 header_size = _gvn.transform(new AddINode(header_size, mask));
8948
8949 // There is no need to mask or shift this value.
8950 // The semantics of LShiftINode include an implicit mask to 0x1F.
8951 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8952 Node* elem_shift = layout_val;
8953
8954 Node* lengthx = ConvI2X(arr_length);
8955 Node* headerx = ConvI2X(header_size);
8956
8957 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8958 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8959 if (round_mask != 0) {
8960 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8961 }
8962 size = ConvX2L(size);
8963
8964 result_reg->init_req(_array_path, control());
8965 result_val->init_req(_array_path, size);
8966 }
8967
8968 if (!stopped()) {
8969 // Instance case: the layout helper gives us instance size almost directly,
8970 // but we need to mask out the _lh_instance_slow_path_bit.
8971 Node* size = ConvI2X(layout_val);
8972 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8973 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8974 size = _gvn.transform(new AndXNode(size, mask));
8975 size = ConvX2L(size);
8976
8977 result_reg->init_req(_instance_path, control());
8978 result_val->init_req(_instance_path, size);
8979 }
8980
8981 set_result(result_reg, result_val);
8982 }
8983
8984 return true;
8985 }
8986
8987 //------------------------------- inline_blackhole --------------------------------------
8988 //
8989 // Make sure all arguments to this node are alive.
8990 // This matches methods that were requested to be blackholed through compile commands.
8991 //
8992 bool LibraryCallKit::inline_blackhole() {
8993 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8994 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8995 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8996
8997 // Blackhole node pinches only the control, not memory. This allows
8998 // the blackhole to be pinned in the loop that computes blackholed
8999 // values, but have no other side effects, like breaking the optimizations
9000 // across the blackhole.
9001
9002 Node* bh = _gvn.transform(new BlackholeNode(control()));
9003 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9004
9005 // Bind call arguments as blackhole arguments to keep them alive
9006 uint nargs = callee()->arg_size();
9007 for (uint i = 0; i < nargs; i++) {
9008 bh->add_req(argument(i));
9009 }
9010
9011 return true;
9012 }
9013
9014 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9015 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9016 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9017 return nullptr; // box klass is not Float16
9018 }
9019
9020 // Null check; get notnull casted pointer
9021 Node* null_ctl = top();
9022 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9023 // If not_null_box is dead, only null-path is taken
9024 if (stopped()) {
9025 set_control(null_ctl);
9026 return nullptr;
9027 }
9028 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9029 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9030 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9031 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9032 }
9033
9034 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9035 PreserveReexecuteState preexecs(this);
9036 jvms()->set_should_reexecute(true);
9037
9038 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9039 Node* klass_node = makecon(klass_type);
9040 Node* box = new_instance(klass_node);
9041
9042 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9043 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9044
9045 Node* field_store = _gvn.transform(access_store_at(box,
9046 value_field,
9047 value_adr_type,
9048 value,
9049 TypeInt::SHORT,
9050 T_SHORT,
9051 IN_HEAP));
9052 set_memory(field_store, value_adr_type);
9053 return box;
9054 }
9055
9056 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9057 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9058 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9059 return false;
9060 }
9061
9062 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9063 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9064 return false;
9065 }
9066
9067 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9068 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9069 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9070 ciSymbols::short_signature(),
9071 false);
9072 assert(field != nullptr, "");
9073
9074 // Transformed nodes
9075 Node* fld1 = nullptr;
9076 Node* fld2 = nullptr;
9077 Node* fld3 = nullptr;
9078 switch(num_args) {
9079 case 3:
9080 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9081 if (fld3 == nullptr) {
9082 return false;
9083 }
9084 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9085 // fall-through
9086 case 2:
9087 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9088 if (fld2 == nullptr) {
9089 return false;
9090 }
9091 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9092 // fall-through
9093 case 1:
9094 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9095 if (fld1 == nullptr) {
9096 return false;
9097 }
9098 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9099 break;
9100 default: fatal("Unsupported number of arguments %d", num_args);
9101 }
9102
9103 Node* result = nullptr;
9104 switch (id) {
9105 // Unary operations
9106 case vmIntrinsics::_sqrt_float16:
9107 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9108 break;
9109 // Ternary operations
9110 case vmIntrinsics::_fma_float16:
9111 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9112 break;
9113 default:
9114 fatal_unexpected_iid(id);
9115 break;
9116 }
9117 result = _gvn.transform(new ReinterpretHF2SNode(result));
9118 set_result(box_fp16_value(float16_box_type, field, result));
9119 return true;
9120 }
9121