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/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciSymbols.hpp"
30 #include "ci/ciUtilities.inline.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/inlinetypenode.hpp"
52 #include "opto/library_call.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/rootnode.hpp"
60 #include "opto/runtime.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/globals.hpp"
68 #include "runtime/jniHandles.inline.hpp"
69 #include "runtime/mountUnmountDisabler.hpp"
70 #include "runtime/objectMonitor.hpp"
71 #include "runtime/sharedRuntime.hpp"
72 #include "runtime/stubRoutines.hpp"
73 #include "utilities/globalDefinitions.hpp"
74 #include "utilities/macros.hpp"
75 #include "utilities/powerOfTwo.hpp"
76
77 //---------------------------make_vm_intrinsic----------------------------
78 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
79 vmIntrinsicID id = m->intrinsic_id();
80 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
81
82 if (!m->is_loaded()) {
83 // Do not attempt to inline unloaded methods.
84 return nullptr;
85 }
86
87 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
88 bool is_available = false;
89
90 {
91 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
92 // the compiler must transition to '_thread_in_vm' state because both
93 // methods access VM-internal data.
94 VM_ENTRY_MARK;
95 methodHandle mh(THREAD, m->get_Method());
96 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
97 if (is_available && is_virtual) {
98 is_available = vmIntrinsics::does_virtual_dispatch(id);
99 }
100 }
101
102 if (is_available) {
103 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
104 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
105 return new LibraryIntrinsic(m, is_virtual,
106 vmIntrinsics::predicates_needed(id),
107 vmIntrinsics::does_virtual_dispatch(id),
108 id);
109 } else {
110 return nullptr;
111 }
112 }
113
114 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
115 LibraryCallKit kit(jvms, this);
116 Compile* C = kit.C;
117 int nodes = C->unique();
118 #ifndef PRODUCT
119 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
120 char buf[1000];
121 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
122 tty->print_cr("Intrinsic %s", str);
123 }
124 #endif
125 ciMethod* callee = kit.callee();
126 const int bci = kit.bci();
127 #ifdef ASSERT
128 Node* ctrl = kit.control();
129 #endif
130 // Try to inline the intrinsic.
131 if (callee->check_intrinsic_candidate() &&
132 kit.try_to_inline(_last_predicate)) {
133 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
134 : "(intrinsic)";
135 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
136 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
137 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
138 if (C->log()) {
139 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
140 vmIntrinsics::name_at(intrinsic_id()),
141 (is_virtual() ? " virtual='1'" : ""),
142 C->unique() - nodes);
143 }
144 // Push the result from the inlined method onto the stack.
145 kit.push_result();
146 return kit.transfer_exceptions_into_jvms();
147 }
148
149 // The intrinsic bailed out
150 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
151 assert(jvms->map() == kit.map(), "Out of sync JVM state");
152 if (jvms->has_method()) {
153 // Not a root compile.
154 const char* msg;
155 if (callee->intrinsic_candidate()) {
156 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
157 } else {
158 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
159 : "failed to inline (intrinsic), method not annotated";
160 }
161 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
162 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
163 } else {
164 // Root compile
165 ResourceMark rm;
166 stringStream msg_stream;
167 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
168 vmIntrinsics::name_at(intrinsic_id()),
169 is_virtual() ? " (virtual)" : "", bci);
170 const char *msg = msg_stream.freeze();
171 log_debug(jit, inlining)("%s", msg);
172 if (C->print_intrinsics() || C->print_inlining()) {
173 tty->print("%s", msg);
174 }
175 }
176 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
177
178 return nullptr;
179 }
180
181 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
182 LibraryCallKit kit(jvms, this);
183 Compile* C = kit.C;
184 int nodes = C->unique();
185 _last_predicate = predicate;
186 #ifndef PRODUCT
187 assert(is_predicated() && predicate < predicates_count(), "sanity");
188 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
189 char buf[1000];
190 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
191 tty->print_cr("Predicate for intrinsic %s", str);
192 }
193 #endif
194 ciMethod* callee = kit.callee();
195 const int bci = kit.bci();
196
197 Node* slow_ctl = kit.try_to_predicate(predicate);
198 if (!kit.failing()) {
199 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
200 : "(intrinsic, predicate)";
201 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
202 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
203
204 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
205 if (C->log()) {
206 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
207 vmIntrinsics::name_at(intrinsic_id()),
208 (is_virtual() ? " virtual='1'" : ""),
209 C->unique() - nodes);
210 }
211 return slow_ctl; // Could be null if the check folds.
212 }
213
214 // The intrinsic bailed out
215 if (jvms->has_method()) {
216 // Not a root compile.
217 const char* msg = "failed to generate predicate for intrinsic";
218 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
219 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
220 } else {
221 // Root compile
222 ResourceMark rm;
223 stringStream msg_stream;
224 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
225 vmIntrinsics::name_at(intrinsic_id()),
226 is_virtual() ? " (virtual)" : "", bci);
227 const char *msg = msg_stream.freeze();
228 log_debug(jit, inlining)("%s", msg);
229 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
230 }
231 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
232 return nullptr;
233 }
234
235 bool LibraryCallKit::try_to_inline(int predicate) {
236 // Handle symbolic names for otherwise undistinguished boolean switches:
237 const bool is_store = true;
238 const bool is_compress = true;
239 const bool is_static = true;
240 const bool is_volatile = true;
241
242 if (!jvms()->has_method()) {
243 // Root JVMState has a null method.
244 assert(map()->memory()->Opcode() == Op_Parm, "");
245 // Insert the memory aliasing node
246 set_all_memory(reset_memory());
247 }
248 assert(merged_memory(), "");
249
250 switch (intrinsic_id()) {
251 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
252 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
253 case vmIntrinsics::_getClass: return inline_native_getClass();
254
255 case vmIntrinsics::_ceil:
256 case vmIntrinsics::_floor:
257 case vmIntrinsics::_rint:
258 case vmIntrinsics::_dsin:
259 case vmIntrinsics::_dcos:
260 case vmIntrinsics::_dtan:
261 case vmIntrinsics::_dsinh:
262 case vmIntrinsics::_dtanh:
263 case vmIntrinsics::_dcbrt:
264 case vmIntrinsics::_dabs:
265 case vmIntrinsics::_fabs:
266 case vmIntrinsics::_iabs:
267 case vmIntrinsics::_labs:
268 case vmIntrinsics::_datan2:
269 case vmIntrinsics::_dsqrt:
270 case vmIntrinsics::_dsqrt_strict:
271 case vmIntrinsics::_dexp:
272 case vmIntrinsics::_dlog:
273 case vmIntrinsics::_dlog10:
274 case vmIntrinsics::_dpow:
275 case vmIntrinsics::_dcopySign:
276 case vmIntrinsics::_fcopySign:
277 case vmIntrinsics::_dsignum:
278 case vmIntrinsics::_roundF:
279 case vmIntrinsics::_roundD:
280 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
281
282 case vmIntrinsics::_notify:
283 case vmIntrinsics::_notifyAll:
284 return inline_notify(intrinsic_id());
285
286 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
287 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
288 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
289 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
290 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
291 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
292 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
293 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
294 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
295 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
296 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
297 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
298 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
299 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
300
301 case vmIntrinsics::_arraycopy: return inline_arraycopy();
302
303 case vmIntrinsics::_arraySort: return inline_array_sort();
304 case vmIntrinsics::_arrayPartition: return inline_array_partition();
305
306 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
307 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
308 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
309 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
310
311 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
312 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
313 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
314 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
315 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
316 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
317 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
318 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
319
320 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
321
322 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
323
324 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
325 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
326 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
327 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
328
329 case vmIntrinsics::_compressStringC:
330 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
331 case vmIntrinsics::_inflateStringC:
332 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
333
334 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
335 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
336 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
337 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
338 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
339 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
340 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
341 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
342 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
343 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
344 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
345
346 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
347 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
348 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
349 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
350 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
351 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
352 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
353 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
354 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
355
356 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
357 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
358 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
359 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
360 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
361 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
362 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
363 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
364 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
365
366 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
367 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
368 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
369 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
370 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
371 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
372 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
373 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
374 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
375
376 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
377 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
378 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
379 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
380
381 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
382 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
383 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
384 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
385
386 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
387 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
388 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
389 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
390 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
391 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
392 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
393 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
394 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
395
396 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
397 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
398 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
399 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
400 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
401 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
402 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
403 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
404 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
405
406 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
407 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
408 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
409 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
410 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
411 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
412 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
413 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
414 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
415
416 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
417 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
418 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
419 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
420 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
421 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
422 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
423 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
424 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
425
426 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
427 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
428
429 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
432 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
433 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
434
435 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
443 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
444 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
445 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
446 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
447 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
448 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
449 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
450 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
451 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
452 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
453 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
454 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
455
456 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
465 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
466 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
467 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
468 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
469 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
470 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
471
472 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
474 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
475 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
476
477 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
480 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
481 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
482
483 case vmIntrinsics::_loadFence:
484 case vmIntrinsics::_storeFence:
485 case vmIntrinsics::_storeStoreFence:
486 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
487
488 case vmIntrinsics::_arrayInstanceBaseOffset: return inline_arrayInstanceBaseOffset();
489 case vmIntrinsics::_arrayInstanceIndexScale: return inline_arrayInstanceIndexScale();
490 case vmIntrinsics::_arrayLayout: return inline_arrayLayout();
491 case vmIntrinsics::_getFieldMap: return inline_getFieldMap();
492
493 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
494
495 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
496 case vmIntrinsics::_currentThread: return inline_native_currentThread();
497 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
498
499 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
500 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
501
502 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
503 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
504
505 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
506 "endFirstTransition", true);
507 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
508 "startFinalTransition", true);
509 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
510 "startTransition", false);
511 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
512 "endTransition", false);
513 #if INCLUDE_JVMTI
514 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
515 #endif
516
517 #ifdef JFR_HAVE_INTRINSICS
518 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
519 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
520 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
521 #endif
522 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
523 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
524 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
525 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
526 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
527 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
528 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
529 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
530 case vmIntrinsics::_getLength: return inline_native_getLength();
531 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
532 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
533 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
534 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
535 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
536 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
537 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
538
539 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
540 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
541 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
542 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
543 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
544 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
545 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
546 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
547
548 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
549
550 case vmIntrinsics::_isInstance:
551 case vmIntrinsics::_isHidden:
552 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
553
554 case vmIntrinsics::_floatToRawIntBits:
555 case vmIntrinsics::_floatToIntBits:
556 case vmIntrinsics::_intBitsToFloat:
557 case vmIntrinsics::_doubleToRawLongBits:
558 case vmIntrinsics::_doubleToLongBits:
559 case vmIntrinsics::_longBitsToDouble:
560 case vmIntrinsics::_floatToFloat16:
561 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
562 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
563 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
564 case vmIntrinsics::_floatIsFinite:
565 case vmIntrinsics::_floatIsInfinite:
566 case vmIntrinsics::_doubleIsFinite:
567 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
568
569 case vmIntrinsics::_numberOfLeadingZeros_i:
570 case vmIntrinsics::_numberOfLeadingZeros_l:
571 case vmIntrinsics::_numberOfTrailingZeros_i:
572 case vmIntrinsics::_numberOfTrailingZeros_l:
573 case vmIntrinsics::_bitCount_i:
574 case vmIntrinsics::_bitCount_l:
575 case vmIntrinsics::_reverse_i:
576 case vmIntrinsics::_reverse_l:
577 case vmIntrinsics::_reverseBytes_i:
578 case vmIntrinsics::_reverseBytes_l:
579 case vmIntrinsics::_reverseBytes_s:
580 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
581
582 case vmIntrinsics::_compress_i:
583 case vmIntrinsics::_compress_l:
584 case vmIntrinsics::_expand_i:
585 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
586
587 case vmIntrinsics::_compareUnsigned_i:
588 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
589
590 case vmIntrinsics::_divideUnsigned_i:
591 case vmIntrinsics::_divideUnsigned_l:
592 case vmIntrinsics::_remainderUnsigned_i:
593 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
594
595 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
596
597 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
598 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
599 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
600 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
601 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
602
603 case vmIntrinsics::_Class_cast: return inline_Class_cast();
604
605 case vmIntrinsics::_aescrypt_encryptBlock:
606 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
607
608 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
609 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
610 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
611
612 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
613 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
614 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
615
616 case vmIntrinsics::_counterMode_AESCrypt:
617 return inline_counterMode_AESCrypt(intrinsic_id());
618
619 case vmIntrinsics::_galoisCounterMode_AESCrypt:
620 return inline_galoisCounterMode_AESCrypt();
621
622 case vmIntrinsics::_md5_implCompress:
623 case vmIntrinsics::_sha_implCompress:
624 case vmIntrinsics::_sha2_implCompress:
625 case vmIntrinsics::_sha5_implCompress:
626 case vmIntrinsics::_sha3_implCompress:
627 return inline_digestBase_implCompress(intrinsic_id());
628 case vmIntrinsics::_double_keccak:
629 return inline_double_keccak();
630
631 case vmIntrinsics::_digestBase_implCompressMB:
632 return inline_digestBase_implCompressMB(predicate);
633
634 case vmIntrinsics::_multiplyToLen:
635 return inline_multiplyToLen();
636
637 case vmIntrinsics::_squareToLen:
638 return inline_squareToLen();
639
640 case vmIntrinsics::_mulAdd:
641 return inline_mulAdd();
642
643 case vmIntrinsics::_montgomeryMultiply:
644 return inline_montgomeryMultiply();
645 case vmIntrinsics::_montgomerySquare:
646 return inline_montgomerySquare();
647
648 case vmIntrinsics::_bigIntegerRightShiftWorker:
649 return inline_bigIntegerShift(true);
650 case vmIntrinsics::_bigIntegerLeftShiftWorker:
651 return inline_bigIntegerShift(false);
652
653 case vmIntrinsics::_vectorizedMismatch:
654 return inline_vectorizedMismatch();
655
656 case vmIntrinsics::_ghash_processBlocks:
657 return inline_ghash_processBlocks();
658 case vmIntrinsics::_chacha20Block:
659 return inline_chacha20Block();
660 case vmIntrinsics::_kyberNtt:
661 return inline_kyberNtt();
662 case vmIntrinsics::_kyberInverseNtt:
663 return inline_kyberInverseNtt();
664 case vmIntrinsics::_kyberNttMult:
665 return inline_kyberNttMult();
666 case vmIntrinsics::_kyberAddPoly_2:
667 return inline_kyberAddPoly_2();
668 case vmIntrinsics::_kyberAddPoly_3:
669 return inline_kyberAddPoly_3();
670 case vmIntrinsics::_kyber12To16:
671 return inline_kyber12To16();
672 case vmIntrinsics::_kyberBarrettReduce:
673 return inline_kyberBarrettReduce();
674 case vmIntrinsics::_dilithiumAlmostNtt:
675 return inline_dilithiumAlmostNtt();
676 case vmIntrinsics::_dilithiumAlmostInverseNtt:
677 return inline_dilithiumAlmostInverseNtt();
678 case vmIntrinsics::_dilithiumNttMult:
679 return inline_dilithiumNttMult();
680 case vmIntrinsics::_dilithiumMontMulByConstant:
681 return inline_dilithiumMontMulByConstant();
682 case vmIntrinsics::_dilithiumDecomposePoly:
683 return inline_dilithiumDecomposePoly();
684 case vmIntrinsics::_base64_encodeBlock:
685 return inline_base64_encodeBlock();
686 case vmIntrinsics::_base64_decodeBlock:
687 return inline_base64_decodeBlock();
688 case vmIntrinsics::_poly1305_processBlocks:
689 return inline_poly1305_processBlocks();
690 case vmIntrinsics::_intpoly_montgomeryMult_P256:
691 return inline_intpoly_montgomeryMult_P256();
692 case vmIntrinsics::_intpoly_assign:
693 return inline_intpoly_assign();
694 case vmIntrinsics::_encodeISOArray:
695 case vmIntrinsics::_encodeByteISOArray:
696 return inline_encodeISOArray(false);
697 case vmIntrinsics::_encodeAsciiArray:
698 return inline_encodeISOArray(true);
699
700 case vmIntrinsics::_updateCRC32:
701 return inline_updateCRC32();
702 case vmIntrinsics::_updateBytesCRC32:
703 return inline_updateBytesCRC32();
704 case vmIntrinsics::_updateByteBufferCRC32:
705 return inline_updateByteBufferCRC32();
706
707 case vmIntrinsics::_updateBytesCRC32C:
708 return inline_updateBytesCRC32C();
709 case vmIntrinsics::_updateDirectByteBufferCRC32C:
710 return inline_updateDirectByteBufferCRC32C();
711
712 case vmIntrinsics::_updateBytesAdler32:
713 return inline_updateBytesAdler32();
714 case vmIntrinsics::_updateByteBufferAdler32:
715 return inline_updateByteBufferAdler32();
716
717 case vmIntrinsics::_profileBoolean:
718 return inline_profileBoolean();
719 case vmIntrinsics::_isCompileConstant:
720 return inline_isCompileConstant();
721
722 case vmIntrinsics::_countPositives:
723 return inline_countPositives();
724
725 case vmIntrinsics::_fmaD:
726 case vmIntrinsics::_fmaF:
727 return inline_fma(intrinsic_id());
728
729 case vmIntrinsics::_isDigit:
730 case vmIntrinsics::_isLowerCase:
731 case vmIntrinsics::_isUpperCase:
732 case vmIntrinsics::_isWhitespace:
733 return inline_character_compare(intrinsic_id());
734
735 case vmIntrinsics::_min:
736 case vmIntrinsics::_max:
737 case vmIntrinsics::_min_strict:
738 case vmIntrinsics::_max_strict:
739 case vmIntrinsics::_minL:
740 case vmIntrinsics::_maxL:
741 case vmIntrinsics::_minF:
742 case vmIntrinsics::_maxF:
743 case vmIntrinsics::_minD:
744 case vmIntrinsics::_maxD:
745 case vmIntrinsics::_minF_strict:
746 case vmIntrinsics::_maxF_strict:
747 case vmIntrinsics::_minD_strict:
748 case vmIntrinsics::_maxD_strict:
749 return inline_min_max(intrinsic_id());
750
751 case vmIntrinsics::_VectorUnaryOp:
752 return inline_vector_nary_operation(1);
753 case vmIntrinsics::_VectorBinaryOp:
754 return inline_vector_nary_operation(2);
755 case vmIntrinsics::_VectorUnaryLibOp:
756 return inline_vector_call(1);
757 case vmIntrinsics::_VectorBinaryLibOp:
758 return inline_vector_call(2);
759 case vmIntrinsics::_VectorTernaryOp:
760 return inline_vector_nary_operation(3);
761 case vmIntrinsics::_VectorFromBitsCoerced:
762 return inline_vector_frombits_coerced();
763 case vmIntrinsics::_VectorMaskOp:
764 return inline_vector_mask_operation();
765 case vmIntrinsics::_VectorLoadOp:
766 return inline_vector_mem_operation(/*is_store=*/false);
767 case vmIntrinsics::_VectorLoadMaskedOp:
768 return inline_vector_mem_masked_operation(/*is_store*/false);
769 case vmIntrinsics::_VectorStoreOp:
770 return inline_vector_mem_operation(/*is_store=*/true);
771 case vmIntrinsics::_VectorStoreMaskedOp:
772 return inline_vector_mem_masked_operation(/*is_store=*/true);
773 case vmIntrinsics::_VectorGatherOp:
774 return inline_vector_gather_scatter(/*is_scatter*/ false);
775 case vmIntrinsics::_VectorScatterOp:
776 return inline_vector_gather_scatter(/*is_scatter*/ true);
777 case vmIntrinsics::_VectorReductionCoerced:
778 return inline_vector_reduction();
779 case vmIntrinsics::_VectorTest:
780 return inline_vector_test();
781 case vmIntrinsics::_VectorBlend:
782 return inline_vector_blend();
783 case vmIntrinsics::_VectorRearrange:
784 return inline_vector_rearrange();
785 case vmIntrinsics::_VectorSelectFrom:
786 return inline_vector_select_from();
787 case vmIntrinsics::_VectorCompare:
788 return inline_vector_compare();
789 case vmIntrinsics::_VectorBroadcastInt:
790 return inline_vector_broadcast_int();
791 case vmIntrinsics::_VectorConvert:
792 return inline_vector_convert();
793 case vmIntrinsics::_VectorInsert:
794 return inline_vector_insert();
795 case vmIntrinsics::_VectorExtract:
796 return inline_vector_extract();
797 case vmIntrinsics::_VectorCompressExpand:
798 return inline_vector_compress_expand();
799 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
800 return inline_vector_select_from_two_vectors();
801 case vmIntrinsics::_IndexVector:
802 return inline_index_vector();
803 case vmIntrinsics::_IndexPartiallyInUpperRange:
804 return inline_index_partially_in_upper_range();
805
806 case vmIntrinsics::_getObjectSize:
807 return inline_getObjectSize();
808
809 case vmIntrinsics::_blackhole:
810 return inline_blackhole();
811
812 default:
813 // If you get here, it may be that someone has added a new intrinsic
814 // to the list in vmIntrinsics.hpp without implementing it here.
815 #ifndef PRODUCT
816 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
817 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
818 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
819 }
820 #endif
821 return false;
822 }
823 }
824
825 Node* LibraryCallKit::try_to_predicate(int predicate) {
826 if (!jvms()->has_method()) {
827 // Root JVMState has a null method.
828 assert(map()->memory()->Opcode() == Op_Parm, "");
829 // Insert the memory aliasing node
830 set_all_memory(reset_memory());
831 }
832 assert(merged_memory(), "");
833
834 switch (intrinsic_id()) {
835 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
836 return inline_cipherBlockChaining_AESCrypt_predicate(false);
837 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
838 return inline_cipherBlockChaining_AESCrypt_predicate(true);
839 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
840 return inline_electronicCodeBook_AESCrypt_predicate(false);
841 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
842 return inline_electronicCodeBook_AESCrypt_predicate(true);
843 case vmIntrinsics::_counterMode_AESCrypt:
844 return inline_counterMode_AESCrypt_predicate();
845 case vmIntrinsics::_digestBase_implCompressMB:
846 return inline_digestBase_implCompressMB_predicate(predicate);
847 case vmIntrinsics::_galoisCounterMode_AESCrypt:
848 return inline_galoisCounterMode_AESCrypt_predicate();
849
850 default:
851 // If you get here, it may be that someone has added a new intrinsic
852 // to the list in vmIntrinsics.hpp without implementing it here.
853 #ifndef PRODUCT
854 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
855 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
856 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
857 }
858 #endif
859 Node* slow_ctl = control();
860 set_control(top()); // No fast path intrinsic
861 return slow_ctl;
862 }
863 }
864
865 //------------------------------set_result-------------------------------
866 // Helper function for finishing intrinsics.
867 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
868 record_for_igvn(region);
869 set_control(_gvn.transform(region));
870 set_result( _gvn.transform(value));
871 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
872 }
873
874 //------------------------------generate_guard---------------------------
875 // Helper function for generating guarded fast-slow graph structures.
876 // The given 'test', if true, guards a slow path. If the test fails
877 // then a fast path can be taken. (We generally hope it fails.)
878 // In all cases, GraphKit::control() is updated to the fast path.
879 // The returned value represents the control for the slow path.
880 // The return value is never 'top'; it is either a valid control
881 // or null if it is obvious that the slow path can never be taken.
882 // Also, if region and the slow control are not null, the slow edge
883 // is appended to the region.
884 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
885 if (stopped()) {
886 // Already short circuited.
887 return nullptr;
888 }
889
890 // Build an if node and its projections.
891 // If test is true we take the slow path, which we assume is uncommon.
892 if (_gvn.type(test) == TypeInt::ZERO) {
893 // The slow branch is never taken. No need to build this guard.
894 return nullptr;
895 }
896
897 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
898
899 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
900 if (if_slow == top()) {
901 // The slow branch is never taken. No need to build this guard.
902 return nullptr;
903 }
904
905 if (region != nullptr)
906 region->add_req(if_slow);
907
908 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
909 set_control(if_fast);
910
911 return if_slow;
912 }
913
914 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
915 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
916 }
917 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
918 return generate_guard(test, region, PROB_FAIR);
919 }
920
921 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
922 Node* *pos_index) {
923 if (stopped())
924 return nullptr; // already stopped
925 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
926 return nullptr; // index is already adequately typed
927 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
928 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
929 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
930 if (is_neg != nullptr && pos_index != nullptr) {
931 // Emulate effect of Parse::adjust_map_after_if.
932 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
933 (*pos_index) = _gvn.transform(ccast);
934 }
935 return is_neg;
936 }
937
938 // Make sure that 'position' is a valid limit index, in [0..length].
939 // There are two equivalent plans for checking this:
940 // A. (offset + copyLength) unsigned<= arrayLength
941 // B. offset <= (arrayLength - copyLength)
942 // We require that all of the values above, except for the sum and
943 // difference, are already known to be non-negative.
944 // Plan A is robust in the face of overflow, if offset and copyLength
945 // are both hugely positive.
946 //
947 // Plan B is less direct and intuitive, but it does not overflow at
948 // all, since the difference of two non-negatives is always
949 // representable. Whenever Java methods must perform the equivalent
950 // check they generally use Plan B instead of Plan A.
951 // For the moment we use Plan A.
952 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
953 Node* subseq_length,
954 Node* array_length,
955 RegionNode* region) {
956 if (stopped())
957 return nullptr; // already stopped
958 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
959 if (zero_offset && subseq_length->eqv_uncast(array_length))
960 return nullptr; // common case of whole-array copy
961 Node* last = subseq_length;
962 if (!zero_offset) // last += offset
963 last = _gvn.transform(new AddINode(last, offset));
964 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
965 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
966 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
967 return is_over;
968 }
969
970 // Emit range checks for the given String.value byte array
971 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
972 if (stopped()) {
973 return; // already stopped
974 }
975 RegionNode* bailout = new RegionNode(1);
976 record_for_igvn(bailout);
977 if (char_count) {
978 // Convert char count to byte count
979 count = _gvn.transform(new LShiftINode(count, intcon(1)));
980 }
981
982 // Offset and count must not be negative
983 generate_negative_guard(offset, bailout);
984 generate_negative_guard(count, bailout);
985 // Offset + count must not exceed length of array
986 generate_limit_guard(offset, count, load_array_length(array), bailout);
987
988 if (bailout->req() > 1) {
989 PreserveJVMState pjvms(this);
990 set_control(_gvn.transform(bailout));
991 uncommon_trap(Deoptimization::Reason_intrinsic,
992 Deoptimization::Action_maybe_recompile);
993 }
994 }
995
996 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
997 bool is_immutable) {
998 ciKlass* thread_klass = env()->Thread_klass();
999 const Type* thread_type
1000 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1001
1002 Node* thread = _gvn.transform(new ThreadLocalNode());
1003 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
1004 tls_output = thread;
1005
1006 Node* thread_obj_handle
1007 = (is_immutable
1008 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1009 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1010 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1011 thread_obj_handle = _gvn.transform(thread_obj_handle);
1012
1013 DecoratorSet decorators = IN_NATIVE;
1014 if (is_immutable) {
1015 decorators |= C2_IMMUTABLE_MEMORY;
1016 }
1017 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1018 }
1019
1020 //--------------------------generate_current_thread--------------------
1021 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1022 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1023 /*is_immutable*/false);
1024 }
1025
1026 //--------------------------generate_virtual_thread--------------------
1027 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1028 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1029 !C->method()->changes_current_thread());
1030 }
1031
1032 //------------------------------make_string_method_node------------------------
1033 // Helper method for String intrinsic functions. This version is called with
1034 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1035 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1036 // containing the lengths of str1 and str2.
1037 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1038 Node* result = nullptr;
1039 switch (opcode) {
1040 case Op_StrIndexOf:
1041 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1042 str1_start, cnt1, str2_start, cnt2, ae);
1043 break;
1044 case Op_StrComp:
1045 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1046 str1_start, cnt1, str2_start, cnt2, ae);
1047 break;
1048 case Op_StrEquals:
1049 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1050 // Use the constant length if there is one because optimized match rule may exist.
1051 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1052 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1053 break;
1054 default:
1055 ShouldNotReachHere();
1056 return nullptr;
1057 }
1058
1059 // All these intrinsics have checks.
1060 C->set_has_split_ifs(true); // Has chance for split-if optimization
1061 clear_upper_avx();
1062
1063 return _gvn.transform(result);
1064 }
1065
1066 //------------------------------inline_string_compareTo------------------------
1067 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1068 Node* arg1 = argument(0);
1069 Node* arg2 = argument(1);
1070
1071 arg1 = must_be_not_null(arg1, true);
1072 arg2 = must_be_not_null(arg2, true);
1073
1074 // Get start addr and length of first argument
1075 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1076 Node* arg1_cnt = load_array_length(arg1);
1077
1078 // Get start addr and length of second argument
1079 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1080 Node* arg2_cnt = load_array_length(arg2);
1081
1082 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1083 set_result(result);
1084 return true;
1085 }
1086
1087 //------------------------------inline_string_equals------------------------
1088 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1089 Node* arg1 = argument(0);
1090 Node* arg2 = argument(1);
1091
1092 // paths (plus control) merge
1093 RegionNode* region = new RegionNode(3);
1094 Node* phi = new PhiNode(region, TypeInt::BOOL);
1095
1096 if (!stopped()) {
1097
1098 arg1 = must_be_not_null(arg1, true);
1099 arg2 = must_be_not_null(arg2, true);
1100
1101 // Get start addr and length of first argument
1102 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1103 Node* arg1_cnt = load_array_length(arg1);
1104
1105 // Get start addr and length of second argument
1106 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1107 Node* arg2_cnt = load_array_length(arg2);
1108
1109 // Check for arg1_cnt != arg2_cnt
1110 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1111 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1112 Node* if_ne = generate_slow_guard(bol, nullptr);
1113 if (if_ne != nullptr) {
1114 phi->init_req(2, intcon(0));
1115 region->init_req(2, if_ne);
1116 }
1117
1118 // Check for count == 0 is done by assembler code for StrEquals.
1119
1120 if (!stopped()) {
1121 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1122 phi->init_req(1, equals);
1123 region->init_req(1, control());
1124 }
1125 }
1126
1127 // post merge
1128 set_control(_gvn.transform(region));
1129 record_for_igvn(region);
1130
1131 set_result(_gvn.transform(phi));
1132 return true;
1133 }
1134
1135 //------------------------------inline_array_equals----------------------------
1136 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1137 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1138 Node* arg1 = argument(0);
1139 Node* arg2 = argument(1);
1140
1141 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1142 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1143 clear_upper_avx();
1144
1145 return true;
1146 }
1147
1148
1149 //------------------------------inline_countPositives------------------------------
1150 bool LibraryCallKit::inline_countPositives() {
1151 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1152 return false;
1153 }
1154
1155 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1156 // no receiver since it is static method
1157 Node* ba = argument(0);
1158 Node* offset = argument(1);
1159 Node* len = argument(2);
1160
1161 ba = must_be_not_null(ba, true);
1162
1163 // Range checks
1164 generate_string_range_check(ba, offset, len, false);
1165 if (stopped()) {
1166 return true;
1167 }
1168 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1169 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1170 set_result(_gvn.transform(result));
1171 clear_upper_avx();
1172 return true;
1173 }
1174
1175 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1176 Node* index = argument(0);
1177 Node* length = bt == T_INT ? argument(1) : argument(2);
1178 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1179 return false;
1180 }
1181
1182 // check that length is positive
1183 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1184 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1185
1186 {
1187 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1188 uncommon_trap(Deoptimization::Reason_intrinsic,
1189 Deoptimization::Action_make_not_entrant);
1190 }
1191
1192 if (stopped()) {
1193 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1194 return true;
1195 }
1196
1197 // length is now known positive, add a cast node to make this explicit
1198 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1199 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1200 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1201 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1202 casted_length = _gvn.transform(casted_length);
1203 replace_in_map(length, casted_length);
1204 length = casted_length;
1205
1206 // Use an unsigned comparison for the range check itself
1207 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1208 BoolTest::mask btest = BoolTest::lt;
1209 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1210 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1211 _gvn.set_type(rc, rc->Value(&_gvn));
1212 if (!rc_bool->is_Con()) {
1213 record_for_igvn(rc);
1214 }
1215 set_control(_gvn.transform(new IfTrueNode(rc)));
1216 {
1217 PreserveJVMState pjvms(this);
1218 set_control(_gvn.transform(new IfFalseNode(rc)));
1219 uncommon_trap(Deoptimization::Reason_range_check,
1220 Deoptimization::Action_make_not_entrant);
1221 }
1222
1223 if (stopped()) {
1224 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1225 return true;
1226 }
1227
1228 // index is now known to be >= 0 and < length, cast it
1229 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1230 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1231 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1232 result = _gvn.transform(result);
1233 set_result(result);
1234 replace_in_map(index, result);
1235 return true;
1236 }
1237
1238 //------------------------------inline_string_indexOf------------------------
1239 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1240 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1241 return false;
1242 }
1243 Node* src = argument(0);
1244 Node* tgt = argument(1);
1245
1246 // Make the merge point
1247 RegionNode* result_rgn = new RegionNode(4);
1248 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1249
1250 src = must_be_not_null(src, true);
1251 tgt = must_be_not_null(tgt, true);
1252
1253 // Get start addr and length of source string
1254 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1255 Node* src_count = load_array_length(src);
1256
1257 // Get start addr and length of substring
1258 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1259 Node* tgt_count = load_array_length(tgt);
1260
1261 Node* result = nullptr;
1262 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1263
1264 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1265 // Divide src size by 2 if String is UTF16 encoded
1266 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1267 }
1268 if (ae == StrIntrinsicNode::UU) {
1269 // Divide substring size by 2 if String is UTF16 encoded
1270 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1271 }
1272
1273 if (call_opt_stub) {
1274 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1275 StubRoutines::_string_indexof_array[ae],
1276 "stringIndexOf", TypePtr::BOTTOM, src_start,
1277 src_count, tgt_start, tgt_count);
1278 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1279 } else {
1280 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1281 result_rgn, result_phi, ae);
1282 }
1283 if (result != nullptr) {
1284 result_phi->init_req(3, result);
1285 result_rgn->init_req(3, control());
1286 }
1287 set_control(_gvn.transform(result_rgn));
1288 record_for_igvn(result_rgn);
1289 set_result(_gvn.transform(result_phi));
1290
1291 return true;
1292 }
1293
1294 //-----------------------------inline_string_indexOfI-----------------------
1295 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1296 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1297 return false;
1298 }
1299 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1300 return false;
1301 }
1302
1303 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1304 Node* src = argument(0); // byte[]
1305 Node* src_count = argument(1); // char count
1306 Node* tgt = argument(2); // byte[]
1307 Node* tgt_count = argument(3); // char count
1308 Node* from_index = argument(4); // char index
1309
1310 src = must_be_not_null(src, true);
1311 tgt = must_be_not_null(tgt, true);
1312
1313 // Multiply byte array index by 2 if String is UTF16 encoded
1314 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1315 src_count = _gvn.transform(new SubINode(src_count, from_index));
1316 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1317 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1318
1319 // Range checks
1320 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1321 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1322 if (stopped()) {
1323 return true;
1324 }
1325
1326 RegionNode* region = new RegionNode(5);
1327 Node* phi = new PhiNode(region, TypeInt::INT);
1328 Node* result = nullptr;
1329
1330 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1331
1332 if (call_opt_stub) {
1333 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1334 StubRoutines::_string_indexof_array[ae],
1335 "stringIndexOf", TypePtr::BOTTOM, src_start,
1336 src_count, tgt_start, tgt_count);
1337 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1338 } else {
1339 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1340 region, phi, ae);
1341 }
1342 if (result != nullptr) {
1343 // The result is index relative to from_index if substring was found, -1 otherwise.
1344 // Generate code which will fold into cmove.
1345 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1346 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1347
1348 Node* if_lt = generate_slow_guard(bol, nullptr);
1349 if (if_lt != nullptr) {
1350 // result == -1
1351 phi->init_req(3, result);
1352 region->init_req(3, if_lt);
1353 }
1354 if (!stopped()) {
1355 result = _gvn.transform(new AddINode(result, from_index));
1356 phi->init_req(4, result);
1357 region->init_req(4, control());
1358 }
1359 }
1360
1361 set_control(_gvn.transform(region));
1362 record_for_igvn(region);
1363 set_result(_gvn.transform(phi));
1364 clear_upper_avx();
1365
1366 return true;
1367 }
1368
1369 // Create StrIndexOfNode with fast path checks
1370 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1371 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1372 // Check for substr count > string count
1373 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1374 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1375 Node* if_gt = generate_slow_guard(bol, nullptr);
1376 if (if_gt != nullptr) {
1377 phi->init_req(1, intcon(-1));
1378 region->init_req(1, if_gt);
1379 }
1380 if (!stopped()) {
1381 // Check for substr count == 0
1382 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1383 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1384 Node* if_zero = generate_slow_guard(bol, nullptr);
1385 if (if_zero != nullptr) {
1386 phi->init_req(2, intcon(0));
1387 region->init_req(2, if_zero);
1388 }
1389 }
1390 if (!stopped()) {
1391 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1392 }
1393 return nullptr;
1394 }
1395
1396 //-----------------------------inline_string_indexOfChar-----------------------
1397 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1398 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1399 return false;
1400 }
1401 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1402 return false;
1403 }
1404 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1405 Node* src = argument(0); // byte[]
1406 Node* int_ch = argument(1);
1407 Node* from_index = argument(2);
1408 Node* max = argument(3);
1409
1410 src = must_be_not_null(src, true);
1411
1412 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1413 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1414 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1415
1416 // Range checks
1417 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1418
1419 // Check for int_ch >= 0
1420 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1421 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1422 {
1423 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1424 uncommon_trap(Deoptimization::Reason_intrinsic,
1425 Deoptimization::Action_maybe_recompile);
1426 }
1427 if (stopped()) {
1428 return true;
1429 }
1430
1431 RegionNode* region = new RegionNode(3);
1432 Node* phi = new PhiNode(region, TypeInt::INT);
1433
1434 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1435 C->set_has_split_ifs(true); // Has chance for split-if optimization
1436 _gvn.transform(result);
1437
1438 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1439 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1440
1441 Node* if_lt = generate_slow_guard(bol, nullptr);
1442 if (if_lt != nullptr) {
1443 // result == -1
1444 phi->init_req(2, result);
1445 region->init_req(2, if_lt);
1446 }
1447 if (!stopped()) {
1448 result = _gvn.transform(new AddINode(result, from_index));
1449 phi->init_req(1, result);
1450 region->init_req(1, control());
1451 }
1452 set_control(_gvn.transform(region));
1453 record_for_igvn(region);
1454 set_result(_gvn.transform(phi));
1455 clear_upper_avx();
1456
1457 return true;
1458 }
1459 //---------------------------inline_string_copy---------------------
1460 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1461 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1462 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1463 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1464 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1465 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1466 bool LibraryCallKit::inline_string_copy(bool compress) {
1467 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1468 return false;
1469 }
1470 int nargs = 5; // 2 oops, 3 ints
1471 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1472
1473 Node* src = argument(0);
1474 Node* src_offset = argument(1);
1475 Node* dst = argument(2);
1476 Node* dst_offset = argument(3);
1477 Node* length = argument(4);
1478
1479 // Check for allocation before we add nodes that would confuse
1480 // tightly_coupled_allocation()
1481 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1482
1483 // Figure out the size and type of the elements we will be copying.
1484 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1485 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1486 if (src_type == nullptr || dst_type == nullptr) {
1487 return false;
1488 }
1489 BasicType src_elem = src_type->elem()->array_element_basic_type();
1490 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1491 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1492 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1493 "Unsupported array types for inline_string_copy");
1494
1495 src = must_be_not_null(src, true);
1496 dst = must_be_not_null(dst, true);
1497
1498 // Convert char[] offsets to byte[] offsets
1499 bool convert_src = (compress && src_elem == T_BYTE);
1500 bool convert_dst = (!compress && dst_elem == T_BYTE);
1501 if (convert_src) {
1502 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1503 } else if (convert_dst) {
1504 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1505 }
1506
1507 // Range checks
1508 generate_string_range_check(src, src_offset, length, convert_src);
1509 generate_string_range_check(dst, dst_offset, length, convert_dst);
1510 if (stopped()) {
1511 return true;
1512 }
1513
1514 Node* src_start = array_element_address(src, src_offset, src_elem);
1515 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1516 // 'src_start' points to src array + scaled offset
1517 // 'dst_start' points to dst array + scaled offset
1518 Node* count = nullptr;
1519 if (compress) {
1520 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1521 } else {
1522 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1523 }
1524
1525 if (alloc != nullptr) {
1526 if (alloc->maybe_set_complete(&_gvn)) {
1527 // "You break it, you buy it."
1528 InitializeNode* init = alloc->initialization();
1529 assert(init->is_complete(), "we just did this");
1530 init->set_complete_with_arraycopy();
1531 assert(dst->is_CheckCastPP(), "sanity");
1532 assert(dst->in(0)->in(0) == init, "dest pinned");
1533 }
1534 // Do not let stores that initialize this object be reordered with
1535 // a subsequent store that would make this object accessible by
1536 // other threads.
1537 // Record what AllocateNode this StoreStore protects so that
1538 // escape analysis can go from the MemBarStoreStoreNode to the
1539 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1540 // based on the escape status of the AllocateNode.
1541 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1542 }
1543 if (compress) {
1544 set_result(_gvn.transform(count));
1545 }
1546 clear_upper_avx();
1547
1548 return true;
1549 }
1550
1551 #ifdef _LP64
1552 #define XTOP ,top() /*additional argument*/
1553 #else //_LP64
1554 #define XTOP /*no additional argument*/
1555 #endif //_LP64
1556
1557 //------------------------inline_string_toBytesU--------------------------
1558 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1559 bool LibraryCallKit::inline_string_toBytesU() {
1560 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1561 return false;
1562 }
1563 // Get the arguments.
1564 Node* value = argument(0);
1565 Node* offset = argument(1);
1566 Node* length = argument(2);
1567
1568 Node* newcopy = nullptr;
1569
1570 // Set the original stack and the reexecute bit for the interpreter to reexecute
1571 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1572 { PreserveReexecuteState preexecs(this);
1573 jvms()->set_should_reexecute(true);
1574
1575 // Check if a null path was taken unconditionally.
1576 value = null_check(value);
1577
1578 RegionNode* bailout = new RegionNode(1);
1579 record_for_igvn(bailout);
1580
1581 // Range checks
1582 generate_negative_guard(offset, bailout);
1583 generate_negative_guard(length, bailout);
1584 generate_limit_guard(offset, length, load_array_length(value), bailout);
1585 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1586 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1587
1588 if (bailout->req() > 1) {
1589 PreserveJVMState pjvms(this);
1590 set_control(_gvn.transform(bailout));
1591 uncommon_trap(Deoptimization::Reason_intrinsic,
1592 Deoptimization::Action_maybe_recompile);
1593 }
1594 if (stopped()) {
1595 return true;
1596 }
1597
1598 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1599 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1600 newcopy = new_array(klass_node, size, 0); // no arguments to push
1601 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1602 guarantee(alloc != nullptr, "created above");
1603
1604 // Calculate starting addresses.
1605 Node* src_start = array_element_address(value, offset, T_CHAR);
1606 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1607
1608 // Check if dst array address is aligned to HeapWordSize
1609 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1610 // If true, then check if src array address is aligned to HeapWordSize
1611 if (aligned) {
1612 const TypeInt* toffset = gvn().type(offset)->is_int();
1613 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1614 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1615 }
1616
1617 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1618 const char* copyfunc_name = "arraycopy";
1619 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1620 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1621 OptoRuntime::fast_arraycopy_Type(),
1622 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1623 src_start, dst_start, ConvI2X(length) XTOP);
1624 // Do not let reads from the cloned object float above the arraycopy.
1625 if (alloc->maybe_set_complete(&_gvn)) {
1626 // "You break it, you buy it."
1627 InitializeNode* init = alloc->initialization();
1628 assert(init->is_complete(), "we just did this");
1629 init->set_complete_with_arraycopy();
1630 assert(newcopy->is_CheckCastPP(), "sanity");
1631 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1632 }
1633 // Do not let stores that initialize this object be reordered with
1634 // a subsequent store that would make this object accessible by
1635 // other threads.
1636 // Record what AllocateNode this StoreStore protects so that
1637 // escape analysis can go from the MemBarStoreStoreNode to the
1638 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1639 // based on the escape status of the AllocateNode.
1640 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1641 } // original reexecute is set back here
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 if (!stopped()) {
1645 set_result(newcopy);
1646 }
1647 clear_upper_avx();
1648
1649 return true;
1650 }
1651
1652 //------------------------inline_string_getCharsU--------------------------
1653 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1654 bool LibraryCallKit::inline_string_getCharsU() {
1655 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1656 return false;
1657 }
1658
1659 // Get the arguments.
1660 Node* src = argument(0);
1661 Node* src_begin = argument(1);
1662 Node* src_end = argument(2); // exclusive offset (i < src_end)
1663 Node* dst = argument(3);
1664 Node* dst_begin = argument(4);
1665
1666 // Check for allocation before we add nodes that would confuse
1667 // tightly_coupled_allocation()
1668 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1669
1670 // Check if a null path was taken unconditionally.
1671 src = null_check(src);
1672 dst = null_check(dst);
1673 if (stopped()) {
1674 return true;
1675 }
1676
1677 // Get length and convert char[] offset to byte[] offset
1678 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1679 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1680
1681 // Range checks
1682 generate_string_range_check(src, src_begin, length, true);
1683 generate_string_range_check(dst, dst_begin, length, false);
1684 if (stopped()) {
1685 return true;
1686 }
1687
1688 if (!stopped()) {
1689 // Calculate starting addresses.
1690 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1691 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1692
1693 // Check if array addresses are aligned to HeapWordSize
1694 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1695 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1696 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1697 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1698
1699 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1700 const char* copyfunc_name = "arraycopy";
1701 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1702 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1703 OptoRuntime::fast_arraycopy_Type(),
1704 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1705 src_start, dst_start, ConvI2X(length) XTOP);
1706 // Do not let reads from the cloned object float above the arraycopy.
1707 if (alloc != nullptr) {
1708 if (alloc->maybe_set_complete(&_gvn)) {
1709 // "You break it, you buy it."
1710 InitializeNode* init = alloc->initialization();
1711 assert(init->is_complete(), "we just did this");
1712 init->set_complete_with_arraycopy();
1713 assert(dst->is_CheckCastPP(), "sanity");
1714 assert(dst->in(0)->in(0) == init, "dest pinned");
1715 }
1716 // Do not let stores that initialize this object be reordered with
1717 // a subsequent store that would make this object accessible by
1718 // other threads.
1719 // Record what AllocateNode this StoreStore protects so that
1720 // escape analysis can go from the MemBarStoreStoreNode to the
1721 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1722 // based on the escape status of the AllocateNode.
1723 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1724 } else {
1725 insert_mem_bar(Op_MemBarCPUOrder);
1726 }
1727 }
1728
1729 C->set_has_split_ifs(true); // Has chance for split-if optimization
1730 return true;
1731 }
1732
1733 //----------------------inline_string_char_access----------------------------
1734 // Store/Load char to/from byte[] array.
1735 // static void StringUTF16.putChar(byte[] val, int index, int c)
1736 // static char StringUTF16.getChar(byte[] val, int index)
1737 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1738 Node* value = argument(0);
1739 Node* index = argument(1);
1740 Node* ch = is_store ? argument(2) : nullptr;
1741
1742 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1743 // correctly requires matched array shapes.
1744 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1745 "sanity: byte[] and char[] bases agree");
1746 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1747 "sanity: byte[] and char[] scales agree");
1748
1749 // Bail when getChar over constants is requested: constant folding would
1750 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1751 // Java method would constant fold nicely instead.
1752 if (!is_store && value->is_Con() && index->is_Con()) {
1753 return false;
1754 }
1755
1756 // Save state and restore on bailout
1757 SavedState old_state(this);
1758
1759 value = must_be_not_null(value, true);
1760
1761 Node* adr = array_element_address(value, index, T_CHAR);
1762 if (adr->is_top()) {
1763 return false;
1764 }
1765 old_state.discard();
1766 if (is_store) {
1767 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1768 } else {
1769 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);
1770 set_result(ch);
1771 }
1772 return true;
1773 }
1774
1775
1776 //------------------------------inline_math-----------------------------------
1777 // public static double Math.abs(double)
1778 // public static double Math.sqrt(double)
1779 // public static double Math.log(double)
1780 // public static double Math.log10(double)
1781 // public static double Math.round(double)
1782 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1783 Node* arg = argument(0);
1784 Node* n = nullptr;
1785 switch (id) {
1786 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1787 case vmIntrinsics::_dsqrt:
1788 case vmIntrinsics::_dsqrt_strict:
1789 n = new SqrtDNode(C, control(), arg); break;
1790 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1791 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1792 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1793 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1794 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1795 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1796 default: fatal_unexpected_iid(id); break;
1797 }
1798 set_result(_gvn.transform(n));
1799 return true;
1800 }
1801
1802 //------------------------------inline_math-----------------------------------
1803 // public static float Math.abs(float)
1804 // public static int Math.abs(int)
1805 // public static long Math.abs(long)
1806 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1807 Node* arg = argument(0);
1808 Node* n = nullptr;
1809 switch (id) {
1810 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1811 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1812 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1813 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1814 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1815 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1816 default: fatal_unexpected_iid(id); break;
1817 }
1818 set_result(_gvn.transform(n));
1819 return true;
1820 }
1821
1822 //------------------------------runtime_math-----------------------------
1823 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1824 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1825 "must be (DD)D or (D)D type");
1826
1827 // Inputs
1828 Node* a = argument(0);
1829 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1830
1831 const TypePtr* no_memory_effects = nullptr;
1832 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1833 no_memory_effects,
1834 a, top(), b, b ? top() : nullptr);
1835 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1836 #ifdef ASSERT
1837 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1838 assert(value_top == top(), "second value must be top");
1839 #endif
1840
1841 set_result(value);
1842 return true;
1843 }
1844
1845 //------------------------------inline_math_pow-----------------------------
1846 bool LibraryCallKit::inline_math_pow() {
1847 Node* exp = argument(2);
1848 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1849 if (d != nullptr) {
1850 if (d->getd() == 2.0) {
1851 // Special case: pow(x, 2.0) => x * x
1852 Node* base = argument(0);
1853 set_result(_gvn.transform(new MulDNode(base, base)));
1854 return true;
1855 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1856 // Special case: pow(x, 0.5) => sqrt(x)
1857 Node* base = argument(0);
1858 Node* zero = _gvn.zerocon(T_DOUBLE);
1859
1860 RegionNode* region = new RegionNode(3);
1861 Node* phi = new PhiNode(region, Type::DOUBLE);
1862
1863 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1864 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1865 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1866 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1867 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1868
1869 Node* if_pow = generate_slow_guard(test, nullptr);
1870 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1871 phi->init_req(1, value_sqrt);
1872 region->init_req(1, control());
1873
1874 if (if_pow != nullptr) {
1875 set_control(if_pow);
1876 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1877 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1878 const TypePtr* no_memory_effects = nullptr;
1879 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1880 no_memory_effects, base, top(), exp, top());
1881 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1882 #ifdef ASSERT
1883 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1884 assert(value_top == top(), "second value must be top");
1885 #endif
1886 phi->init_req(2, value_pow);
1887 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1888 }
1889
1890 C->set_has_split_ifs(true); // Has chance for split-if optimization
1891 set_control(_gvn.transform(region));
1892 record_for_igvn(region);
1893 set_result(_gvn.transform(phi));
1894
1895 return true;
1896 }
1897 }
1898
1899 return StubRoutines::dpow() != nullptr ?
1900 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1901 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1902 }
1903
1904 //------------------------------inline_math_native-----------------------------
1905 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1906 switch (id) {
1907 case vmIntrinsics::_dsin:
1908 return StubRoutines::dsin() != nullptr ?
1909 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1910 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1911 case vmIntrinsics::_dcos:
1912 return StubRoutines::dcos() != nullptr ?
1913 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1914 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1915 case vmIntrinsics::_dtan:
1916 return StubRoutines::dtan() != nullptr ?
1917 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1918 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1919 case vmIntrinsics::_dsinh:
1920 return StubRoutines::dsinh() != nullptr ?
1921 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1922 case vmIntrinsics::_dtanh:
1923 return StubRoutines::dtanh() != nullptr ?
1924 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1925 case vmIntrinsics::_dcbrt:
1926 return StubRoutines::dcbrt() != nullptr ?
1927 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1928 case vmIntrinsics::_dexp:
1929 return StubRoutines::dexp() != nullptr ?
1930 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1931 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1932 case vmIntrinsics::_dlog:
1933 return StubRoutines::dlog() != nullptr ?
1934 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1935 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1936 case vmIntrinsics::_dlog10:
1937 return StubRoutines::dlog10() != nullptr ?
1938 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1939 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1940
1941 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1942 case vmIntrinsics::_ceil:
1943 case vmIntrinsics::_floor:
1944 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1945
1946 case vmIntrinsics::_dsqrt:
1947 case vmIntrinsics::_dsqrt_strict:
1948 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1949 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1950 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1951 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1952 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1953
1954 case vmIntrinsics::_dpow: return inline_math_pow();
1955 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1956 case vmIntrinsics::_fcopySign: return inline_math(id);
1957 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1958 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1959 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1960
1961 // These intrinsics are not yet correctly implemented
1962 case vmIntrinsics::_datan2:
1963 return false;
1964
1965 default:
1966 fatal_unexpected_iid(id);
1967 return false;
1968 }
1969 }
1970
1971 //----------------------------inline_notify-----------------------------------*
1972 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1973 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1974 address func;
1975 if (id == vmIntrinsics::_notify) {
1976 func = OptoRuntime::monitor_notify_Java();
1977 } else {
1978 func = OptoRuntime::monitor_notifyAll_Java();
1979 }
1980 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1981 make_slow_call_ex(call, env()->Throwable_klass(), false);
1982 return true;
1983 }
1984
1985
1986 //----------------------------inline_min_max-----------------------------------
1987 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1988 Node* a = nullptr;
1989 Node* b = nullptr;
1990 Node* n = nullptr;
1991 switch (id) {
1992 case vmIntrinsics::_min:
1993 case vmIntrinsics::_max:
1994 case vmIntrinsics::_minF:
1995 case vmIntrinsics::_maxF:
1996 case vmIntrinsics::_minF_strict:
1997 case vmIntrinsics::_maxF_strict:
1998 case vmIntrinsics::_min_strict:
1999 case vmIntrinsics::_max_strict:
2000 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
2001 a = argument(0);
2002 b = argument(1);
2003 break;
2004 case vmIntrinsics::_minD:
2005 case vmIntrinsics::_maxD:
2006 case vmIntrinsics::_minD_strict:
2007 case vmIntrinsics::_maxD_strict:
2008 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2009 a = argument(0);
2010 b = argument(2);
2011 break;
2012 case vmIntrinsics::_minL:
2013 case vmIntrinsics::_maxL:
2014 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2015 a = argument(0);
2016 b = argument(2);
2017 break;
2018 default:
2019 fatal_unexpected_iid(id);
2020 break;
2021 }
2022
2023 switch (id) {
2024 case vmIntrinsics::_min:
2025 case vmIntrinsics::_min_strict:
2026 n = new MinINode(a, b);
2027 break;
2028 case vmIntrinsics::_max:
2029 case vmIntrinsics::_max_strict:
2030 n = new MaxINode(a, b);
2031 break;
2032 case vmIntrinsics::_minF:
2033 case vmIntrinsics::_minF_strict:
2034 n = new MinFNode(a, b);
2035 break;
2036 case vmIntrinsics::_maxF:
2037 case vmIntrinsics::_maxF_strict:
2038 n = new MaxFNode(a, b);
2039 break;
2040 case vmIntrinsics::_minD:
2041 case vmIntrinsics::_minD_strict:
2042 n = new MinDNode(a, b);
2043 break;
2044 case vmIntrinsics::_maxD:
2045 case vmIntrinsics::_maxD_strict:
2046 n = new MaxDNode(a, b);
2047 break;
2048 case vmIntrinsics::_minL:
2049 n = new MinLNode(_gvn.C, a, b);
2050 break;
2051 case vmIntrinsics::_maxL:
2052 n = new MaxLNode(_gvn.C, a, b);
2053 break;
2054 default:
2055 fatal_unexpected_iid(id);
2056 break;
2057 }
2058
2059 set_result(_gvn.transform(n));
2060 return true;
2061 }
2062
2063 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2064 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2065 env()->ArithmeticException_instance())) {
2066 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2067 // so let's bail out intrinsic rather than risking deopting again.
2068 return false;
2069 }
2070
2071 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2072 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2073 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2074 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2075
2076 {
2077 PreserveJVMState pjvms(this);
2078 PreserveReexecuteState preexecs(this);
2079 jvms()->set_should_reexecute(true);
2080
2081 set_control(slow_path);
2082 set_i_o(i_o());
2083
2084 builtin_throw(Deoptimization::Reason_intrinsic,
2085 env()->ArithmeticException_instance(),
2086 /*allow_too_many_traps*/ false);
2087 }
2088
2089 set_control(fast_path);
2090 set_result(math);
2091 return true;
2092 }
2093
2094 template <typename OverflowOp>
2095 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2096 typedef typename OverflowOp::MathOp MathOp;
2097
2098 MathOp* mathOp = new MathOp(arg1, arg2);
2099 Node* operation = _gvn.transform( mathOp );
2100 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2101 return inline_math_mathExact(operation, ofcheck);
2102 }
2103
2104 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2105 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2106 }
2107
2108 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2109 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2110 }
2111
2112 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2113 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2114 }
2115
2116 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2117 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2118 }
2119
2120 bool LibraryCallKit::inline_math_negateExactI() {
2121 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2122 }
2123
2124 bool LibraryCallKit::inline_math_negateExactL() {
2125 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2126 }
2127
2128 bool LibraryCallKit::inline_math_multiplyExactI() {
2129 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2130 }
2131
2132 bool LibraryCallKit::inline_math_multiplyExactL() {
2133 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2134 }
2135
2136 bool LibraryCallKit::inline_math_multiplyHigh() {
2137 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2138 return true;
2139 }
2140
2141 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2142 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2143 return true;
2144 }
2145
2146 inline int
2147 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2148 const TypePtr* base_type = TypePtr::NULL_PTR;
2149 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2150 if (base_type == nullptr) {
2151 // Unknown type.
2152 return Type::AnyPtr;
2153 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2154 // Since this is a null+long form, we have to switch to a rawptr.
2155 base = _gvn.transform(new CastX2PNode(offset));
2156 offset = MakeConX(0);
2157 return Type::RawPtr;
2158 } else if (base_type->base() == Type::RawPtr) {
2159 return Type::RawPtr;
2160 } else if (base_type->isa_oopptr()) {
2161 // Base is never null => always a heap address.
2162 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2163 return Type::OopPtr;
2164 }
2165 // Offset is small => always a heap address.
2166 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2167 if (offset_type != nullptr &&
2168 base_type->offset() == 0 && // (should always be?)
2169 offset_type->_lo >= 0 &&
2170 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2171 return Type::OopPtr;
2172 } else if (type == T_OBJECT) {
2173 // off heap access to an oop doesn't make any sense. Has to be on
2174 // heap.
2175 return Type::OopPtr;
2176 }
2177 // Otherwise, it might either be oop+off or null+addr.
2178 return Type::AnyPtr;
2179 } else {
2180 // No information:
2181 return Type::AnyPtr;
2182 }
2183 }
2184
2185 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2186 Node* uncasted_base = base;
2187 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2188 if (kind == Type::RawPtr) {
2189 return basic_plus_adr(top(), uncasted_base, offset);
2190 } else if (kind == Type::AnyPtr) {
2191 assert(base == uncasted_base, "unexpected base change");
2192 if (can_cast) {
2193 if (!_gvn.type(base)->speculative_maybe_null() &&
2194 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2195 // According to profiling, this access is always on
2196 // heap. Casting the base to not null and thus avoiding membars
2197 // around the access should allow better optimizations
2198 Node* null_ctl = top();
2199 base = null_check_oop(base, &null_ctl, true, true, true);
2200 assert(null_ctl->is_top(), "no null control here");
2201 return basic_plus_adr(base, offset);
2202 } else if (_gvn.type(base)->speculative_always_null() &&
2203 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2204 // According to profiling, this access is always off
2205 // heap.
2206 base = null_assert(base);
2207 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2208 offset = MakeConX(0);
2209 return basic_plus_adr(top(), raw_base, offset);
2210 }
2211 }
2212 // We don't know if it's an on heap or off heap access. Fall back
2213 // to raw memory access.
2214 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2215 return basic_plus_adr(top(), raw, offset);
2216 } else {
2217 assert(base == uncasted_base, "unexpected base change");
2218 // We know it's an on heap access so base can't be null
2219 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2220 base = must_be_not_null(base, true);
2221 }
2222 return basic_plus_adr(base, offset);
2223 }
2224 }
2225
2226 //--------------------------inline_number_methods-----------------------------
2227 // inline int Integer.numberOfLeadingZeros(int)
2228 // inline int Long.numberOfLeadingZeros(long)
2229 //
2230 // inline int Integer.numberOfTrailingZeros(int)
2231 // inline int Long.numberOfTrailingZeros(long)
2232 //
2233 // inline int Integer.bitCount(int)
2234 // inline int Long.bitCount(long)
2235 //
2236 // inline char Character.reverseBytes(char)
2237 // inline short Short.reverseBytes(short)
2238 // inline int Integer.reverseBytes(int)
2239 // inline long Long.reverseBytes(long)
2240 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2241 Node* arg = argument(0);
2242 Node* n = nullptr;
2243 switch (id) {
2244 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2245 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2246 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2247 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2248 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2249 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2250 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2251 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2252 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2253 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2254 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2255 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2256 default: fatal_unexpected_iid(id); break;
2257 }
2258 set_result(_gvn.transform(n));
2259 return true;
2260 }
2261
2262 //--------------------------inline_bitshuffle_methods-----------------------------
2263 // inline int Integer.compress(int, int)
2264 // inline int Integer.expand(int, int)
2265 // inline long Long.compress(long, long)
2266 // inline long Long.expand(long, long)
2267 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2268 Node* n = nullptr;
2269 switch (id) {
2270 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2271 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2272 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2273 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2274 default: fatal_unexpected_iid(id); break;
2275 }
2276 set_result(_gvn.transform(n));
2277 return true;
2278 }
2279
2280 //--------------------------inline_number_methods-----------------------------
2281 // inline int Integer.compareUnsigned(int, int)
2282 // inline int Long.compareUnsigned(long, long)
2283 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2284 Node* arg1 = argument(0);
2285 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2286 Node* n = nullptr;
2287 switch (id) {
2288 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2289 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2290 default: fatal_unexpected_iid(id); break;
2291 }
2292 set_result(_gvn.transform(n));
2293 return true;
2294 }
2295
2296 //--------------------------inline_unsigned_divmod_methods-----------------------------
2297 // inline int Integer.divideUnsigned(int, int)
2298 // inline int Integer.remainderUnsigned(int, int)
2299 // inline long Long.divideUnsigned(long, long)
2300 // inline long Long.remainderUnsigned(long, long)
2301 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2302 Node* n = nullptr;
2303 switch (id) {
2304 case vmIntrinsics::_divideUnsigned_i: {
2305 zero_check_int(argument(1));
2306 // Compile-time detect of null-exception
2307 if (stopped()) {
2308 return true; // keep the graph constructed so far
2309 }
2310 n = new UDivINode(control(), argument(0), argument(1));
2311 break;
2312 }
2313 case vmIntrinsics::_divideUnsigned_l: {
2314 zero_check_long(argument(2));
2315 // Compile-time detect of null-exception
2316 if (stopped()) {
2317 return true; // keep the graph constructed so far
2318 }
2319 n = new UDivLNode(control(), argument(0), argument(2));
2320 break;
2321 }
2322 case vmIntrinsics::_remainderUnsigned_i: {
2323 zero_check_int(argument(1));
2324 // Compile-time detect of null-exception
2325 if (stopped()) {
2326 return true; // keep the graph constructed so far
2327 }
2328 n = new UModINode(control(), argument(0), argument(1));
2329 break;
2330 }
2331 case vmIntrinsics::_remainderUnsigned_l: {
2332 zero_check_long(argument(2));
2333 // Compile-time detect of null-exception
2334 if (stopped()) {
2335 return true; // keep the graph constructed so far
2336 }
2337 n = new UModLNode(control(), argument(0), argument(2));
2338 break;
2339 }
2340 default: fatal_unexpected_iid(id); break;
2341 }
2342 set_result(_gvn.transform(n));
2343 return true;
2344 }
2345
2346 //----------------------------inline_unsafe_access----------------------------
2347
2348 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2349 // Attempt to infer a sharper value type from the offset and base type.
2350 ciKlass* sharpened_klass = nullptr;
2351 bool null_free = false;
2352
2353 // See if it is an instance field, with an object type.
2354 if (alias_type->field() != nullptr) {
2355 if (alias_type->field()->type()->is_klass()) {
2356 sharpened_klass = alias_type->field()->type()->as_klass();
2357 null_free = alias_type->field()->is_null_free();
2358 }
2359 }
2360
2361 const TypeOopPtr* result = nullptr;
2362 // See if it is a narrow oop array.
2363 if (adr_type->isa_aryptr()) {
2364 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2365 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2366 null_free = adr_type->is_aryptr()->is_null_free();
2367 if (elem_type != nullptr && elem_type->is_loaded()) {
2368 // Sharpen the value type.
2369 result = elem_type;
2370 }
2371 }
2372 }
2373
2374 // The sharpened class might be unloaded if there is no class loader
2375 // contraint in place.
2376 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2377 // Sharpen the value type.
2378 result = TypeOopPtr::make_from_klass(sharpened_klass);
2379 if (null_free) {
2380 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2381 }
2382 }
2383 if (result != nullptr) {
2384 #ifndef PRODUCT
2385 if (C->print_intrinsics() || C->print_inlining()) {
2386 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2387 tty->print(" sharpened value: "); result->dump(); tty->cr();
2388 }
2389 #endif
2390 }
2391 return result;
2392 }
2393
2394 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2395 switch (kind) {
2396 case Relaxed:
2397 return MO_UNORDERED;
2398 case Opaque:
2399 return MO_RELAXED;
2400 case Acquire:
2401 return MO_ACQUIRE;
2402 case Release:
2403 return MO_RELEASE;
2404 case Volatile:
2405 return MO_SEQ_CST;
2406 default:
2407 ShouldNotReachHere();
2408 return 0;
2409 }
2410 }
2411
2412 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2413 _kit(kit),
2414 _sp(kit->sp()),
2415 _jvms(kit->jvms()),
2416 _map(kit->clone_map()),
2417 _discarded(false)
2418 {
2419 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2420 Node* out = kit->control()->fast_out(i);
2421 if (out->is_CFG()) {
2422 _ctrl_succ.push(out);
2423 }
2424 }
2425 }
2426
2427 LibraryCallKit::SavedState::~SavedState() {
2428 if (_discarded) {
2429 _kit->destruct_map_clone(_map);
2430 return;
2431 }
2432 _kit->jvms()->set_map(_map);
2433 _kit->jvms()->set_sp(_sp);
2434 _map->set_jvms(_kit->jvms());
2435 _kit->set_map(_map);
2436 _kit->set_sp(_sp);
2437 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2438 Node* out = _kit->control()->fast_out(i);
2439 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2440 _kit->_gvn.hash_delete(out);
2441 out->set_req(0, _kit->C->top());
2442 _kit->C->record_for_igvn(out);
2443 --i; --imax;
2444 _kit->_gvn.hash_find_insert(out);
2445 }
2446 }
2447 }
2448
2449 void LibraryCallKit::SavedState::discard() {
2450 _discarded = true;
2451 }
2452
2453 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2454 if (callee()->is_static()) return false; // caller must have the capability!
2455 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2456 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2457 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2458 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2459
2460 if (is_reference_type(type)) {
2461 decorators |= ON_UNKNOWN_OOP_REF;
2462 }
2463
2464 if (unaligned) {
2465 decorators |= C2_UNALIGNED;
2466 }
2467
2468 #ifndef PRODUCT
2469 {
2470 ResourceMark rm;
2471 // Check the signatures.
2472 ciSignature* sig = callee()->signature();
2473 #ifdef ASSERT
2474 if (!is_store) {
2475 // Object getReference(Object base, int/long offset), etc.
2476 BasicType rtype = sig->return_type()->basic_type();
2477 assert(rtype == type, "getter must return the expected value");
2478 assert(sig->count() == 2, "oop getter has 2 arguments");
2479 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2480 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2481 } else {
2482 // void putReference(Object base, int/long offset, Object x), etc.
2483 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2484 assert(sig->count() == 3, "oop putter has 3 arguments");
2485 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2486 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2487 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2488 assert(vtype == type, "putter must accept the expected value");
2489 }
2490 #endif // ASSERT
2491 }
2492 #endif //PRODUCT
2493
2494 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2495
2496 Node* receiver = argument(0); // type: oop
2497
2498 // Build address expression.
2499 Node* heap_base_oop = top();
2500
2501 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2502 Node* base = argument(1); // type: oop
2503 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2504 Node* offset = argument(2); // type: long
2505 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2506 // to be plain byte offsets, which are also the same as those accepted
2507 // by oopDesc::field_addr.
2508 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2509 "fieldOffset must be byte-scaled");
2510
2511 if (base->is_InlineType()) {
2512 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2513 InlineTypeNode* vt = base->as_InlineType();
2514 if (offset->is_Con()) {
2515 long off = find_long_con(offset, 0);
2516 ciInlineKlass* vk = vt->type()->inline_klass();
2517 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2518 return false;
2519 }
2520
2521 ciField* field = vk->get_non_flat_field_by_offset(off);
2522 if (field != nullptr) {
2523 BasicType bt = type2field[field->type()->basic_type()];
2524 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2525 bt = T_OBJECT;
2526 }
2527 if (bt == type && !field->is_flat()) {
2528 Node* value = vt->field_value_by_offset(off, false);
2529 if (value->is_InlineType()) {
2530 value = value->as_InlineType()->adjust_scalarization_depth(this);
2531 }
2532 set_result(value);
2533 return true;
2534 }
2535 }
2536 }
2537 {
2538 // Re-execute the unsafe access if allocation triggers deoptimization.
2539 PreserveReexecuteState preexecs(this);
2540 jvms()->set_should_reexecute(true);
2541 vt = vt->buffer(this);
2542 }
2543 base = vt->get_oop();
2544 }
2545
2546 // 32-bit machines ignore the high half!
2547 offset = ConvL2X(offset);
2548
2549 // Save state and restore on bailout
2550 SavedState old_state(this);
2551
2552 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2553 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2554
2555 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2556 if (type != T_OBJECT) {
2557 decorators |= IN_NATIVE; // off-heap primitive access
2558 } else {
2559 return false; // off-heap oop accesses are not supported
2560 }
2561 } else {
2562 heap_base_oop = base; // on-heap or mixed access
2563 }
2564
2565 // Can base be null? Otherwise, always on-heap access.
2566 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2567
2568 if (!can_access_non_heap) {
2569 decorators |= IN_HEAP;
2570 }
2571
2572 Node* val = is_store ? argument(4) : nullptr;
2573
2574 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2575 if (adr_type == TypePtr::NULL_PTR) {
2576 return false; // off-heap access with zero address
2577 }
2578
2579 // Try to categorize the address.
2580 Compile::AliasType* alias_type = C->alias_type(adr_type);
2581 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2582
2583 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2584 alias_type->adr_type() == TypeAryPtr::RANGE) {
2585 return false; // not supported
2586 }
2587
2588 bool mismatched = false;
2589 BasicType bt = T_ILLEGAL;
2590 ciField* field = nullptr;
2591 if (adr_type->isa_instptr()) {
2592 const TypeInstPtr* instptr = adr_type->is_instptr();
2593 ciInstanceKlass* k = instptr->instance_klass();
2594 int off = instptr->offset();
2595 if (instptr->const_oop() != nullptr &&
2596 k == ciEnv::current()->Class_klass() &&
2597 instptr->offset() >= (k->size_helper() * wordSize)) {
2598 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2599 field = k->get_field_by_offset(off, true);
2600 } else {
2601 field = k->get_non_flat_field_by_offset(off);
2602 }
2603 if (field != nullptr) {
2604 bt = type2field[field->type()->basic_type()];
2605 }
2606 if (bt != alias_type->basic_type()) {
2607 // Type mismatch. Is it an access to a nested flat field?
2608 field = k->get_field_by_offset(off, false);
2609 if (field != nullptr) {
2610 bt = type2field[field->type()->basic_type()];
2611 }
2612 }
2613 assert(bt == alias_type->basic_type(), "should match");
2614 } else {
2615 bt = alias_type->basic_type();
2616 }
2617
2618 if (bt != T_ILLEGAL) {
2619 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2620 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2621 // Alias type doesn't differentiate between byte[] and boolean[]).
2622 // Use address type to get the element type.
2623 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2624 }
2625 if (is_reference_type(bt, true)) {
2626 // accessing an array field with getReference is not a mismatch
2627 bt = T_OBJECT;
2628 }
2629 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2630 // Don't intrinsify mismatched object accesses
2631 return false;
2632 }
2633 mismatched = (bt != type);
2634 } else if (alias_type->adr_type()->isa_oopptr()) {
2635 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2636 }
2637
2638 old_state.discard();
2639 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2640
2641 if (mismatched) {
2642 decorators |= C2_MISMATCHED;
2643 }
2644
2645 // First guess at the value type.
2646 const Type *value_type = Type::get_const_basic_type(type);
2647
2648 // Figure out the memory ordering.
2649 decorators |= mo_decorator_for_access_kind(kind);
2650
2651 if (!is_store) {
2652 if (type == T_OBJECT) {
2653 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2654 if (tjp != nullptr) {
2655 value_type = tjp;
2656 }
2657 }
2658 }
2659
2660 receiver = null_check(receiver);
2661 if (stopped()) {
2662 return true;
2663 }
2664 // Heap pointers get a null-check from the interpreter,
2665 // as a courtesy. However, this is not guaranteed by Unsafe,
2666 // and it is not possible to fully distinguish unintended nulls
2667 // from intended ones in this API.
2668
2669 if (!is_store) {
2670 Node* p = nullptr;
2671 // Try to constant fold a load from a constant field
2672
2673 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2674 // final or stable field
2675 p = make_constant_from_field(field, heap_base_oop);
2676 }
2677
2678 if (p == nullptr) { // Could not constant fold the load
2679 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2680 const TypeOopPtr* ptr = value_type->make_oopptr();
2681 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2682 // Load a non-flattened inline type from memory
2683 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2684 }
2685 // Normalize the value returned by getBoolean in the following cases
2686 if (type == T_BOOLEAN &&
2687 (mismatched ||
2688 heap_base_oop == top() || // - heap_base_oop is null or
2689 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2690 // and the unsafe access is made to large offset
2691 // (i.e., larger than the maximum offset necessary for any
2692 // field access)
2693 ) {
2694 IdealKit ideal = IdealKit(this);
2695 #define __ ideal.
2696 IdealVariable normalized_result(ideal);
2697 __ declarations_done();
2698 __ set(normalized_result, p);
2699 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2700 __ set(normalized_result, ideal.ConI(1));
2701 ideal.end_if();
2702 final_sync(ideal);
2703 p = __ value(normalized_result);
2704 #undef __
2705 }
2706 }
2707 if (type == T_ADDRESS) {
2708 p = gvn().transform(new CastP2XNode(nullptr, p));
2709 p = ConvX2UL(p);
2710 }
2711 // The load node has the control of the preceding MemBarCPUOrder. All
2712 // following nodes will have the control of the MemBarCPUOrder inserted at
2713 // the end of this method. So, pushing the load onto the stack at a later
2714 // point is fine.
2715 set_result(p);
2716 } else {
2717 if (bt == T_ADDRESS) {
2718 // Repackage the long as a pointer.
2719 val = ConvL2X(val);
2720 val = gvn().transform(new CastX2PNode(val));
2721 }
2722 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2723 }
2724
2725 return true;
2726 }
2727
2728 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2729 #ifdef ASSERT
2730 {
2731 ResourceMark rm;
2732 // Check the signatures.
2733 ciSignature* sig = callee()->signature();
2734 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2735 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2736 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2737 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2738 if (is_store) {
2739 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2740 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2741 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2742 } else {
2743 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2744 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2745 }
2746 }
2747 #endif // ASSERT
2748
2749 assert(kind == Relaxed, "Only plain accesses for now");
2750 if (callee()->is_static()) {
2751 // caller must have the capability!
2752 return false;
2753 }
2754 C->set_has_unsafe_access(true);
2755
2756 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2757 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2758 // parameter valueType is not a constant
2759 return false;
2760 }
2761 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2762 if (!mirror_type->is_inlinetype()) {
2763 // Dead code
2764 return false;
2765 }
2766 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2767
2768 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2769 if (layout_type == nullptr || !layout_type->is_con()) {
2770 // parameter layoutKind is not a constant
2771 return false;
2772 }
2773 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2774 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2775 "invalid layoutKind %d", layout_type->get_con());
2776 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2777 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NULL_FREE_NON_ATOMIC_FLAT ||
2778 layout == LayoutKind::NULL_FREE_ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2779 "unexpected layoutKind %d", layout_type->get_con());
2780
2781 null_check(argument(0));
2782 if (stopped()) {
2783 return true;
2784 }
2785
2786 Node* base = must_be_not_null(argument(1), true);
2787 Node* offset = argument(2);
2788 const Type* base_type = _gvn.type(base);
2789
2790 Node* ptr;
2791 bool immutable_memory = false;
2792 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2793 if (base_type->isa_instptr()) {
2794 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2795 if (offset_type == nullptr || !offset_type->is_con()) {
2796 // Offset into a non-array should be a constant
2797 decorators |= C2_MISMATCHED;
2798 } else {
2799 int offset_con = checked_cast<int>(offset_type->get_con());
2800 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2801 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2802 if (field == nullptr) {
2803 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2804 decorators |= C2_MISMATCHED;
2805 } else {
2806 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2807 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2808 immutable_memory = field->is_strict() && field->is_final();
2809
2810 if (base->is_InlineType()) {
2811 assert(!is_store, "Cannot store into a non-larval value object");
2812 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2813 return true;
2814 }
2815 }
2816 }
2817
2818 if (base->is_InlineType()) {
2819 assert(!is_store, "Cannot store into a non-larval value object");
2820 base = base->as_InlineType()->buffer(this, true);
2821 }
2822 ptr = basic_plus_adr(base, ConvL2X(offset));
2823 } else if (base_type->isa_aryptr()) {
2824 decorators |= IS_ARRAY;
2825 if (layout == LayoutKind::REFERENCE) {
2826 if (!base_type->is_aryptr()->is_not_flat()) {
2827 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2828 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2829 replace_in_map(base, new_base);
2830 base = new_base;
2831 }
2832 ptr = basic_plus_adr(base, ConvL2X(offset));
2833 } else {
2834 if (UseArrayFlattening) {
2835 // Flat array must have an exact type
2836 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2837 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2838 Node* new_base = cast_to_flat_array_exact(base, value_klass, is_null_free, is_atomic);
2839 replace_in_map(base, new_base);
2840 base = new_base;
2841 ptr = basic_plus_adr(base, ConvL2X(offset));
2842 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2843 if (ptr_type->field_offset().get() != 0) {
2844 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2845 }
2846 } else {
2847 uncommon_trap(Deoptimization::Reason_intrinsic,
2848 Deoptimization::Action_none);
2849 return true;
2850 }
2851 }
2852 } else {
2853 decorators |= C2_MISMATCHED;
2854 ptr = basic_plus_adr(base, ConvL2X(offset));
2855 }
2856
2857 if (is_store) {
2858 Node* value = argument(6);
2859 const Type* value_type = _gvn.type(value);
2860 if (!value_type->is_inlinetypeptr()) {
2861 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2862 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::DependencyType::NonFloatingNarrowing));
2863 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2864 replace_in_map(value, new_value);
2865 value = new_value;
2866 }
2867
2868 assert(value_type->inline_klass() == value_klass, "value is of type %s while valueType is %s", value_type->inline_klass()->name()->as_utf8(), value_klass->name()->as_utf8());
2869 if (layout == LayoutKind::REFERENCE) {
2870 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2871 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2872 } else {
2873 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2874 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2875 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2876 }
2877
2878 return true;
2879 } else {
2880 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2881 InlineTypeNode* result;
2882 if (layout == LayoutKind::REFERENCE) {
2883 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2884 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2885 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2886 } else {
2887 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2888 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2889 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2890 }
2891
2892 set_result(result);
2893 return true;
2894 }
2895 }
2896
2897 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2898 Node* receiver = argument(0);
2899 Node* value = argument(1);
2900
2901 const Type* type = gvn().type(value);
2902 if (!type->is_inlinetypeptr()) {
2903 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2904 return false;
2905 }
2906
2907 null_check(receiver);
2908 if (stopped()) {
2909 return true;
2910 }
2911
2912 value = null_check(value);
2913 if (stopped()) {
2914 return true;
2915 }
2916
2917 ciInlineKlass* vk = type->inline_klass();
2918 Node* klass = makecon(TypeKlassPtr::make(vk));
2919 Node* obj = new_instance(klass);
2920 AllocateNode::Ideal_allocation(obj)->_larval = true;
2921
2922 assert(value->is_InlineType(), "must be an InlineTypeNode");
2923 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2924 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2925
2926 set_result(obj);
2927 return true;
2928 }
2929
2930 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2931 Node* receiver = argument(0);
2932 Node* buffer = argument(1);
2933
2934 const Type* type = gvn().type(buffer);
2935 if (!type->is_inlinetypeptr()) {
2936 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2937 return false;
2938 }
2939
2940 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2941 if (alloc == nullptr) {
2942 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2943 return false;
2944 }
2945
2946 null_check(receiver);
2947 if (stopped()) {
2948 return true;
2949 }
2950
2951 // Unset the larval bit in the object header
2952 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2953 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2954 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2955
2956 // We must ensure that the buffer is properly published
2957 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2958 assert(!type->maybe_null(), "result of an allocation should not be null");
2959 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2960 return true;
2961 }
2962
2963 //----------------------------inline_unsafe_load_store----------------------------
2964 // This method serves a couple of different customers (depending on LoadStoreKind):
2965 //
2966 // LS_cmp_swap:
2967 //
2968 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2969 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2970 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2971 //
2972 // LS_cmp_swap_weak:
2973 //
2974 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2975 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2976 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2977 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2978 //
2979 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2980 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2981 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2982 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2983 //
2984 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2985 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2986 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2987 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2988 //
2989 // LS_cmp_exchange:
2990 //
2991 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2992 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2993 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2994 //
2995 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2996 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2997 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2998 //
2999 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
3000 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
3001 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
3002 //
3003 // LS_get_add:
3004 //
3005 // int getAndAddInt( Object o, long offset, int delta)
3006 // long getAndAddLong(Object o, long offset, long delta)
3007 //
3008 // LS_get_set:
3009 //
3010 // int getAndSet(Object o, long offset, int newValue)
3011 // long getAndSet(Object o, long offset, long newValue)
3012 // Object getAndSet(Object o, long offset, Object newValue)
3013 //
3014 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
3015 // This basic scheme here is the same as inline_unsafe_access, but
3016 // differs in enough details that combining them would make the code
3017 // overly confusing. (This is a true fact! I originally combined
3018 // them, but even I was confused by it!) As much code/comments as
3019 // possible are retained from inline_unsafe_access though to make
3020 // the correspondences clearer. - dl
3021
3022 if (callee()->is_static()) return false; // caller must have the capability!
3023
3024 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3025 decorators |= mo_decorator_for_access_kind(access_kind);
3026
3027 #ifndef PRODUCT
3028 BasicType rtype;
3029 {
3030 ResourceMark rm;
3031 // Check the signatures.
3032 ciSignature* sig = callee()->signature();
3033 rtype = sig->return_type()->basic_type();
3034 switch(kind) {
3035 case LS_get_add:
3036 case LS_get_set: {
3037 // Check the signatures.
3038 #ifdef ASSERT
3039 assert(rtype == type, "get and set must return the expected type");
3040 assert(sig->count() == 3, "get and set has 3 arguments");
3041 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3042 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3043 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3044 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3045 #endif // ASSERT
3046 break;
3047 }
3048 case LS_cmp_swap:
3049 case LS_cmp_swap_weak: {
3050 // Check the signatures.
3051 #ifdef ASSERT
3052 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3053 assert(sig->count() == 4, "CAS has 4 arguments");
3054 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3055 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3056 #endif // ASSERT
3057 break;
3058 }
3059 case LS_cmp_exchange: {
3060 // Check the signatures.
3061 #ifdef ASSERT
3062 assert(rtype == type, "CAS must return the expected type");
3063 assert(sig->count() == 4, "CAS has 4 arguments");
3064 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3065 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3066 #endif // ASSERT
3067 break;
3068 }
3069 default:
3070 ShouldNotReachHere();
3071 }
3072 }
3073 #endif //PRODUCT
3074
3075 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3076
3077 // Get arguments:
3078 Node* receiver = nullptr;
3079 Node* base = nullptr;
3080 Node* offset = nullptr;
3081 Node* oldval = nullptr;
3082 Node* newval = nullptr;
3083 switch(kind) {
3084 case LS_cmp_swap:
3085 case LS_cmp_swap_weak:
3086 case LS_cmp_exchange: {
3087 const bool two_slot_type = type2size[type] == 2;
3088 receiver = argument(0); // type: oop
3089 base = argument(1); // type: oop
3090 offset = argument(2); // type: long
3091 oldval = argument(4); // type: oop, int, or long
3092 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3093 break;
3094 }
3095 case LS_get_add:
3096 case LS_get_set: {
3097 receiver = argument(0); // type: oop
3098 base = argument(1); // type: oop
3099 offset = argument(2); // type: long
3100 oldval = nullptr;
3101 newval = argument(4); // type: oop, int, or long
3102 break;
3103 }
3104 default:
3105 ShouldNotReachHere();
3106 }
3107
3108 // Build field offset expression.
3109 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3110 // to be plain byte offsets, which are also the same as those accepted
3111 // by oopDesc::field_addr.
3112 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3113 // 32-bit machines ignore the high half of long offsets
3114 offset = ConvL2X(offset);
3115 // Save state and restore on bailout
3116 SavedState old_state(this);
3117 Node* adr = make_unsafe_address(base, offset,type, false);
3118 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3119
3120 Compile::AliasType* alias_type = C->alias_type(adr_type);
3121 BasicType bt = alias_type->basic_type();
3122 if (bt != T_ILLEGAL &&
3123 (is_reference_type(bt) != (type == T_OBJECT))) {
3124 // Don't intrinsify mismatched object accesses.
3125 return false;
3126 }
3127
3128 old_state.discard();
3129
3130 // For CAS, unlike inline_unsafe_access, there seems no point in
3131 // trying to refine types. Just use the coarse types here.
3132 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3133 const Type *value_type = Type::get_const_basic_type(type);
3134
3135 switch (kind) {
3136 case LS_get_set:
3137 case LS_cmp_exchange: {
3138 if (type == T_OBJECT) {
3139 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3140 if (tjp != nullptr) {
3141 value_type = tjp;
3142 }
3143 }
3144 break;
3145 }
3146 case LS_cmp_swap:
3147 case LS_cmp_swap_weak:
3148 case LS_get_add:
3149 break;
3150 default:
3151 ShouldNotReachHere();
3152 }
3153
3154 // Null check receiver.
3155 receiver = null_check(receiver);
3156 if (stopped()) {
3157 return true;
3158 }
3159
3160 int alias_idx = C->get_alias_index(adr_type);
3161
3162 if (is_reference_type(type)) {
3163 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3164
3165 if (oldval != nullptr && oldval->is_InlineType()) {
3166 // Re-execute the unsafe access if allocation triggers deoptimization.
3167 PreserveReexecuteState preexecs(this);
3168 jvms()->set_should_reexecute(true);
3169 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3170 }
3171 if (newval != nullptr && newval->is_InlineType()) {
3172 // Re-execute the unsafe access if allocation triggers deoptimization.
3173 PreserveReexecuteState preexecs(this);
3174 jvms()->set_should_reexecute(true);
3175 newval = newval->as_InlineType()->buffer(this)->get_oop();
3176 }
3177
3178 // Transformation of a value which could be null pointer (CastPP #null)
3179 // could be delayed during Parse (for example, in adjust_map_after_if()).
3180 // Execute transformation here to avoid barrier generation in such case.
3181 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3182 newval = _gvn.makecon(TypePtr::NULL_PTR);
3183
3184 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3185 // Refine the value to a null constant, when it is known to be null
3186 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3187 }
3188 }
3189
3190 Node* result = nullptr;
3191 switch (kind) {
3192 case LS_cmp_exchange: {
3193 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3194 oldval, newval, value_type, type, decorators);
3195 break;
3196 }
3197 case LS_cmp_swap_weak:
3198 decorators |= C2_WEAK_CMPXCHG;
3199 case LS_cmp_swap: {
3200 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3201 oldval, newval, value_type, type, decorators);
3202 break;
3203 }
3204 case LS_get_set: {
3205 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3206 newval, value_type, type, decorators);
3207 break;
3208 }
3209 case LS_get_add: {
3210 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3211 newval, value_type, type, decorators);
3212 break;
3213 }
3214 default:
3215 ShouldNotReachHere();
3216 }
3217
3218 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3219 set_result(result);
3220 return true;
3221 }
3222
3223 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3224 // Regardless of form, don't allow previous ld/st to move down,
3225 // then issue acquire, release, or volatile mem_bar.
3226 insert_mem_bar(Op_MemBarCPUOrder);
3227 switch(id) {
3228 case vmIntrinsics::_loadFence:
3229 insert_mem_bar(Op_LoadFence);
3230 return true;
3231 case vmIntrinsics::_storeFence:
3232 insert_mem_bar(Op_StoreFence);
3233 return true;
3234 case vmIntrinsics::_storeStoreFence:
3235 insert_mem_bar(Op_StoreStoreFence);
3236 return true;
3237 case vmIntrinsics::_fullFence:
3238 insert_mem_bar(Op_MemBarVolatile);
3239 return true;
3240 default:
3241 fatal_unexpected_iid(id);
3242 return false;
3243 }
3244 }
3245
3246 // private native int arrayInstanceBaseOffset0(Object[] array);
3247 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3248 Node* array = argument(1);
3249 Node* klass_node = load_object_klass(array);
3250
3251 jint layout_con = Klass::_lh_neutral_value;
3252 Node* layout_val = get_layout_helper(klass_node, layout_con);
3253 int layout_is_con = (layout_val == nullptr);
3254
3255 Node* header_size = nullptr;
3256 if (layout_is_con) {
3257 int hsize = Klass::layout_helper_header_size(layout_con);
3258 header_size = intcon(hsize);
3259 } else {
3260 Node* hss = intcon(Klass::_lh_header_size_shift);
3261 Node* hsm = intcon(Klass::_lh_header_size_mask);
3262 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3263 header_size = _gvn.transform(new AndINode(header_size, hsm));
3264 }
3265 set_result(header_size);
3266 return true;
3267 }
3268
3269 // private native int arrayInstanceIndexScale0(Object[] array);
3270 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3271 Node* array = argument(1);
3272 Node* klass_node = load_object_klass(array);
3273
3274 jint layout_con = Klass::_lh_neutral_value;
3275 Node* layout_val = get_layout_helper(klass_node, layout_con);
3276 int layout_is_con = (layout_val == nullptr);
3277
3278 Node* element_size = nullptr;
3279 if (layout_is_con) {
3280 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3281 int elem_size = 1 << log_element_size;
3282 element_size = intcon(elem_size);
3283 } else {
3284 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3285 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3286 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3287 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3288 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3289 }
3290 set_result(element_size);
3291 return true;
3292 }
3293
3294 // private native int arrayLayout0(Object[] array);
3295 bool LibraryCallKit::inline_arrayLayout() {
3296 RegionNode* region = new RegionNode(2);
3297 Node* phi = new PhiNode(region, TypeInt::POS);
3298
3299 Node* array = argument(1);
3300 Node* klass_node = load_object_klass(array);
3301 generate_refArray_guard(klass_node, region);
3302 if (region->req() == 3) {
3303 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3304 }
3305
3306 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3307 Node* layout_kind_addr = basic_plus_adr(klass_node, layout_kind_offset);
3308 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3309
3310 region->init_req(1, control());
3311 phi->init_req(1, layout_kind);
3312
3313 set_control(_gvn.transform(region));
3314 set_result(_gvn.transform(phi));
3315 return true;
3316 }
3317
3318 // private native int[] getFieldMap0(Class <?> c);
3319 // int offset = c._klass._acmp_maps_offset;
3320 // return (int[])c.obj_field(offset);
3321 bool LibraryCallKit::inline_getFieldMap() {
3322 Node* mirror = argument(1);
3323 Node* klass = load_klass_from_mirror(mirror, false, nullptr, 0);
3324
3325 int field_map_offset_offset = in_bytes(InstanceKlass::acmp_maps_offset_offset());
3326 Node* field_map_offset_addr = basic_plus_adr(klass, field_map_offset_offset);
3327 Node* field_map_offset = make_load(nullptr, field_map_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
3328 field_map_offset = _gvn.transform(ConvI2L(field_map_offset));
3329
3330 Node* map_addr = basic_plus_adr(mirror, field_map_offset);
3331 const TypeAryPtr* val_type = TypeAryPtr::INTS->cast_to_ptr_type(TypePtr::NotNull)->with_offset(0);
3332 // TODO 8350865 Remove this
3333 val_type = val_type->cast_to_not_flat(true)->cast_to_not_null_free(true);
3334 Node* map = access_load_at(mirror, map_addr, TypeAryPtr::INTS, val_type, T_ARRAY, IN_HEAP | MO_UNORDERED);
3335
3336 set_result(map);
3337 return true;
3338 }
3339
3340 bool LibraryCallKit::inline_onspinwait() {
3341 insert_mem_bar(Op_OnSpinWait);
3342 return true;
3343 }
3344
3345 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3346 if (!kls->is_Con()) {
3347 return true;
3348 }
3349 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3350 if (klsptr == nullptr) {
3351 return true;
3352 }
3353 ciInstanceKlass* ik = klsptr->instance_klass();
3354 // don't need a guard for a klass that is already initialized
3355 return !ik->is_initialized();
3356 }
3357
3358 //----------------------------inline_unsafe_writeback0-------------------------
3359 // public native void Unsafe.writeback0(long address)
3360 bool LibraryCallKit::inline_unsafe_writeback0() {
3361 if (!Matcher::has_match_rule(Op_CacheWB)) {
3362 return false;
3363 }
3364 #ifndef PRODUCT
3365 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3366 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3367 ciSignature* sig = callee()->signature();
3368 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3369 #endif
3370 null_check_receiver(); // null-check, then ignore
3371 Node *addr = argument(1);
3372 addr = new CastX2PNode(addr);
3373 addr = _gvn.transform(addr);
3374 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3375 flush = _gvn.transform(flush);
3376 set_memory(flush, TypeRawPtr::BOTTOM);
3377 return true;
3378 }
3379
3380 //----------------------------inline_unsafe_writeback0-------------------------
3381 // public native void Unsafe.writeback0(long address)
3382 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3383 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3384 return false;
3385 }
3386 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3387 return false;
3388 }
3389 #ifndef PRODUCT
3390 assert(Matcher::has_match_rule(Op_CacheWB),
3391 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3392 : "found match rule for CacheWBPostSync but not CacheWB"));
3393
3394 #endif
3395 null_check_receiver(); // null-check, then ignore
3396 Node *sync;
3397 if (is_pre) {
3398 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3399 } else {
3400 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3401 }
3402 sync = _gvn.transform(sync);
3403 set_memory(sync, TypeRawPtr::BOTTOM);
3404 return true;
3405 }
3406
3407 //----------------------------inline_unsafe_allocate---------------------------
3408 // public native Object Unsafe.allocateInstance(Class<?> cls);
3409 bool LibraryCallKit::inline_unsafe_allocate() {
3410
3411 #if INCLUDE_JVMTI
3412 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3413 return false;
3414 }
3415 #endif //INCLUDE_JVMTI
3416
3417 if (callee()->is_static()) return false; // caller must have the capability!
3418
3419 null_check_receiver(); // null-check, then ignore
3420 Node* cls = null_check(argument(1));
3421 if (stopped()) return true;
3422
3423 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3424 kls = null_check(kls);
3425 if (stopped()) return true; // argument was like int.class
3426
3427 #if INCLUDE_JVMTI
3428 // Don't try to access new allocated obj in the intrinsic.
3429 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3430 // Deoptimize and allocate in interpreter instead.
3431 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3432 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3433 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3434 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3435 {
3436 BuildCutout unless(this, tst, PROB_MAX);
3437 uncommon_trap(Deoptimization::Reason_intrinsic,
3438 Deoptimization::Action_make_not_entrant);
3439 }
3440 if (stopped()) {
3441 return true;
3442 }
3443 #endif //INCLUDE_JVMTI
3444
3445 Node* test = nullptr;
3446 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3447 // Note: The argument might still be an illegal value like
3448 // Serializable.class or Object[].class. The runtime will handle it.
3449 // But we must make an explicit check for initialization.
3450 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3451 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3452 // can generate code to load it as unsigned byte.
3453 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3454 Node* bits = intcon(InstanceKlass::fully_initialized);
3455 test = _gvn.transform(new SubINode(inst, bits));
3456 // The 'test' is non-zero if we need to take a slow path.
3457 }
3458 Node* obj = nullptr;
3459 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3460 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3461 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3462 } else {
3463 obj = new_instance(kls, test);
3464 }
3465 set_result(obj);
3466 return true;
3467 }
3468
3469 //------------------------inline_native_time_funcs--------------
3470 // inline code for System.currentTimeMillis() and System.nanoTime()
3471 // these have the same type and signature
3472 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3473 const TypeFunc* tf = OptoRuntime::void_long_Type();
3474 const TypePtr* no_memory_effects = nullptr;
3475 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3476 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3477 #ifdef ASSERT
3478 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3479 assert(value_top == top(), "second value must be top");
3480 #endif
3481 set_result(value);
3482 return true;
3483 }
3484
3485 //--------------------inline_native_vthread_start_transition--------------------
3486 // inline void startTransition(boolean is_mount);
3487 // inline void startFinalTransition();
3488 // Pseudocode of implementation:
3489 //
3490 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3491 // carrier->set_is_in_vthread_transition(true);
3492 // OrderAccess::storeload();
3493 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3494 // + global_vthread_transition_disable_count();
3495 // if (disable_requests > 0) {
3496 // slow path: runtime call
3497 // }
3498 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3499 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3500 IdealKit ideal(this);
3501
3502 Node* thread = ideal.thread();
3503 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3504 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3505 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3506 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3507 insert_mem_bar(Op_MemBarVolatile);
3508 ideal.sync_kit(this);
3509
3510 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3511 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3512 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3513 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3514 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3515
3516 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3517 sync_kit(ideal);
3518 Node* is_mount = is_final_transition ? ideal.ConI(0) : _gvn.transform(argument(1));
3519 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3520 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3521 ideal.sync_kit(this);
3522 }
3523 ideal.end_if();
3524
3525 final_sync(ideal);
3526 return true;
3527 }
3528
3529 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3530 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3531 IdealKit ideal(this);
3532
3533 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3534 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3535
3536 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3537 sync_kit(ideal);
3538 Node* is_mount = is_first_transition ? ideal.ConI(1) : _gvn.transform(argument(1));
3539 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3540 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3541 ideal.sync_kit(this);
3542 } ideal.else_(); {
3543 Node* thread = ideal.thread();
3544 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3545 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3546
3547 sync_kit(ideal);
3548 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3549 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3550 ideal.sync_kit(this);
3551 } ideal.end_if();
3552
3553 final_sync(ideal);
3554 return true;
3555 }
3556
3557 #if INCLUDE_JVMTI
3558
3559 // Always update the is_disable_suspend bit.
3560 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3561 if (!DoJVMTIVirtualThreadTransitions) {
3562 return true;
3563 }
3564 IdealKit ideal(this);
3565
3566 {
3567 // unconditionally update the is_disable_suspend bit in current JavaThread
3568 Node* thread = ideal.thread();
3569 Node* arg = _gvn.transform(argument(0)); // argument for notification
3570 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3571 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3572
3573 sync_kit(ideal);
3574 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3575 ideal.sync_kit(this);
3576 }
3577 final_sync(ideal);
3578
3579 return true;
3580 }
3581
3582 #endif // INCLUDE_JVMTI
3583
3584 #ifdef JFR_HAVE_INTRINSICS
3585
3586 /**
3587 * if oop->klass != null
3588 * // normal class
3589 * epoch = _epoch_state ? 2 : 1
3590 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3591 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3592 * }
3593 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3594 * else
3595 * // primitive class
3596 * if oop->array_klass != null
3597 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3598 * else
3599 * id = LAST_TYPE_ID + 1 // void class path
3600 * if (!signaled)
3601 * signaled = true
3602 */
3603 bool LibraryCallKit::inline_native_classID() {
3604 Node* cls = argument(0);
3605
3606 IdealKit ideal(this);
3607 #define __ ideal.
3608 IdealVariable result(ideal); __ declarations_done();
3609 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3610 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3611 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3612
3613
3614 __ if_then(kls, BoolTest::ne, null()); {
3615 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3616 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3617
3618 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3619 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3620 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3621 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3622 mask = _gvn.transform(new OrLNode(mask, epoch));
3623 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3624
3625 float unlikely = PROB_UNLIKELY(0.999);
3626 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3627 sync_kit(ideal);
3628 make_runtime_call(RC_LEAF,
3629 OptoRuntime::class_id_load_barrier_Type(),
3630 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3631 "class id load barrier",
3632 TypePtr::BOTTOM,
3633 kls);
3634 ideal.sync_kit(this);
3635 } __ end_if();
3636
3637 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3638 } __ else_(); {
3639 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3640 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3641 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3642 __ if_then(array_kls, BoolTest::ne, null()); {
3643 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3644 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3645 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3646 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3647 } __ else_(); {
3648 // void class case
3649 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3650 } __ end_if();
3651
3652 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3653 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3654 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3655 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3656 } __ end_if();
3657 } __ end_if();
3658
3659 final_sync(ideal);
3660 set_result(ideal.value(result));
3661 #undef __
3662 return true;
3663 }
3664
3665 //------------------------inline_native_jvm_commit------------------
3666 bool LibraryCallKit::inline_native_jvm_commit() {
3667 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3668
3669 // Save input memory and i_o state.
3670 Node* input_memory_state = reset_memory();
3671 set_all_memory(input_memory_state);
3672 Node* input_io_state = i_o();
3673
3674 // TLS.
3675 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3676 // Jfr java buffer.
3677 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3678 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3679 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3680
3681 // Load the current value of the notified field in the JfrThreadLocal.
3682 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3683 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3684
3685 // Test for notification.
3686 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3687 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3688 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3689
3690 // True branch, is notified.
3691 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3692 set_control(is_notified);
3693
3694 // Reset notified state.
3695 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3696 Node* notified_reset_memory = reset_memory();
3697
3698 // 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.
3699 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3700 // Convert the machine-word to a long.
3701 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3702
3703 // False branch, not notified.
3704 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3705 set_control(not_notified);
3706 set_all_memory(input_memory_state);
3707
3708 // Arg is the next position as a long.
3709 Node* arg = argument(0);
3710 // Convert long to machine-word.
3711 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3712
3713 // Store the next_position to the underlying jfr java buffer.
3714 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3715
3716 Node* commit_memory = reset_memory();
3717 set_all_memory(commit_memory);
3718
3719 // 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.
3720 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3721 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3722 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3723
3724 // And flags with lease constant.
3725 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3726
3727 // Branch on lease to conditionalize returning the leased java buffer.
3728 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3729 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3730 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3731
3732 // False branch, not a lease.
3733 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3734
3735 // True branch, is lease.
3736 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3737 set_control(is_lease);
3738
3739 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3740 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3741 OptoRuntime::void_void_Type(),
3742 SharedRuntime::jfr_return_lease(),
3743 "return_lease", TypePtr::BOTTOM);
3744 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3745
3746 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3747 record_for_igvn(lease_compare_rgn);
3748 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3749 record_for_igvn(lease_compare_mem);
3750 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3751 record_for_igvn(lease_compare_io);
3752 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3753 record_for_igvn(lease_result_value);
3754
3755 // Update control and phi nodes.
3756 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3757 lease_compare_rgn->init_req(_false_path, not_lease);
3758
3759 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3760 lease_compare_mem->init_req(_false_path, commit_memory);
3761
3762 lease_compare_io->init_req(_true_path, i_o());
3763 lease_compare_io->init_req(_false_path, input_io_state);
3764
3765 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3766 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3767
3768 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3769 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3770 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3771 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3772
3773 // Update control and phi nodes.
3774 result_rgn->init_req(_true_path, is_notified);
3775 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3776
3777 result_mem->init_req(_true_path, notified_reset_memory);
3778 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3779
3780 result_io->init_req(_true_path, input_io_state);
3781 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3782
3783 result_value->init_req(_true_path, current_pos);
3784 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3785
3786 // Set output state.
3787 set_control(_gvn.transform(result_rgn));
3788 set_all_memory(_gvn.transform(result_mem));
3789 set_i_o(_gvn.transform(result_io));
3790 set_result(result_rgn, result_value);
3791 return true;
3792 }
3793
3794 /*
3795 * The intrinsic is a model of this pseudo-code:
3796 *
3797 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3798 * jobject h_event_writer = tl->java_event_writer();
3799 * if (h_event_writer == nullptr) {
3800 * return nullptr;
3801 * }
3802 * oop threadObj = Thread::threadObj();
3803 * oop vthread = java_lang_Thread::vthread(threadObj);
3804 * traceid tid;
3805 * bool pinVirtualThread;
3806 * bool excluded;
3807 * if (vthread != threadObj) { // i.e. current thread is virtual
3808 * tid = java_lang_Thread::tid(vthread);
3809 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3810 * pinVirtualThread = VMContinuations;
3811 * excluded = vthread_epoch_raw & excluded_mask;
3812 * if (!excluded) {
3813 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3814 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3815 * if (vthread_epoch != current_epoch) {
3816 * write_checkpoint();
3817 * }
3818 * }
3819 * } else {
3820 * tid = java_lang_Thread::tid(threadObj);
3821 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3822 * pinVirtualThread = false;
3823 * excluded = thread_epoch_raw & excluded_mask;
3824 * }
3825 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3826 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3827 * if (tid_in_event_writer != tid) {
3828 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3829 * setField(event_writer, "excluded", excluded);
3830 * setField(event_writer, "threadID", tid);
3831 * }
3832 * return event_writer
3833 */
3834 bool LibraryCallKit::inline_native_getEventWriter() {
3835 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3836
3837 // Save input memory and i_o state.
3838 Node* input_memory_state = reset_memory();
3839 set_all_memory(input_memory_state);
3840 Node* input_io_state = i_o();
3841
3842 // The most significant bit of the u2 is used to denote thread exclusion
3843 Node* excluded_shift = _gvn.intcon(15);
3844 Node* excluded_mask = _gvn.intcon(1 << 15);
3845 // The epoch generation is the range [1-32767]
3846 Node* epoch_mask = _gvn.intcon(32767);
3847
3848 // TLS
3849 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3850
3851 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3852 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3853
3854 // Load the eventwriter jobject handle.
3855 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3856
3857 // Null check the jobject handle.
3858 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3859 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3860 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3861
3862 // False path, jobj is null.
3863 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3864
3865 // True path, jobj is not null.
3866 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3867
3868 set_control(jobj_is_not_null);
3869
3870 // Load the threadObj for the CarrierThread.
3871 Node* threadObj = generate_current_thread(tls_ptr);
3872
3873 // Load the vthread.
3874 Node* vthread = generate_virtual_thread(tls_ptr);
3875
3876 // If vthread != threadObj, this is a virtual thread.
3877 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3878 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3879 IfNode* iff_vthread_not_equal_threadObj =
3880 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3881
3882 // False branch, fallback to threadObj.
3883 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3884 set_control(vthread_equal_threadObj);
3885
3886 // Load the tid field from the vthread object.
3887 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3888
3889 // Load the raw epoch value from the threadObj.
3890 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3891 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3892 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3893 TypeInt::CHAR, T_CHAR,
3894 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3895
3896 // Mask off the excluded information from the epoch.
3897 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3898
3899 // True branch, this is a virtual thread.
3900 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3901 set_control(vthread_not_equal_threadObj);
3902
3903 // Load the tid field from the vthread object.
3904 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3905
3906 // Continuation support determines if a virtual thread should be pinned.
3907 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3908 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3909
3910 // Load the raw epoch value from the vthread.
3911 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3912 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3913 TypeInt::CHAR, T_CHAR,
3914 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3915
3916 // Mask off the excluded information from the epoch.
3917 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3918
3919 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3920 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3921 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3922 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3923
3924 // False branch, vthread is excluded, no need to write epoch info.
3925 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3926
3927 // True branch, vthread is included, update epoch info.
3928 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3929 set_control(included);
3930
3931 // Get epoch value.
3932 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3933
3934 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3935 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3936 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3937
3938 // Compare the epoch in the vthread to the current epoch generation.
3939 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3940 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3941 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3942
3943 // False path, epoch is equal, checkpoint information is valid.
3944 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3945
3946 // True path, epoch is not equal, write a checkpoint for the vthread.
3947 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3948
3949 set_control(epoch_is_not_equal);
3950
3951 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3952 // The call also updates the native thread local thread id and the vthread with the current epoch.
3953 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3954 OptoRuntime::jfr_write_checkpoint_Type(),
3955 SharedRuntime::jfr_write_checkpoint(),
3956 "write_checkpoint", TypePtr::BOTTOM);
3957 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3958
3959 // vthread epoch != current epoch
3960 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3961 record_for_igvn(epoch_compare_rgn);
3962 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3963 record_for_igvn(epoch_compare_mem);
3964 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3965 record_for_igvn(epoch_compare_io);
3966
3967 // Update control and phi nodes.
3968 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3969 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3970 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3971 epoch_compare_mem->init_req(_false_path, input_memory_state);
3972 epoch_compare_io->init_req(_true_path, i_o());
3973 epoch_compare_io->init_req(_false_path, input_io_state);
3974
3975 // excluded != true
3976 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3977 record_for_igvn(exclude_compare_rgn);
3978 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3979 record_for_igvn(exclude_compare_mem);
3980 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3981 record_for_igvn(exclude_compare_io);
3982
3983 // Update control and phi nodes.
3984 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3985 exclude_compare_rgn->init_req(_false_path, excluded);
3986 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3987 exclude_compare_mem->init_req(_false_path, input_memory_state);
3988 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3989 exclude_compare_io->init_req(_false_path, input_io_state);
3990
3991 // vthread != threadObj
3992 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3993 record_for_igvn(vthread_compare_rgn);
3994 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3995 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3996 record_for_igvn(vthread_compare_io);
3997 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3998 record_for_igvn(tid);
3999 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
4000 record_for_igvn(exclusion);
4001 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
4002 record_for_igvn(pinVirtualThread);
4003
4004 // Update control and phi nodes.
4005 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
4006 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
4007 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
4008 vthread_compare_mem->init_req(_false_path, input_memory_state);
4009 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
4010 vthread_compare_io->init_req(_false_path, input_io_state);
4011 tid->init_req(_true_path, _gvn.transform(vthread_tid));
4012 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
4013 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4014 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
4015 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
4016 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
4017
4018 // Update branch state.
4019 set_control(_gvn.transform(vthread_compare_rgn));
4020 set_all_memory(_gvn.transform(vthread_compare_mem));
4021 set_i_o(_gvn.transform(vthread_compare_io));
4022
4023 // Load the event writer oop by dereferencing the jobject handle.
4024 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
4025 assert(klass_EventWriter->is_loaded(), "invariant");
4026 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
4027 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
4028 const TypeOopPtr* const xtype = aklass->as_instance_type();
4029 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
4030 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
4031
4032 // Load the current thread id from the event writer object.
4033 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
4034 // Get the field offset to, conditionally, store an updated tid value later.
4035 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
4036 // Get the field offset to, conditionally, store an updated exclusion value later.
4037 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
4038 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
4039 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
4040
4041 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
4042 record_for_igvn(event_writer_tid_compare_rgn);
4043 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4044 record_for_igvn(event_writer_tid_compare_mem);
4045 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
4046 record_for_igvn(event_writer_tid_compare_io);
4047
4048 // Compare the current tid from the thread object to what is currently stored in the event writer object.
4049 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
4050 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
4051 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
4052
4053 // False path, tids are the same.
4054 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
4055
4056 // True path, tid is not equal, need to update the tid in the event writer.
4057 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
4058 record_for_igvn(tid_is_not_equal);
4059
4060 // Store the pin state to the event writer.
4061 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
4062
4063 // Store the exclusion state to the event writer.
4064 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
4065 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
4066
4067 // Store the tid to the event writer.
4068 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
4069
4070 // Update control and phi nodes.
4071 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
4072 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
4073 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4074 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
4075 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
4076 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
4077
4078 // Result of top level CFG, Memory, IO and Value.
4079 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4080 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4081 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4082 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4083
4084 // Result control.
4085 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4086 result_rgn->init_req(_false_path, jobj_is_null);
4087
4088 // Result memory.
4089 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4090 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4091
4092 // Result IO.
4093 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4094 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4095
4096 // Result value.
4097 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4098 result_value->init_req(_false_path, null()); // return null
4099
4100 // Set output state.
4101 set_control(_gvn.transform(result_rgn));
4102 set_all_memory(_gvn.transform(result_mem));
4103 set_i_o(_gvn.transform(result_io));
4104 set_result(result_rgn, result_value);
4105 return true;
4106 }
4107
4108 /*
4109 * The intrinsic is a model of this pseudo-code:
4110 *
4111 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4112 * if (carrierThread != thread) { // is virtual thread
4113 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4114 * bool excluded = vthread_epoch_raw & excluded_mask;
4115 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4116 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
4117 * if (!excluded) {
4118 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4119 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
4120 * }
4121 * AtomicAccess::release_store(&tl->_vthread, true);
4122 * return;
4123 * }
4124 * AtomicAccess::release_store(&tl->_vthread, false);
4125 */
4126 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4127 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4128
4129 Node* input_memory_state = reset_memory();
4130 set_all_memory(input_memory_state);
4131
4132 // The most significant bit of the u2 is used to denote thread exclusion
4133 Node* excluded_mask = _gvn.intcon(1 << 15);
4134 // The epoch generation is the range [1-32767]
4135 Node* epoch_mask = _gvn.intcon(32767);
4136
4137 Node* const carrierThread = generate_current_thread(jt);
4138 // If thread != carrierThread, this is a virtual thread.
4139 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4140 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4141 IfNode* iff_thread_not_equal_carrierThread =
4142 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4143
4144 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4145
4146 // False branch, is carrierThread.
4147 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4148 // Store release
4149 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4150
4151 set_all_memory(input_memory_state);
4152
4153 // True branch, is virtual thread.
4154 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4155 set_control(thread_not_equal_carrierThread);
4156
4157 // Load the raw epoch value from the vthread.
4158 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4159 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4160 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4161
4162 // Mask off the excluded information from the epoch.
4163 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4164
4165 // Load the tid field from the thread.
4166 Node* tid = load_field_from_object(thread, "tid", "J");
4167
4168 // Store the vthread tid to the jfr thread local.
4169 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4170 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4171
4172 // Branch is_excluded to conditionalize updating the epoch .
4173 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4174 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4175 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4176
4177 // True branch, vthread is excluded, no need to write epoch info.
4178 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4179 set_control(excluded);
4180 Node* vthread_is_excluded = _gvn.intcon(1);
4181
4182 // False branch, vthread is included, update epoch info.
4183 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4184 set_control(included);
4185 Node* vthread_is_included = _gvn.intcon(0);
4186
4187 // Get epoch value.
4188 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4189
4190 // Store the vthread epoch to the jfr thread local.
4191 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4192 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4193
4194 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4195 record_for_igvn(excluded_rgn);
4196 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4197 record_for_igvn(excluded_mem);
4198 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4199 record_for_igvn(exclusion);
4200
4201 // Merge the excluded control and memory.
4202 excluded_rgn->init_req(_true_path, excluded);
4203 excluded_rgn->init_req(_false_path, included);
4204 excluded_mem->init_req(_true_path, tid_memory);
4205 excluded_mem->init_req(_false_path, included_memory);
4206 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4207 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4208
4209 // Set intermediate state.
4210 set_control(_gvn.transform(excluded_rgn));
4211 set_all_memory(excluded_mem);
4212
4213 // Store the vthread exclusion state to the jfr thread local.
4214 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4215 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4216
4217 // Store release
4218 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4219
4220 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4221 record_for_igvn(thread_compare_rgn);
4222 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4223 record_for_igvn(thread_compare_mem);
4224 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4225 record_for_igvn(vthread);
4226
4227 // Merge the thread_compare control and memory.
4228 thread_compare_rgn->init_req(_true_path, control());
4229 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4230 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4231 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4232
4233 // Set output state.
4234 set_control(_gvn.transform(thread_compare_rgn));
4235 set_all_memory(_gvn.transform(thread_compare_mem));
4236 }
4237
4238 #endif // JFR_HAVE_INTRINSICS
4239
4240 //------------------------inline_native_currentCarrierThread------------------
4241 bool LibraryCallKit::inline_native_currentCarrierThread() {
4242 Node* junk = nullptr;
4243 set_result(generate_current_thread(junk));
4244 return true;
4245 }
4246
4247 //------------------------inline_native_currentThread------------------
4248 bool LibraryCallKit::inline_native_currentThread() {
4249 Node* junk = nullptr;
4250 set_result(generate_virtual_thread(junk));
4251 return true;
4252 }
4253
4254 //------------------------inline_native_setVthread------------------
4255 bool LibraryCallKit::inline_native_setCurrentThread() {
4256 assert(C->method()->changes_current_thread(),
4257 "method changes current Thread but is not annotated ChangesCurrentThread");
4258 Node* arr = argument(1);
4259 Node* thread = _gvn.transform(new ThreadLocalNode());
4260 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4261 Node* thread_obj_handle
4262 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4263 thread_obj_handle = _gvn.transform(thread_obj_handle);
4264 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4265 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4266
4267 // Change the _monitor_owner_id of the JavaThread
4268 Node* tid = load_field_from_object(arr, "tid", "J");
4269 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4270 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4271
4272 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4273 return true;
4274 }
4275
4276 const Type* LibraryCallKit::scopedValueCache_type() {
4277 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4278 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4279 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true, true);
4280
4281 // Because we create the scopedValue cache lazily we have to make the
4282 // type of the result BotPTR.
4283 bool xk = etype->klass_is_exact();
4284 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4285 return objects_type;
4286 }
4287
4288 Node* LibraryCallKit::scopedValueCache_helper() {
4289 Node* thread = _gvn.transform(new ThreadLocalNode());
4290 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4291 // We cannot use immutable_memory() because we might flip onto a
4292 // different carrier thread, at which point we'll need to use that
4293 // carrier thread's cache.
4294 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4295 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4296 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4297 }
4298
4299 //------------------------inline_native_scopedValueCache------------------
4300 bool LibraryCallKit::inline_native_scopedValueCache() {
4301 Node* cache_obj_handle = scopedValueCache_helper();
4302 const Type* objects_type = scopedValueCache_type();
4303 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4304
4305 return true;
4306 }
4307
4308 //------------------------inline_native_setScopedValueCache------------------
4309 bool LibraryCallKit::inline_native_setScopedValueCache() {
4310 Node* arr = argument(0);
4311 Node* cache_obj_handle = scopedValueCache_helper();
4312 const Type* objects_type = scopedValueCache_type();
4313
4314 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4315 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4316
4317 return true;
4318 }
4319
4320 //------------------------inline_native_Continuation_pin and unpin-----------
4321
4322 // Shared implementation routine for both pin and unpin.
4323 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4324 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4325
4326 // Save input memory.
4327 Node* input_memory_state = reset_memory();
4328 set_all_memory(input_memory_state);
4329
4330 // TLS
4331 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4332 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4333 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4334
4335 // Null check the last continuation object.
4336 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4337 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4338 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4339
4340 // False path, last continuation is null.
4341 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4342
4343 // True path, last continuation is not null.
4344 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4345
4346 set_control(continuation_is_not_null);
4347
4348 // Load the pin count from the last continuation.
4349 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4350 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4351
4352 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4353 Node* pin_count_rhs;
4354 if (unpin) {
4355 pin_count_rhs = _gvn.intcon(0);
4356 } else {
4357 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4358 }
4359 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4360 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4361 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4362
4363 // True branch, pin count over/underflow.
4364 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4365 {
4366 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4367 // which will throw IllegalStateException for pin count over/underflow.
4368 // No memory changed so far - we can use memory create by reset_memory()
4369 // at the beginning of this intrinsic. No need to call reset_memory() again.
4370 PreserveJVMState pjvms(this);
4371 set_control(pin_count_over_underflow);
4372 uncommon_trap(Deoptimization::Reason_intrinsic,
4373 Deoptimization::Action_none);
4374 assert(stopped(), "invariant");
4375 }
4376
4377 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4378 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4379 set_control(valid_pin_count);
4380
4381 Node* next_pin_count;
4382 if (unpin) {
4383 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4384 } else {
4385 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4386 }
4387
4388 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4389
4390 // Result of top level CFG and Memory.
4391 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4392 record_for_igvn(result_rgn);
4393 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4394 record_for_igvn(result_mem);
4395
4396 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4397 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4398 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4399 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4400
4401 // Set output state.
4402 set_control(_gvn.transform(result_rgn));
4403 set_all_memory(_gvn.transform(result_mem));
4404
4405 return true;
4406 }
4407
4408 //---------------------------load_mirror_from_klass----------------------------
4409 // Given a klass oop, load its java mirror (a java.lang.Class oop).
4410 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
4411 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
4412 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4413 // mirror = ((OopHandle)mirror)->resolve();
4414 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
4415 }
4416
4417 //-----------------------load_klass_from_mirror_common-------------------------
4418 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4419 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4420 // and branch to the given path on the region.
4421 // If never_see_null, take an uncommon trap on null, so we can optimistically
4422 // compile for the non-null case.
4423 // If the region is null, force never_see_null = true.
4424 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4425 bool never_see_null,
4426 RegionNode* region,
4427 int null_path,
4428 int offset) {
4429 if (region == nullptr) never_see_null = true;
4430 Node* p = basic_plus_adr(mirror, offset);
4431 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4432 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4433 Node* null_ctl = top();
4434 kls = null_check_oop(kls, &null_ctl, never_see_null);
4435 if (region != nullptr) {
4436 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4437 region->init_req(null_path, null_ctl);
4438 } else {
4439 assert(null_ctl == top(), "no loose ends");
4440 }
4441 return kls;
4442 }
4443
4444 //--------------------(inline_native_Class_query helpers)---------------------
4445 // Use this for JVM_ACC_INTERFACE.
4446 // Fall through if (mods & mask) == bits, take the guard otherwise.
4447 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4448 ByteSize offset, const Type* type, BasicType bt) {
4449 // Branch around if the given klass has the given modifier bit set.
4450 // Like generate_guard, adds a new path onto the region.
4451 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4452 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4453 Node* mask = intcon(modifier_mask);
4454 Node* bits = intcon(modifier_bits);
4455 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4456 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4457 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4458 return generate_fair_guard(bol, region);
4459 }
4460
4461 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4462 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4463 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4464 }
4465
4466 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4467 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4468 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4469 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4470 }
4471
4472 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4473 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4474 }
4475
4476 //-------------------------inline_native_Class_query-------------------
4477 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4478 const Type* return_type = TypeInt::BOOL;
4479 Node* prim_return_value = top(); // what happens if it's a primitive class?
4480 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4481 bool expect_prim = false; // most of these guys expect to work on refs
4482
4483 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4484
4485 Node* mirror = argument(0);
4486 Node* obj = top();
4487
4488 switch (id) {
4489 case vmIntrinsics::_isInstance:
4490 // nothing is an instance of a primitive type
4491 prim_return_value = intcon(0);
4492 obj = argument(1);
4493 break;
4494 case vmIntrinsics::_isHidden:
4495 prim_return_value = intcon(0);
4496 break;
4497 case vmIntrinsics::_getSuperclass:
4498 prim_return_value = null();
4499 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4500 break;
4501 default:
4502 fatal_unexpected_iid(id);
4503 break;
4504 }
4505
4506 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4507 if (mirror_con == nullptr) return false; // cannot happen?
4508
4509 #ifndef PRODUCT
4510 if (C->print_intrinsics() || C->print_inlining()) {
4511 ciType* k = mirror_con->java_mirror_type();
4512 if (k) {
4513 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4514 k->print_name();
4515 tty->cr();
4516 }
4517 }
4518 #endif
4519
4520 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4521 RegionNode* region = new RegionNode(PATH_LIMIT);
4522 record_for_igvn(region);
4523 PhiNode* phi = new PhiNode(region, return_type);
4524
4525 // The mirror will never be null of Reflection.getClassAccessFlags, however
4526 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4527 // if it is. See bug 4774291.
4528
4529 // For Reflection.getClassAccessFlags(), the null check occurs in
4530 // the wrong place; see inline_unsafe_access(), above, for a similar
4531 // situation.
4532 mirror = null_check(mirror);
4533 // If mirror or obj is dead, only null-path is taken.
4534 if (stopped()) return true;
4535
4536 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4537
4538 // Now load the mirror's klass metaobject, and null-check it.
4539 // Side-effects region with the control path if the klass is null.
4540 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4541 // If kls is null, we have a primitive mirror.
4542 phi->init_req(_prim_path, prim_return_value);
4543 if (stopped()) { set_result(region, phi); return true; }
4544 bool safe_for_replace = (region->in(_prim_path) == top());
4545
4546 Node* p; // handy temp
4547 Node* null_ctl;
4548
4549 // Now that we have the non-null klass, we can perform the real query.
4550 // For constant classes, the query will constant-fold in LoadNode::Value.
4551 Node* query_value = top();
4552 switch (id) {
4553 case vmIntrinsics::_isInstance:
4554 // nothing is an instance of a primitive type
4555 query_value = gen_instanceof(obj, kls, safe_for_replace);
4556 break;
4557
4558 case vmIntrinsics::_isHidden:
4559 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4560 if (generate_hidden_class_guard(kls, region) != nullptr)
4561 // A guard was added. If the guard is taken, it was an hidden class.
4562 phi->add_req(intcon(1));
4563 // If we fall through, it's a plain class.
4564 query_value = intcon(0);
4565 break;
4566
4567
4568 case vmIntrinsics::_getSuperclass:
4569 // The rules here are somewhat unfortunate, but we can still do better
4570 // with random logic than with a JNI call.
4571 // Interfaces store null or Object as _super, but must report null.
4572 // Arrays store an intermediate super as _super, but must report Object.
4573 // Other types can report the actual _super.
4574 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4575 if (generate_array_guard(kls, region) != nullptr) {
4576 // A guard was added. If the guard is taken, it was an array.
4577 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4578 }
4579 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4580 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4581 if (generate_interface_guard(kls, region) != nullptr) {
4582 // A guard was added. If the guard is taken, it was an interface.
4583 phi->add_req(null());
4584 }
4585 // If we fall through, it's a plain class. Get its _super.
4586 if (!stopped()) {
4587 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4588 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4589 null_ctl = top();
4590 kls = null_check_oop(kls, &null_ctl);
4591 if (null_ctl != top()) {
4592 // If the guard is taken, Object.superClass is null (both klass and mirror).
4593 region->add_req(null_ctl);
4594 phi ->add_req(null());
4595 }
4596 if (!stopped()) {
4597 query_value = load_mirror_from_klass(kls);
4598 }
4599 }
4600 break;
4601
4602 default:
4603 fatal_unexpected_iid(id);
4604 break;
4605 }
4606
4607 // Fall-through is the normal case of a query to a real class.
4608 phi->init_req(1, query_value);
4609 region->init_req(1, control());
4610
4611 C->set_has_split_ifs(true); // Has chance for split-if optimization
4612 set_result(region, phi);
4613 return true;
4614 }
4615
4616
4617 //-------------------------inline_Class_cast-------------------
4618 bool LibraryCallKit::inline_Class_cast() {
4619 Node* mirror = argument(0); // Class
4620 Node* obj = argument(1);
4621 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4622 if (mirror_con == nullptr) {
4623 return false; // dead path (mirror->is_top()).
4624 }
4625 if (obj == nullptr || obj->is_top()) {
4626 return false; // dead path
4627 }
4628 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4629
4630 // First, see if Class.cast() can be folded statically.
4631 // java_mirror_type() returns non-null for compile-time Class constants.
4632 ciType* tm = mirror_con->java_mirror_type();
4633 if (tm != nullptr && tm->is_klass() &&
4634 tp != nullptr) {
4635 if (!tp->is_loaded()) {
4636 // Don't use intrinsic when class is not loaded.
4637 return false;
4638 } else {
4639 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4640 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4641 if (static_res == Compile::SSC_always_true) {
4642 // isInstance() is true - fold the code.
4643 set_result(obj);
4644 return true;
4645 } else if (static_res == Compile::SSC_always_false) {
4646 // Don't use intrinsic, have to throw ClassCastException.
4647 // If the reference is null, the non-intrinsic bytecode will
4648 // be optimized appropriately.
4649 return false;
4650 }
4651 }
4652 }
4653
4654 // Bailout intrinsic and do normal inlining if exception path is frequent.
4655 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4656 return false;
4657 }
4658
4659 // Generate dynamic checks.
4660 // Class.cast() is java implementation of _checkcast bytecode.
4661 // Do checkcast (Parse::do_checkcast()) optimizations here.
4662
4663 mirror = null_check(mirror);
4664 // If mirror is dead, only null-path is taken.
4665 if (stopped()) {
4666 return true;
4667 }
4668
4669 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4670 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4671 RegionNode* region = new RegionNode(PATH_LIMIT);
4672 record_for_igvn(region);
4673
4674 // Now load the mirror's klass metaobject, and null-check it.
4675 // If kls is null, we have a primitive mirror and
4676 // nothing is an instance of a primitive type.
4677 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4678
4679 Node* res = top();
4680 Node* io = i_o();
4681 Node* mem = merged_memory();
4682 if (!stopped()) {
4683
4684 Node* bad_type_ctrl = top();
4685 // Do checkcast optimizations.
4686 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4687 region->init_req(_bad_type_path, bad_type_ctrl);
4688 }
4689 if (region->in(_prim_path) != top() ||
4690 region->in(_bad_type_path) != top() ||
4691 region->in(_npe_path) != top()) {
4692 // Let Interpreter throw ClassCastException.
4693 PreserveJVMState pjvms(this);
4694 set_control(_gvn.transform(region));
4695 // Set IO and memory because gen_checkcast may override them when buffering inline types
4696 set_i_o(io);
4697 set_all_memory(mem);
4698 uncommon_trap(Deoptimization::Reason_intrinsic,
4699 Deoptimization::Action_maybe_recompile);
4700 }
4701 if (!stopped()) {
4702 set_result(res);
4703 }
4704 return true;
4705 }
4706
4707
4708 //--------------------------inline_native_subtype_check------------------------
4709 // This intrinsic takes the JNI calls out of the heart of
4710 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4711 bool LibraryCallKit::inline_native_subtype_check() {
4712 // Pull both arguments off the stack.
4713 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4714 args[0] = argument(0);
4715 args[1] = argument(1);
4716 Node* klasses[2]; // corresponding Klasses: superk, subk
4717 klasses[0] = klasses[1] = top();
4718
4719 enum {
4720 // A full decision tree on {superc is prim, subc is prim}:
4721 _prim_0_path = 1, // {P,N} => false
4722 // {P,P} & superc!=subc => false
4723 _prim_same_path, // {P,P} & superc==subc => true
4724 _prim_1_path, // {N,P} => false
4725 _ref_subtype_path, // {N,N} & subtype check wins => true
4726 _both_ref_path, // {N,N} & subtype check loses => false
4727 PATH_LIMIT
4728 };
4729
4730 RegionNode* region = new RegionNode(PATH_LIMIT);
4731 RegionNode* prim_region = new RegionNode(2);
4732 Node* phi = new PhiNode(region, TypeInt::BOOL);
4733 record_for_igvn(region);
4734 record_for_igvn(prim_region);
4735
4736 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4737 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4738 int class_klass_offset = java_lang_Class::klass_offset();
4739
4740 // First null-check both mirrors and load each mirror's klass metaobject.
4741 int which_arg;
4742 for (which_arg = 0; which_arg <= 1; which_arg++) {
4743 Node* arg = args[which_arg];
4744 arg = null_check(arg);
4745 if (stopped()) break;
4746 args[which_arg] = arg;
4747
4748 Node* p = basic_plus_adr(arg, class_klass_offset);
4749 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4750 klasses[which_arg] = _gvn.transform(kls);
4751 }
4752
4753 // Having loaded both klasses, test each for null.
4754 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4755 for (which_arg = 0; which_arg <= 1; which_arg++) {
4756 Node* kls = klasses[which_arg];
4757 Node* null_ctl = top();
4758 kls = null_check_oop(kls, &null_ctl, never_see_null);
4759 if (which_arg == 0) {
4760 prim_region->init_req(1, null_ctl);
4761 } else {
4762 region->init_req(_prim_1_path, null_ctl);
4763 }
4764 if (stopped()) break;
4765 klasses[which_arg] = kls;
4766 }
4767
4768 if (!stopped()) {
4769 // now we have two reference types, in klasses[0..1]
4770 Node* subk = klasses[1]; // the argument to isAssignableFrom
4771 Node* superk = klasses[0]; // the receiver
4772 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4773 region->set_req(_ref_subtype_path, control());
4774 }
4775
4776 // If both operands are primitive (both klasses null), then
4777 // we must return true when they are identical primitives.
4778 // It is convenient to test this after the first null klass check.
4779 // This path is also used if superc is a value mirror.
4780 set_control(_gvn.transform(prim_region));
4781 if (!stopped()) {
4782 // Since superc is primitive, make a guard for the superc==subc case.
4783 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4784 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4785 generate_fair_guard(bol_eq, region);
4786 if (region->req() == PATH_LIMIT+1) {
4787 // A guard was added. If the added guard is taken, superc==subc.
4788 region->swap_edges(PATH_LIMIT, _prim_same_path);
4789 region->del_req(PATH_LIMIT);
4790 }
4791 region->set_req(_prim_0_path, control()); // Not equal after all.
4792 }
4793
4794 // these are the only paths that produce 'true':
4795 phi->set_req(_prim_same_path, intcon(1));
4796 phi->set_req(_ref_subtype_path, intcon(1));
4797
4798 // pull together the cases:
4799 assert(region->req() == PATH_LIMIT, "sane region");
4800 for (uint i = 1; i < region->req(); i++) {
4801 Node* ctl = region->in(i);
4802 if (ctl == nullptr || ctl == top()) {
4803 region->set_req(i, top());
4804 phi ->set_req(i, top());
4805 } else if (phi->in(i) == nullptr) {
4806 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4807 }
4808 }
4809
4810 set_control(_gvn.transform(region));
4811 set_result(_gvn.transform(phi));
4812 return true;
4813 }
4814
4815 //---------------------generate_array_guard_common------------------------
4816 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4817
4818 if (stopped()) {
4819 return nullptr;
4820 }
4821
4822 // Like generate_guard, adds a new path onto the region.
4823 jint layout_con = 0;
4824 Node* layout_val = get_layout_helper(kls, layout_con);
4825 if (layout_val == nullptr) {
4826 bool query = 0;
4827 switch(kind) {
4828 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4829 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4830 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4831 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4832 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4833 default:
4834 ShouldNotReachHere();
4835 }
4836 if (!query) {
4837 return nullptr; // never a branch
4838 } else { // always a branch
4839 Node* always_branch = control();
4840 if (region != nullptr)
4841 region->add_req(always_branch);
4842 set_control(top());
4843 return always_branch;
4844 }
4845 }
4846 unsigned int value = 0;
4847 BoolTest::mask btest = BoolTest::illegal;
4848 switch(kind) {
4849 case RefArray:
4850 case NonRefArray: {
4851 value = Klass::_lh_array_tag_ref_value;
4852 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4853 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4854 break;
4855 }
4856 case TypeArray: {
4857 value = Klass::_lh_array_tag_type_value;
4858 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4859 btest = BoolTest::eq;
4860 break;
4861 }
4862 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4863 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4864 default:
4865 ShouldNotReachHere();
4866 }
4867 // Now test the correct condition.
4868 jint nval = (jint)value;
4869 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4870 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4871 Node* ctrl = generate_fair_guard(bol, region);
4872 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4873 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4874 // Keep track of the fact that 'obj' is an array to prevent
4875 // array specific accesses from floating above the guard.
4876 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4877 }
4878 return ctrl;
4879 }
4880
4881 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4882 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4883 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4884 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4885 assert(null_free || atomic, "nullable implies atomic");
4886 Node* componentType = argument(0);
4887 Node* length = argument(1);
4888 Node* init_val = null_free ? argument(2) : nullptr;
4889
4890 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4891 if (tp != nullptr) {
4892 ciInstanceKlass* ik = tp->instance_klass();
4893 if (ik == C->env()->Class_klass()) {
4894 ciType* t = tp->java_mirror_type();
4895 if (t != nullptr && t->is_inlinetype()) {
4896
4897 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4898 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4899
4900 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4901 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4902 return false;
4903 }
4904
4905 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4906 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces);
4907 if (null_free) {
4908 if (init_val->is_InlineType()) {
4909 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4910 // Zeroing is enough because the init value is the all-zero value
4911 init_val = nullptr;
4912 } else {
4913 init_val = init_val->as_InlineType()->buffer(this);
4914 }
4915 }
4916 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4917 // If we insert a checkcast here, we can be sure that init_val is an InlineTypeNode, so
4918 // when we folded a field load from an allocation (e.g. during escape analysis), we can
4919 // remove the check init_val->is_InlineType().
4920 }
4921 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4922 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4923 assert(arytype->is_null_free() == null_free, "inconsistency");
4924 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4925 set_result(obj);
4926 return true;
4927 }
4928 }
4929 }
4930 }
4931 return false;
4932 }
4933
4934 // public static native boolean ValueClass::isFlatArray(Object array);
4935 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4936 // public static native boolean ValueClass::isAtomicArray(Object array);
4937 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4938 Node* array = argument(0);
4939
4940 Node* bol;
4941 switch(check) {
4942 case IsFlat:
4943 // TODO 8350865 Use the object version here instead of loading the klass
4944 // The problem is that PhaseMacroExpand::expand_flatarraycheck_node can only handle some IR shapes and will fail, for example, if the bol is directly wired to a ReturnNode
4945 bol = flat_array_test(load_object_klass(array));
4946 break;
4947 case IsNullRestricted:
4948 bol = null_free_array_test(array);
4949 break;
4950 case IsAtomic:
4951 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4952 // Enable TestIntrinsics::test87/88 once this is implemented
4953 // bol = null_free_atomic_array_test
4954 return false;
4955 default:
4956 ShouldNotReachHere();
4957 }
4958
4959 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4960 set_result(res);
4961 return true;
4962 }
4963
4964 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4965 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4966 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4967 RegionNode* region = new RegionNode(2);
4968 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4969
4970 if (type_array_guard) {
4971 generate_typeArray_guard(klass_node, region);
4972 if (region->req() == 3) {
4973 phi->add_req(klass_node);
4974 }
4975 }
4976 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4977 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4978
4979 // Can be null if not initialized yet, just deopt
4980 Node* null_ctl = top();
4981 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4982
4983 region->init_req(1, control());
4984 phi->init_req(1, refined_klass);
4985
4986 set_control(_gvn.transform(region));
4987 return _gvn.transform(phi);
4988 }
4989
4990 // Load the non-refined array klass from an ObjArrayKlass.
4991 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4992 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4993 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4994 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4995 }
4996
4997 RegionNode* region = new RegionNode(2);
4998 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4999
5000 generate_typeArray_guard(klass_node, region);
5001 if (region->req() == 3) {
5002 phi->add_req(klass_node);
5003 }
5004 Node* super_adr = basic_plus_adr(klass_node, in_bytes(Klass::super_offset()));
5005 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
5006
5007 region->init_req(1, control());
5008 phi->init_req(1, super_klass);
5009
5010 set_control(_gvn.transform(region));
5011 return _gvn.transform(phi);
5012 }
5013
5014 //-----------------------inline_native_newArray--------------------------
5015 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5016 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5017 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5018 Node* mirror;
5019 Node* count_val;
5020 if (uninitialized) {
5021 null_check_receiver();
5022 mirror = argument(1);
5023 count_val = argument(2);
5024 } else {
5025 mirror = argument(0);
5026 count_val = argument(1);
5027 }
5028
5029 mirror = null_check(mirror);
5030 // If mirror or obj is dead, only null-path is taken.
5031 if (stopped()) return true;
5032
5033 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5034 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5035 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5036 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5037 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5038
5039 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5040 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5041 result_reg, _slow_path);
5042 Node* normal_ctl = control();
5043 Node* no_array_ctl = result_reg->in(_slow_path);
5044
5045 // Generate code for the slow case. We make a call to newArray().
5046 set_control(no_array_ctl);
5047 if (!stopped()) {
5048 // Either the input type is void.class, or else the
5049 // array klass has not yet been cached. Either the
5050 // ensuing call will throw an exception, or else it
5051 // will cache the array klass for next time.
5052 PreserveJVMState pjvms(this);
5053 CallJavaNode* slow_call = nullptr;
5054 if (uninitialized) {
5055 // Generate optimized virtual call (holder class 'Unsafe' is final)
5056 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5057 } else {
5058 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5059 }
5060 Node* slow_result = set_results_for_java_call(slow_call);
5061 // this->control() comes from set_results_for_java_call
5062 result_reg->set_req(_slow_path, control());
5063 result_val->set_req(_slow_path, slow_result);
5064 result_io ->set_req(_slow_path, i_o());
5065 result_mem->set_req(_slow_path, reset_memory());
5066 }
5067
5068 set_control(normal_ctl);
5069 if (!stopped()) {
5070 // Normal case: The array type has been cached in the java.lang.Class.
5071 // The following call works fine even if the array type is polymorphic.
5072 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5073
5074 klass_node = load_default_refined_array_klass(klass_node);
5075
5076 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5077 result_reg->init_req(_normal_path, control());
5078 result_val->init_req(_normal_path, obj);
5079 result_io ->init_req(_normal_path, i_o());
5080 result_mem->init_req(_normal_path, reset_memory());
5081
5082 if (uninitialized) {
5083 // Mark the allocation so that zeroing is skipped
5084 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5085 alloc->maybe_set_complete(&_gvn);
5086 }
5087 }
5088
5089 // Return the combined state.
5090 set_i_o( _gvn.transform(result_io) );
5091 set_all_memory( _gvn.transform(result_mem));
5092
5093 C->set_has_split_ifs(true); // Has chance for split-if optimization
5094 set_result(result_reg, result_val);
5095 return true;
5096 }
5097
5098 //----------------------inline_native_getLength--------------------------
5099 // public static native int java.lang.reflect.Array.getLength(Object array);
5100 bool LibraryCallKit::inline_native_getLength() {
5101 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5102
5103 Node* array = null_check(argument(0));
5104 // If array is dead, only null-path is taken.
5105 if (stopped()) return true;
5106
5107 // Deoptimize if it is a non-array.
5108 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5109
5110 if (non_array != nullptr) {
5111 PreserveJVMState pjvms(this);
5112 set_control(non_array);
5113 uncommon_trap(Deoptimization::Reason_intrinsic,
5114 Deoptimization::Action_maybe_recompile);
5115 }
5116
5117 // If control is dead, only non-array-path is taken.
5118 if (stopped()) return true;
5119
5120 // The works fine even if the array type is polymorphic.
5121 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5122 Node* result = load_array_length(array);
5123
5124 C->set_has_split_ifs(true); // Has chance for split-if optimization
5125 set_result(result);
5126 return true;
5127 }
5128
5129 //------------------------inline_array_copyOf----------------------------
5130 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5131 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5132 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5133 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5134
5135 // Get the arguments.
5136 Node* original = argument(0);
5137 Node* start = is_copyOfRange? argument(1): intcon(0);
5138 Node* end = is_copyOfRange? argument(2): argument(1);
5139 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5140
5141 Node* newcopy = nullptr;
5142
5143 // Set the original stack and the reexecute bit for the interpreter to reexecute
5144 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5145 { PreserveReexecuteState preexecs(this);
5146 jvms()->set_should_reexecute(true);
5147
5148 array_type_mirror = null_check(array_type_mirror);
5149 original = null_check(original);
5150
5151 // Check if a null path was taken unconditionally.
5152 if (stopped()) return true;
5153
5154 Node* orig_length = load_array_length(original);
5155
5156 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5157 klass_node = null_check(klass_node);
5158
5159 RegionNode* bailout = new RegionNode(1);
5160 record_for_igvn(bailout);
5161
5162 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5163 // Bail out if that is so.
5164 // Inline type array may have object field that would require a
5165 // write barrier. Conservatively, go to slow path.
5166 // TODO 8251971: Optimize for the case when flat src/dst are later found
5167 // to not contain oops (i.e., move this check to the macro expansion phase).
5168 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5169 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5170 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5171 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5172 // Can src array be flat and contain oops?
5173 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5174 // Can dest array be flat and contain oops?
5175 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5176 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5177
5178 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5179
5180 if (not_objArray != nullptr) {
5181 // Improve the klass node's type from the new optimistic assumption:
5182 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5183 bool not_flat = !UseArrayFlattening;
5184 bool not_null_free = !Arguments::is_valhalla_enabled();
5185 const Type* akls = TypeAryKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0), Type::trust_interfaces, not_flat, not_null_free, false, false, not_flat, true);
5186 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5187 refined_klass_node = _gvn.transform(cast);
5188 }
5189
5190 // Bail out if either start or end is negative.
5191 generate_negative_guard(start, bailout, &start);
5192 generate_negative_guard(end, bailout, &end);
5193
5194 Node* length = end;
5195 if (_gvn.type(start) != TypeInt::ZERO) {
5196 length = _gvn.transform(new SubINode(end, start));
5197 }
5198
5199 // Bail out if length is negative (i.e., if start > end).
5200 // Without this the new_array would throw
5201 // NegativeArraySizeException but IllegalArgumentException is what
5202 // should be thrown
5203 generate_negative_guard(length, bailout, &length);
5204
5205 // Handle inline type arrays
5206 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5207 if (!stopped()) {
5208 // TODO 8251971
5209 if (!orig_t->is_null_free()) {
5210 // Not statically known to be null free, add a check
5211 generate_fair_guard(null_free_array_test(original), bailout);
5212 }
5213 orig_t = _gvn.type(original)->isa_aryptr();
5214 if (orig_t != nullptr && orig_t->is_flat()) {
5215 // Src is flat, check that dest is flat as well
5216 if (exclude_flat) {
5217 // Dest can't be flat, bail out
5218 bailout->add_req(control());
5219 set_control(top());
5220 } else {
5221 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5222 }
5223 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5224 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5225 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5226 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5227 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5228 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5229 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5230 if (orig_t != nullptr) {
5231 orig_t = orig_t->cast_to_not_flat();
5232 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5233 }
5234 }
5235 if (!can_validate) {
5236 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5237 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5238 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5239 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5240 generate_fair_guard(null_free_array_test(original), bailout);
5241 }
5242 }
5243
5244 // Bail out if start is larger than the original length
5245 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5246 generate_negative_guard(orig_tail, bailout, &orig_tail);
5247
5248 if (bailout->req() > 1) {
5249 PreserveJVMState pjvms(this);
5250 set_control(_gvn.transform(bailout));
5251 uncommon_trap(Deoptimization::Reason_intrinsic,
5252 Deoptimization::Action_maybe_recompile);
5253 }
5254
5255 if (!stopped()) {
5256 // How many elements will we copy from the original?
5257 // The answer is MinI(orig_tail, length).
5258 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5259
5260 // Generate a direct call to the right arraycopy function(s).
5261 // We know the copy is disjoint but we might not know if the
5262 // oop stores need checking.
5263 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5264 // This will fail a store-check if x contains any non-nulls.
5265
5266 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5267 // loads/stores but it is legal only if we're sure the
5268 // Arrays.copyOf would succeed. So we need all input arguments
5269 // to the copyOf to be validated, including that the copy to the
5270 // new array won't trigger an ArrayStoreException. That subtype
5271 // check can be optimized if we know something on the type of
5272 // the input array from type speculation.
5273 if (_gvn.type(klass_node)->singleton()) {
5274 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5275 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5276
5277 int test = C->static_subtype_check(superk, subk);
5278 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5279 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5280 if (t_original->speculative_type() != nullptr) {
5281 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5282 }
5283 }
5284 }
5285
5286 bool validated = false;
5287 // Reason_class_check rather than Reason_intrinsic because we
5288 // want to intrinsify even if this traps.
5289 if (can_validate) {
5290 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5291
5292 if (not_subtype_ctrl != top()) {
5293 PreserveJVMState pjvms(this);
5294 set_control(not_subtype_ctrl);
5295 uncommon_trap(Deoptimization::Reason_class_check,
5296 Deoptimization::Action_make_not_entrant);
5297 assert(stopped(), "Should be stopped");
5298 }
5299 validated = true;
5300 }
5301
5302 if (!stopped()) {
5303 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5304
5305 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5306 load_object_klass(original), klass_node);
5307 if (!is_copyOfRange) {
5308 ac->set_copyof(validated);
5309 } else {
5310 ac->set_copyofrange(validated);
5311 }
5312 Node* n = _gvn.transform(ac);
5313 if (n == ac) {
5314 ac->connect_outputs(this);
5315 } else {
5316 assert(validated, "shouldn't transform if all arguments not validated");
5317 set_all_memory(n);
5318 }
5319 }
5320 }
5321 } // original reexecute is set back here
5322
5323 C->set_has_split_ifs(true); // Has chance for split-if optimization
5324 if (!stopped()) {
5325 set_result(newcopy);
5326 }
5327 return true;
5328 }
5329
5330
5331 //----------------------generate_virtual_guard---------------------------
5332 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5333 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5334 RegionNode* slow_region) {
5335 ciMethod* method = callee();
5336 int vtable_index = method->vtable_index();
5337 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5338 "bad index %d", vtable_index);
5339 // Get the Method* out of the appropriate vtable entry.
5340 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5341 vtable_index*vtableEntry::size_in_bytes() +
5342 in_bytes(vtableEntry::method_offset());
5343 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5344 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5345
5346 // Compare the target method with the expected method (e.g., Object.hashCode).
5347 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5348
5349 Node* native_call = makecon(native_call_addr);
5350 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5351 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5352
5353 return generate_slow_guard(test_native, slow_region);
5354 }
5355
5356 //-----------------------generate_method_call----------------------------
5357 // Use generate_method_call to make a slow-call to the real
5358 // method if the fast path fails. An alternative would be to
5359 // use a stub like OptoRuntime::slow_arraycopy_Java.
5360 // This only works for expanding the current library call,
5361 // not another intrinsic. (E.g., don't use this for making an
5362 // arraycopy call inside of the copyOf intrinsic.)
5363 CallJavaNode*
5364 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5365 // When compiling the intrinsic method itself, do not use this technique.
5366 guarantee(callee() != C->method(), "cannot make slow-call to self");
5367
5368 ciMethod* method = callee();
5369 // ensure the JVMS we have will be correct for this call
5370 guarantee(method_id == method->intrinsic_id(), "must match");
5371
5372 const TypeFunc* tf = TypeFunc::make(method);
5373 if (res_not_null) {
5374 assert(tf->return_type() == T_OBJECT, "");
5375 const TypeTuple* range = tf->range_cc();
5376 const Type** fields = TypeTuple::fields(range->cnt());
5377 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5378 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5379 tf = TypeFunc::make(tf->domain_cc(), new_range);
5380 }
5381 CallJavaNode* slow_call;
5382 if (is_static) {
5383 assert(!is_virtual, "");
5384 slow_call = new CallStaticJavaNode(C, tf,
5385 SharedRuntime::get_resolve_static_call_stub(), method);
5386 } else if (is_virtual) {
5387 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5388 int vtable_index = Method::invalid_vtable_index;
5389 if (UseInlineCaches) {
5390 // Suppress the vtable call
5391 } else {
5392 // hashCode and clone are not a miranda methods,
5393 // so the vtable index is fixed.
5394 // No need to use the linkResolver to get it.
5395 vtable_index = method->vtable_index();
5396 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5397 "bad index %d", vtable_index);
5398 }
5399 slow_call = new CallDynamicJavaNode(tf,
5400 SharedRuntime::get_resolve_virtual_call_stub(),
5401 method, vtable_index);
5402 } else { // neither virtual nor static: opt_virtual
5403 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5404 slow_call = new CallStaticJavaNode(C, tf,
5405 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5406 slow_call->set_optimized_virtual(true);
5407 }
5408 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5409 // To be able to issue a direct call (optimized virtual or virtual)
5410 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5411 // about the method being invoked should be attached to the call site to
5412 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5413 slow_call->set_override_symbolic_info(true);
5414 }
5415 set_arguments_for_java_call(slow_call);
5416 set_edges_for_java_call(slow_call);
5417 return slow_call;
5418 }
5419
5420
5421 /**
5422 * Build special case code for calls to hashCode on an object. This call may
5423 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5424 * slightly different code.
5425 */
5426 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5427 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5428 assert(!(is_virtual && is_static), "either virtual, special, or static");
5429
5430 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5431
5432 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5433 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5434 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5435 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5436 Node* obj = argument(0);
5437
5438 // Don't intrinsify hashcode on inline types for now.
5439 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5440 if (gvn().type(obj)->is_inlinetypeptr()) {
5441 return false;
5442 }
5443
5444 if (!is_static) {
5445 // Check for hashing null object
5446 obj = null_check_receiver();
5447 if (stopped()) return true; // unconditionally null
5448 result_reg->init_req(_null_path, top());
5449 result_val->init_req(_null_path, top());
5450 } else {
5451 // Do a null check, and return zero if null.
5452 // System.identityHashCode(null) == 0
5453 Node* null_ctl = top();
5454 obj = null_check_oop(obj, &null_ctl);
5455 result_reg->init_req(_null_path, null_ctl);
5456 result_val->init_req(_null_path, _gvn.intcon(0));
5457 }
5458
5459 // Unconditionally null? Then return right away.
5460 if (stopped()) {
5461 set_control( result_reg->in(_null_path));
5462 if (!stopped())
5463 set_result(result_val->in(_null_path));
5464 return true;
5465 }
5466
5467 // We only go to the fast case code if we pass a number of guards. The
5468 // paths which do not pass are accumulated in the slow_region.
5469 RegionNode* slow_region = new RegionNode(1);
5470 record_for_igvn(slow_region);
5471
5472 // If this is a virtual call, we generate a funny guard. We pull out
5473 // the vtable entry corresponding to hashCode() from the target object.
5474 // If the target method which we are calling happens to be the native
5475 // Object hashCode() method, we pass the guard. We do not need this
5476 // guard for non-virtual calls -- the caller is known to be the native
5477 // Object hashCode().
5478 if (is_virtual) {
5479 // After null check, get the object's klass.
5480 Node* obj_klass = load_object_klass(obj);
5481 generate_virtual_guard(obj_klass, slow_region);
5482 }
5483
5484 // Get the header out of the object, use LoadMarkNode when available
5485 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5486 // The control of the load must be null. Otherwise, the load can move before
5487 // the null check after castPP removal.
5488 Node* no_ctrl = nullptr;
5489 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5490
5491 if (!UseObjectMonitorTable) {
5492 // Test the header to see if it is safe to read w.r.t. locking.
5493 // We cannot use the inline type mask as this may check bits that are overriden
5494 // by an object monitor's pointer when inflating locking.
5495 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5496 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5497 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5498 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5499 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5500
5501 generate_slow_guard(test_monitor, slow_region);
5502 }
5503
5504 // Get the hash value and check to see that it has been properly assigned.
5505 // We depend on hash_mask being at most 32 bits and avoid the use of
5506 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5507 // vm: see markWord.hpp.
5508 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5509 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5510 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5511 // This hack lets the hash bits live anywhere in the mark object now, as long
5512 // as the shift drops the relevant bits into the low 32 bits. Note that
5513 // Java spec says that HashCode is an int so there's no point in capturing
5514 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5515 hshifted_header = ConvX2I(hshifted_header);
5516 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5517
5518 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5519 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5520 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5521
5522 generate_slow_guard(test_assigned, slow_region);
5523
5524 Node* init_mem = reset_memory();
5525 // fill in the rest of the null path:
5526 result_io ->init_req(_null_path, i_o());
5527 result_mem->init_req(_null_path, init_mem);
5528
5529 result_val->init_req(_fast_path, hash_val);
5530 result_reg->init_req(_fast_path, control());
5531 result_io ->init_req(_fast_path, i_o());
5532 result_mem->init_req(_fast_path, init_mem);
5533
5534 // Generate code for the slow case. We make a call to hashCode().
5535 set_control(_gvn.transform(slow_region));
5536 if (!stopped()) {
5537 // No need for PreserveJVMState, because we're using up the present state.
5538 set_all_memory(init_mem);
5539 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5540 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5541 Node* slow_result = set_results_for_java_call(slow_call);
5542 // this->control() comes from set_results_for_java_call
5543 result_reg->init_req(_slow_path, control());
5544 result_val->init_req(_slow_path, slow_result);
5545 result_io ->set_req(_slow_path, i_o());
5546 result_mem ->set_req(_slow_path, reset_memory());
5547 }
5548
5549 // Return the combined state.
5550 set_i_o( _gvn.transform(result_io) );
5551 set_all_memory( _gvn.transform(result_mem));
5552
5553 set_result(result_reg, result_val);
5554 return true;
5555 }
5556
5557 //---------------------------inline_native_getClass----------------------------
5558 // public final native Class<?> java.lang.Object.getClass();
5559 //
5560 // Build special case code for calls to getClass on an object.
5561 bool LibraryCallKit::inline_native_getClass() {
5562 Node* obj = argument(0);
5563 if (obj->is_InlineType()) {
5564 const Type* t = _gvn.type(obj);
5565 if (t->maybe_null()) {
5566 null_check(obj);
5567 }
5568 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5569 return true;
5570 }
5571 obj = null_check_receiver();
5572 if (stopped()) return true;
5573 set_result(load_mirror_from_klass(load_object_klass(obj)));
5574 return true;
5575 }
5576
5577 //-----------------inline_native_Reflection_getCallerClass---------------------
5578 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5579 //
5580 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5581 //
5582 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5583 // in that it must skip particular security frames and checks for
5584 // caller sensitive methods.
5585 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5586 #ifndef PRODUCT
5587 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5588 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5589 }
5590 #endif
5591
5592 if (!jvms()->has_method()) {
5593 #ifndef PRODUCT
5594 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5595 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5596 }
5597 #endif
5598 return false;
5599 }
5600
5601 // Walk back up the JVM state to find the caller at the required
5602 // depth.
5603 JVMState* caller_jvms = jvms();
5604
5605 // Cf. JVM_GetCallerClass
5606 // NOTE: Start the loop at depth 1 because the current JVM state does
5607 // not include the Reflection.getCallerClass() frame.
5608 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5609 ciMethod* m = caller_jvms->method();
5610 switch (n) {
5611 case 0:
5612 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5613 break;
5614 case 1:
5615 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5616 if (!m->caller_sensitive()) {
5617 #ifndef PRODUCT
5618 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5619 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5620 }
5621 #endif
5622 return false; // bail-out; let JVM_GetCallerClass do the work
5623 }
5624 break;
5625 default:
5626 if (!m->is_ignored_by_security_stack_walk()) {
5627 // We have reached the desired frame; return the holder class.
5628 // Acquire method holder as java.lang.Class and push as constant.
5629 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5630 ciInstance* caller_mirror = caller_klass->java_mirror();
5631 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5632
5633 #ifndef PRODUCT
5634 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5635 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());
5636 tty->print_cr(" JVM state at this point:");
5637 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5638 ciMethod* m = jvms()->of_depth(i)->method();
5639 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5640 }
5641 }
5642 #endif
5643 return true;
5644 }
5645 break;
5646 }
5647 }
5648
5649 #ifndef PRODUCT
5650 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5651 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5652 tty->print_cr(" JVM state at this point:");
5653 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5654 ciMethod* m = jvms()->of_depth(i)->method();
5655 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5656 }
5657 }
5658 #endif
5659
5660 return false; // bail-out; let JVM_GetCallerClass do the work
5661 }
5662
5663 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5664 Node* arg = argument(0);
5665 Node* result = nullptr;
5666
5667 switch (id) {
5668 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5669 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5670 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5671 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5672 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5673 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5674
5675 case vmIntrinsics::_doubleToLongBits: {
5676 // two paths (plus control) merge in a wood
5677 RegionNode *r = new RegionNode(3);
5678 Node *phi = new PhiNode(r, TypeLong::LONG);
5679
5680 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5681 // Build the boolean node
5682 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5683
5684 // Branch either way.
5685 // NaN case is less traveled, which makes all the difference.
5686 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5687 Node *opt_isnan = _gvn.transform(ifisnan);
5688 assert( opt_isnan->is_If(), "Expect an IfNode");
5689 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5690 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5691
5692 set_control(iftrue);
5693
5694 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5695 Node *slow_result = longcon(nan_bits); // return NaN
5696 phi->init_req(1, _gvn.transform( slow_result ));
5697 r->init_req(1, iftrue);
5698
5699 // Else fall through
5700 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5701 set_control(iffalse);
5702
5703 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5704 r->init_req(2, iffalse);
5705
5706 // Post merge
5707 set_control(_gvn.transform(r));
5708 record_for_igvn(r);
5709
5710 C->set_has_split_ifs(true); // Has chance for split-if optimization
5711 result = phi;
5712 assert(result->bottom_type()->isa_long(), "must be");
5713 break;
5714 }
5715
5716 case vmIntrinsics::_floatToIntBits: {
5717 // two paths (plus control) merge in a wood
5718 RegionNode *r = new RegionNode(3);
5719 Node *phi = new PhiNode(r, TypeInt::INT);
5720
5721 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5722 // Build the boolean node
5723 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5724
5725 // Branch either way.
5726 // NaN case is less traveled, which makes all the difference.
5727 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5728 Node *opt_isnan = _gvn.transform(ifisnan);
5729 assert( opt_isnan->is_If(), "Expect an IfNode");
5730 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5731 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5732
5733 set_control(iftrue);
5734
5735 static const jint nan_bits = 0x7fc00000;
5736 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5737 phi->init_req(1, _gvn.transform( slow_result ));
5738 r->init_req(1, iftrue);
5739
5740 // Else fall through
5741 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5742 set_control(iffalse);
5743
5744 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5745 r->init_req(2, iffalse);
5746
5747 // Post merge
5748 set_control(_gvn.transform(r));
5749 record_for_igvn(r);
5750
5751 C->set_has_split_ifs(true); // Has chance for split-if optimization
5752 result = phi;
5753 assert(result->bottom_type()->isa_int(), "must be");
5754 break;
5755 }
5756
5757 default:
5758 fatal_unexpected_iid(id);
5759 break;
5760 }
5761 set_result(_gvn.transform(result));
5762 return true;
5763 }
5764
5765 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5766 Node* arg = argument(0);
5767 Node* result = nullptr;
5768
5769 switch (id) {
5770 case vmIntrinsics::_floatIsInfinite:
5771 result = new IsInfiniteFNode(arg);
5772 break;
5773 case vmIntrinsics::_floatIsFinite:
5774 result = new IsFiniteFNode(arg);
5775 break;
5776 case vmIntrinsics::_doubleIsInfinite:
5777 result = new IsInfiniteDNode(arg);
5778 break;
5779 case vmIntrinsics::_doubleIsFinite:
5780 result = new IsFiniteDNode(arg);
5781 break;
5782 default:
5783 fatal_unexpected_iid(id);
5784 break;
5785 }
5786 set_result(_gvn.transform(result));
5787 return true;
5788 }
5789
5790 //----------------------inline_unsafe_copyMemory-------------------------
5791 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5792
5793 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5794 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5795 const Type* base_t = gvn.type(base);
5796
5797 bool in_native = (base_t == TypePtr::NULL_PTR);
5798 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5799 bool is_mixed = !in_heap && !in_native;
5800
5801 if (is_mixed) {
5802 return true; // mixed accesses can touch both on-heap and off-heap memory
5803 }
5804 if (in_heap) {
5805 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5806 if (!is_prim_array) {
5807 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5808 // there's not enough type information available to determine proper memory slice for it.
5809 return true;
5810 }
5811 }
5812 return false;
5813 }
5814
5815 bool LibraryCallKit::inline_unsafe_copyMemory() {
5816 if (callee()->is_static()) return false; // caller must have the capability!
5817 null_check_receiver(); // null-check receiver
5818 if (stopped()) return true;
5819
5820 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5821
5822 Node* src_base = argument(1); // type: oop
5823 Node* src_off = ConvL2X(argument(2)); // type: long
5824 Node* dst_base = argument(4); // type: oop
5825 Node* dst_off = ConvL2X(argument(5)); // type: long
5826 Node* size = ConvL2X(argument(7)); // type: long
5827
5828 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5829 "fieldOffset must be byte-scaled");
5830
5831 Node* src_addr = make_unsafe_address(src_base, src_off);
5832 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5833
5834 Node* thread = _gvn.transform(new ThreadLocalNode());
5835 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5836 BasicType doing_unsafe_access_bt = T_BYTE;
5837 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5838
5839 // update volatile field
5840 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5841
5842 int flags = RC_LEAF | RC_NO_FP;
5843
5844 const TypePtr* dst_type = TypePtr::BOTTOM;
5845
5846 // Adjust memory effects of the runtime call based on input values.
5847 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5848 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5849 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5850
5851 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5852 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5853 flags |= RC_NARROW_MEM; // narrow in memory
5854 }
5855 }
5856
5857 // Call it. Note that the length argument is not scaled.
5858 make_runtime_call(flags,
5859 OptoRuntime::fast_arraycopy_Type(),
5860 StubRoutines::unsafe_arraycopy(),
5861 "unsafe_arraycopy",
5862 dst_type,
5863 src_addr, dst_addr, size XTOP);
5864
5865 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5866
5867 return true;
5868 }
5869
5870 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5871 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5872 bool LibraryCallKit::inline_unsafe_setMemory() {
5873 if (callee()->is_static()) return false; // caller must have the capability!
5874 null_check_receiver(); // null-check receiver
5875 if (stopped()) return true;
5876
5877 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5878
5879 Node* dst_base = argument(1); // type: oop
5880 Node* dst_off = ConvL2X(argument(2)); // type: long
5881 Node* size = ConvL2X(argument(4)); // type: long
5882 Node* byte = argument(6); // type: byte
5883
5884 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5885 "fieldOffset must be byte-scaled");
5886
5887 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5888
5889 Node* thread = _gvn.transform(new ThreadLocalNode());
5890 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5891 BasicType doing_unsafe_access_bt = T_BYTE;
5892 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5893
5894 // update volatile field
5895 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5896
5897 int flags = RC_LEAF | RC_NO_FP;
5898
5899 const TypePtr* dst_type = TypePtr::BOTTOM;
5900
5901 // Adjust memory effects of the runtime call based on input values.
5902 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5903 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5904
5905 flags |= RC_NARROW_MEM; // narrow in memory
5906 }
5907
5908 // Call it. Note that the length argument is not scaled.
5909 make_runtime_call(flags,
5910 OptoRuntime::unsafe_setmemory_Type(),
5911 StubRoutines::unsafe_setmemory(),
5912 "unsafe_setmemory",
5913 dst_type,
5914 dst_addr, size XTOP, byte);
5915
5916 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5917
5918 return true;
5919 }
5920
5921 #undef XTOP
5922
5923 //------------------------clone_coping-----------------------------------
5924 // Helper function for inline_native_clone.
5925 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5926 assert(obj_size != nullptr, "");
5927 Node* raw_obj = alloc_obj->in(1);
5928 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5929
5930 AllocateNode* alloc = nullptr;
5931 if (ReduceBulkZeroing &&
5932 // If we are implementing an array clone without knowing its source type
5933 // (can happen when compiling the array-guarded branch of a reflective
5934 // Object.clone() invocation), initialize the array within the allocation.
5935 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5936 // to a runtime clone call that assumes fully initialized source arrays.
5937 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5938 // We will be completely responsible for initializing this object -
5939 // mark Initialize node as complete.
5940 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5941 // The object was just allocated - there should be no any stores!
5942 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5943 // Mark as complete_with_arraycopy so that on AllocateNode
5944 // expansion, we know this AllocateNode is initialized by an array
5945 // copy and a StoreStore barrier exists after the array copy.
5946 alloc->initialization()->set_complete_with_arraycopy();
5947 }
5948
5949 Node* size = _gvn.transform(obj_size);
5950 access_clone(obj, alloc_obj, size, is_array);
5951
5952 // Do not let reads from the cloned object float above the arraycopy.
5953 if (alloc != nullptr) {
5954 // Do not let stores that initialize this object be reordered with
5955 // a subsequent store that would make this object accessible by
5956 // other threads.
5957 // Record what AllocateNode this StoreStore protects so that
5958 // escape analysis can go from the MemBarStoreStoreNode to the
5959 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5960 // based on the escape status of the AllocateNode.
5961 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5962 } else {
5963 insert_mem_bar(Op_MemBarCPUOrder);
5964 }
5965 }
5966
5967 //------------------------inline_native_clone----------------------------
5968 // protected native Object java.lang.Object.clone();
5969 //
5970 // Here are the simple edge cases:
5971 // null receiver => normal trap
5972 // virtual and clone was overridden => slow path to out-of-line clone
5973 // not cloneable or finalizer => slow path to out-of-line Object.clone
5974 //
5975 // The general case has two steps, allocation and copying.
5976 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5977 //
5978 // Copying also has two cases, oop arrays and everything else.
5979 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5980 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5981 //
5982 // These steps fold up nicely if and when the cloned object's klass
5983 // can be sharply typed as an object array, a type array, or an instance.
5984 //
5985 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5986 PhiNode* result_val;
5987
5988 // Set the reexecute bit for the interpreter to reexecute
5989 // the bytecode that invokes Object.clone if deoptimization happens.
5990 { PreserveReexecuteState preexecs(this);
5991 jvms()->set_should_reexecute(true);
5992
5993 Node* obj = argument(0);
5994 obj = null_check_receiver();
5995 if (stopped()) return true;
5996
5997 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5998 if (obj_type->is_inlinetypeptr()) {
5999 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
6000 // no identity.
6001 set_result(obj);
6002 return true;
6003 }
6004
6005 // If we are going to clone an instance, we need its exact type to
6006 // know the number and types of fields to convert the clone to
6007 // loads/stores. Maybe a speculative type can help us.
6008 if (!obj_type->klass_is_exact() &&
6009 obj_type->speculative_type() != nullptr &&
6010 obj_type->speculative_type()->is_instance_klass() &&
6011 !obj_type->speculative_type()->is_inlinetype()) {
6012 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
6013 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
6014 !spec_ik->has_injected_fields()) {
6015 if (!obj_type->isa_instptr() ||
6016 obj_type->is_instptr()->instance_klass()->has_subklass()) {
6017 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6018 }
6019 }
6020 }
6021
6022 // Conservatively insert a memory barrier on all memory slices.
6023 // Do not let writes into the original float below the clone.
6024 insert_mem_bar(Op_MemBarCPUOrder);
6025
6026 // paths into result_reg:
6027 enum {
6028 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6029 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6030 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6031 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6032 PATH_LIMIT
6033 };
6034 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6035 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6036 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6037 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6038 record_for_igvn(result_reg);
6039
6040 Node* obj_klass = load_object_klass(obj);
6041 // We only go to the fast case code if we pass a number of guards.
6042 // The paths which do not pass are accumulated in the slow_region.
6043 RegionNode* slow_region = new RegionNode(1);
6044 record_for_igvn(slow_region);
6045
6046 Node* array_obj = obj;
6047 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6048 if (array_ctl != nullptr) {
6049 // It's an array.
6050 PreserveJVMState pjvms(this);
6051 set_control(array_ctl);
6052
6053 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6054 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6055 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6056 obj_type->can_be_inline_array() &&
6057 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6058 // Flat inline type array may have object field that would require a
6059 // write barrier. Conservatively, go to slow path.
6060 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6061 }
6062
6063 if (!stopped()) {
6064 Node* obj_length = load_array_length(array_obj);
6065 Node* array_size = nullptr; // Size of the array without object alignment padding.
6066 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6067
6068 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6069 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6070 // If it is an oop array, it requires very special treatment,
6071 // because gc barriers are required when accessing the array.
6072 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6073 if (is_obja != nullptr) {
6074 PreserveJVMState pjvms2(this);
6075 set_control(is_obja);
6076 // Generate a direct call to the right arraycopy function(s).
6077 // Clones are always tightly coupled.
6078 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6079 ac->set_clone_oop_array();
6080 Node* n = _gvn.transform(ac);
6081 assert(n == ac, "cannot disappear");
6082 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6083
6084 result_reg->init_req(_objArray_path, control());
6085 result_val->init_req(_objArray_path, alloc_obj);
6086 result_i_o ->set_req(_objArray_path, i_o());
6087 result_mem ->set_req(_objArray_path, reset_memory());
6088 }
6089 }
6090 // Otherwise, there are no barriers to worry about.
6091 // (We can dispense with card marks if we know the allocation
6092 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6093 // causes the non-eden paths to take compensating steps to
6094 // simulate a fresh allocation, so that no further
6095 // card marks are required in compiled code to initialize
6096 // the object.)
6097
6098 if (!stopped()) {
6099 copy_to_clone(obj, alloc_obj, array_size, true);
6100
6101 // Present the results of the copy.
6102 result_reg->init_req(_array_path, control());
6103 result_val->init_req(_array_path, alloc_obj);
6104 result_i_o ->set_req(_array_path, i_o());
6105 result_mem ->set_req(_array_path, reset_memory());
6106 }
6107 }
6108 }
6109
6110 if (!stopped()) {
6111 // It's an instance (we did array above). Make the slow-path tests.
6112 // If this is a virtual call, we generate a funny guard. We grab
6113 // the vtable entry corresponding to clone() from the target object.
6114 // If the target method which we are calling happens to be the
6115 // Object clone() method, we pass the guard. We do not need this
6116 // guard for non-virtual calls; the caller is known to be the native
6117 // Object clone().
6118 if (is_virtual) {
6119 generate_virtual_guard(obj_klass, slow_region);
6120 }
6121
6122 // The object must be easily cloneable and must not have a finalizer.
6123 // Both of these conditions may be checked in a single test.
6124 // We could optimize the test further, but we don't care.
6125 generate_misc_flags_guard(obj_klass,
6126 // Test both conditions:
6127 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6128 // Must be cloneable but not finalizer:
6129 KlassFlags::_misc_is_cloneable_fast,
6130 slow_region);
6131 }
6132
6133 if (!stopped()) {
6134 // It's an instance, and it passed the slow-path tests.
6135 PreserveJVMState pjvms(this);
6136 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6137 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6138 // is reexecuted if deoptimization occurs and there could be problems when merging
6139 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6140 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6141
6142 copy_to_clone(obj, alloc_obj, obj_size, false);
6143
6144 // Present the results of the slow call.
6145 result_reg->init_req(_instance_path, control());
6146 result_val->init_req(_instance_path, alloc_obj);
6147 result_i_o ->set_req(_instance_path, i_o());
6148 result_mem ->set_req(_instance_path, reset_memory());
6149 }
6150
6151 // Generate code for the slow case. We make a call to clone().
6152 set_control(_gvn.transform(slow_region));
6153 if (!stopped()) {
6154 PreserveJVMState pjvms(this);
6155 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6156 // We need to deoptimize on exception (see comment above)
6157 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6158 // this->control() comes from set_results_for_java_call
6159 result_reg->init_req(_slow_path, control());
6160 result_val->init_req(_slow_path, slow_result);
6161 result_i_o ->set_req(_slow_path, i_o());
6162 result_mem ->set_req(_slow_path, reset_memory());
6163 }
6164
6165 // Return the combined state.
6166 set_control( _gvn.transform(result_reg));
6167 set_i_o( _gvn.transform(result_i_o));
6168 set_all_memory( _gvn.transform(result_mem));
6169 } // original reexecute is set back here
6170
6171 set_result(_gvn.transform(result_val));
6172 return true;
6173 }
6174
6175 // If we have a tightly coupled allocation, the arraycopy may take care
6176 // of the array initialization. If one of the guards we insert between
6177 // the allocation and the arraycopy causes a deoptimization, an
6178 // uninitialized array will escape the compiled method. To prevent that
6179 // we set the JVM state for uncommon traps between the allocation and
6180 // the arraycopy to the state before the allocation so, in case of
6181 // deoptimization, we'll reexecute the allocation and the
6182 // initialization.
6183 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6184 if (alloc != nullptr) {
6185 ciMethod* trap_method = alloc->jvms()->method();
6186 int trap_bci = alloc->jvms()->bci();
6187
6188 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6189 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6190 // Make sure there's no store between the allocation and the
6191 // arraycopy otherwise visible side effects could be rexecuted
6192 // in case of deoptimization and cause incorrect execution.
6193 bool no_interfering_store = true;
6194 Node* mem = alloc->in(TypeFunc::Memory);
6195 if (mem->is_MergeMem()) {
6196 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6197 Node* n = mms.memory();
6198 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6199 assert(n->is_Store(), "what else?");
6200 no_interfering_store = false;
6201 break;
6202 }
6203 }
6204 } else {
6205 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6206 Node* n = mms.memory();
6207 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6208 assert(n->is_Store(), "what else?");
6209 no_interfering_store = false;
6210 break;
6211 }
6212 }
6213 }
6214
6215 if (no_interfering_store) {
6216 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6217
6218 JVMState* saved_jvms = jvms();
6219 saved_reexecute_sp = _reexecute_sp;
6220
6221 set_jvms(sfpt->jvms());
6222 _reexecute_sp = jvms()->sp();
6223
6224 return saved_jvms;
6225 }
6226 }
6227 }
6228 return nullptr;
6229 }
6230
6231 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6232 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6233 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6234 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6235 uint size = alloc->req();
6236 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6237 old_jvms->set_map(sfpt);
6238 for (uint i = 0; i < size; i++) {
6239 sfpt->init_req(i, alloc->in(i));
6240 }
6241 int adjustment = 1;
6242 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6243 if (ary_klass_ptr->is_null_free()) {
6244 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6245 // also requires the componentType and initVal on stack for re-execution.
6246 // Re-create and push the componentType.
6247 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6248 ciInstance* instance = klass->component_mirror_instance();
6249 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6250 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6251 adjustment++;
6252 }
6253 // re-push array length for deoptimization
6254 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6255 if (ary_klass_ptr->is_null_free()) {
6256 // Re-create and push the initVal.
6257 Node* init_val = alloc->in(AllocateNode::InitValue);
6258 if (init_val == nullptr) {
6259 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6260 } else if (UseCompressedOops) {
6261 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6262 }
6263 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6264 adjustment++;
6265 }
6266 old_jvms->set_sp(old_jvms->sp() + adjustment);
6267 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6268 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6269 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6270 old_jvms->set_should_reexecute(true);
6271
6272 sfpt->set_i_o(map()->i_o());
6273 sfpt->set_memory(map()->memory());
6274 sfpt->set_control(map()->control());
6275 return sfpt;
6276 }
6277
6278 // In case of a deoptimization, we restart execution at the
6279 // allocation, allocating a new array. We would leave an uninitialized
6280 // array in the heap that GCs wouldn't expect. Move the allocation
6281 // after the traps so we don't allocate the array if we
6282 // deoptimize. This is possible because tightly_coupled_allocation()
6283 // guarantees there's no observer of the allocated array at this point
6284 // and the control flow is simple enough.
6285 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6286 int saved_reexecute_sp, uint new_idx) {
6287 if (saved_jvms_before_guards != nullptr && !stopped()) {
6288 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6289
6290 assert(alloc != nullptr, "only with a tightly coupled allocation");
6291 // restore JVM state to the state at the arraycopy
6292 saved_jvms_before_guards->map()->set_control(map()->control());
6293 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6294 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6295 // If we've improved the types of some nodes (null check) while
6296 // emitting the guards, propagate them to the current state
6297 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6298 set_jvms(saved_jvms_before_guards);
6299 _reexecute_sp = saved_reexecute_sp;
6300
6301 // Remove the allocation from above the guards
6302 CallProjections* callprojs = alloc->extract_projections(true);
6303 InitializeNode* init = alloc->initialization();
6304 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6305 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6306 init->replace_mem_projs_by(alloc_mem, C);
6307
6308 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6309 // the allocation (i.e. is only valid if the allocation succeeds):
6310 // 1) replace CastIINode with AllocateArrayNode's length here
6311 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6312 //
6313 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6314 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6315 Node* init_control = init->proj_out(TypeFunc::Control);
6316 Node* alloc_length = alloc->Ideal_length();
6317 #ifdef ASSERT
6318 Node* prev_cast = nullptr;
6319 #endif
6320 for (uint i = 0; i < init_control->outcnt(); i++) {
6321 Node* init_out = init_control->raw_out(i);
6322 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6323 #ifdef ASSERT
6324 if (prev_cast == nullptr) {
6325 prev_cast = init_out;
6326 } else {
6327 if (prev_cast->cmp(*init_out) == false) {
6328 prev_cast->dump();
6329 init_out->dump();
6330 assert(false, "not equal CastIINode");
6331 }
6332 }
6333 #endif
6334 C->gvn_replace_by(init_out, alloc_length);
6335 }
6336 }
6337 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6338
6339 // move the allocation here (after the guards)
6340 _gvn.hash_delete(alloc);
6341 alloc->set_req(TypeFunc::Control, control());
6342 alloc->set_req(TypeFunc::I_O, i_o());
6343 Node *mem = reset_memory();
6344 set_all_memory(mem);
6345 alloc->set_req(TypeFunc::Memory, mem);
6346 set_control(init->proj_out_or_null(TypeFunc::Control));
6347 set_i_o(callprojs->fallthrough_ioproj);
6348
6349 // Update memory as done in GraphKit::set_output_for_allocation()
6350 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6351 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6352 if (ary_type->isa_aryptr() && length_type != nullptr) {
6353 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6354 }
6355 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6356 int elemidx = C->get_alias_index(telemref);
6357 // Need to properly move every memory projection for the Initialize
6358 #ifdef ASSERT
6359 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6360 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6361 #endif
6362 auto move_proj = [&](ProjNode* proj) {
6363 int alias_idx = C->get_alias_index(proj->adr_type());
6364 assert(alias_idx == Compile::AliasIdxRaw ||
6365 alias_idx == elemidx ||
6366 alias_idx == mark_idx ||
6367 alias_idx == klass_idx, "should be raw memory or array element type");
6368 set_memory(proj, alias_idx);
6369 };
6370 init->for_each_proj(move_proj, TypeFunc::Memory);
6371
6372 Node* allocx = _gvn.transform(alloc);
6373 assert(allocx == alloc, "where has the allocation gone?");
6374 assert(dest->is_CheckCastPP(), "not an allocation result?");
6375
6376 _gvn.hash_delete(dest);
6377 dest->set_req(0, control());
6378 Node* destx = _gvn.transform(dest);
6379 assert(destx == dest, "where has the allocation result gone?");
6380
6381 array_ideal_length(alloc, ary_type, true);
6382 }
6383 }
6384
6385 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6386 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6387 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6388 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6389 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6390 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6391 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6392 JVMState* saved_jvms_before_guards) {
6393 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6394 // There is at least one unrelated uncommon trap which needs to be replaced.
6395 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6396
6397 JVMState* saved_jvms = jvms();
6398 const int saved_reexecute_sp = _reexecute_sp;
6399 set_jvms(sfpt->jvms());
6400 _reexecute_sp = jvms()->sp();
6401
6402 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6403
6404 // Restore state
6405 set_jvms(saved_jvms);
6406 _reexecute_sp = saved_reexecute_sp;
6407 }
6408 }
6409
6410 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6411 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6412 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6413 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6414 while (if_proj->is_IfProj()) {
6415 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6416 if (uncommon_trap != nullptr) {
6417 create_new_uncommon_trap(uncommon_trap);
6418 }
6419 assert(if_proj->in(0)->is_If(), "must be If");
6420 if_proj = if_proj->in(0)->in(0);
6421 }
6422 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6423 "must have reached control projection of init node");
6424 }
6425
6426 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6427 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6428 assert(trap_request != 0, "no valid UCT trap request");
6429 PreserveJVMState pjvms(this);
6430 set_control(uncommon_trap_call->in(0));
6431 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6432 Deoptimization::trap_request_action(trap_request));
6433 assert(stopped(), "Should be stopped");
6434 _gvn.hash_delete(uncommon_trap_call);
6435 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6436 }
6437
6438 // Common checks for array sorting intrinsics arguments.
6439 // Returns `true` if checks passed.
6440 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6441 // check address of the class
6442 if (elementType == nullptr || elementType->is_top()) {
6443 return false; // dead path
6444 }
6445 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6446 if (elem_klass == nullptr) {
6447 return false; // dead path
6448 }
6449 // java_mirror_type() returns non-null for compile-time Class constants only
6450 ciType* elem_type = elem_klass->java_mirror_type();
6451 if (elem_type == nullptr) {
6452 return false;
6453 }
6454 bt = elem_type->basic_type();
6455 // Disable the intrinsic if the CPU does not support SIMD sort
6456 if (!Matcher::supports_simd_sort(bt)) {
6457 return false;
6458 }
6459 // check address of the array
6460 if (obj == nullptr || obj->is_top()) {
6461 return false; // dead path
6462 }
6463 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6464 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6465 return false; // failed input validation
6466 }
6467 return true;
6468 }
6469
6470 //------------------------------inline_array_partition-----------------------
6471 bool LibraryCallKit::inline_array_partition() {
6472 address stubAddr = StubRoutines::select_array_partition_function();
6473 if (stubAddr == nullptr) {
6474 return false; // Intrinsic's stub is not implemented on this platform
6475 }
6476 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6477
6478 // no receiver because it is a static method
6479 Node* elementType = argument(0);
6480 Node* obj = argument(1);
6481 Node* offset = argument(2); // long
6482 Node* fromIndex = argument(4);
6483 Node* toIndex = argument(5);
6484 Node* indexPivot1 = argument(6);
6485 Node* indexPivot2 = argument(7);
6486 // PartitionOperation: argument(8) is ignored
6487
6488 Node* pivotIndices = nullptr;
6489 BasicType bt = T_ILLEGAL;
6490
6491 if (!check_array_sort_arguments(elementType, obj, bt)) {
6492 return false;
6493 }
6494 null_check(obj);
6495 // If obj is dead, only null-path is taken.
6496 if (stopped()) {
6497 return true;
6498 }
6499 // Set the original stack and the reexecute bit for the interpreter to reexecute
6500 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6501 { PreserveReexecuteState preexecs(this);
6502 jvms()->set_should_reexecute(true);
6503
6504 Node* obj_adr = make_unsafe_address(obj, offset);
6505
6506 // create the pivotIndices array of type int and size = 2
6507 Node* size = intcon(2);
6508 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6509 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6510 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6511 guarantee(alloc != nullptr, "created above");
6512 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6513
6514 // pass the basic type enum to the stub
6515 Node* elemType = intcon(bt);
6516
6517 // Call the stub
6518 const char *stubName = "array_partition_stub";
6519 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6520 stubAddr, stubName, TypePtr::BOTTOM,
6521 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6522 indexPivot1, indexPivot2);
6523
6524 } // original reexecute is set back here
6525
6526 if (!stopped()) {
6527 set_result(pivotIndices);
6528 }
6529
6530 return true;
6531 }
6532
6533
6534 //------------------------------inline_array_sort-----------------------
6535 bool LibraryCallKit::inline_array_sort() {
6536 address stubAddr = StubRoutines::select_arraysort_function();
6537 if (stubAddr == nullptr) {
6538 return false; // Intrinsic's stub is not implemented on this platform
6539 }
6540 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6541
6542 // no receiver because it is a static method
6543 Node* elementType = argument(0);
6544 Node* obj = argument(1);
6545 Node* offset = argument(2); // long
6546 Node* fromIndex = argument(4);
6547 Node* toIndex = argument(5);
6548 // SortOperation: argument(6) is ignored
6549
6550 BasicType bt = T_ILLEGAL;
6551
6552 if (!check_array_sort_arguments(elementType, obj, bt)) {
6553 return false;
6554 }
6555 null_check(obj);
6556 // If obj is dead, only null-path is taken.
6557 if (stopped()) {
6558 return true;
6559 }
6560 Node* obj_adr = make_unsafe_address(obj, offset);
6561
6562 // pass the basic type enum to the stub
6563 Node* elemType = intcon(bt);
6564
6565 // Call the stub.
6566 const char *stubName = "arraysort_stub";
6567 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6568 stubAddr, stubName, TypePtr::BOTTOM,
6569 obj_adr, elemType, fromIndex, toIndex);
6570
6571 return true;
6572 }
6573
6574
6575 //------------------------------inline_arraycopy-----------------------
6576 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6577 // Object dest, int destPos,
6578 // int length);
6579 bool LibraryCallKit::inline_arraycopy() {
6580 // Get the arguments.
6581 Node* src = argument(0); // type: oop
6582 Node* src_offset = argument(1); // type: int
6583 Node* dest = argument(2); // type: oop
6584 Node* dest_offset = argument(3); // type: int
6585 Node* length = argument(4); // type: int
6586
6587 uint new_idx = C->unique();
6588
6589 // Check for allocation before we add nodes that would confuse
6590 // tightly_coupled_allocation()
6591 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6592
6593 int saved_reexecute_sp = -1;
6594 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6595 // See arraycopy_restore_alloc_state() comment
6596 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6597 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6598 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6599 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6600
6601 // The following tests must be performed
6602 // (1) src and dest are arrays.
6603 // (2) src and dest arrays must have elements of the same BasicType
6604 // (3) src and dest must not be null.
6605 // (4) src_offset must not be negative.
6606 // (5) dest_offset must not be negative.
6607 // (6) length must not be negative.
6608 // (7) src_offset + length must not exceed length of src.
6609 // (8) dest_offset + length must not exceed length of dest.
6610 // (9) each element of an oop array must be assignable
6611
6612 // (3) src and dest must not be null.
6613 // always do this here because we need the JVM state for uncommon traps
6614 Node* null_ctl = top();
6615 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6616 assert(null_ctl->is_top(), "no null control here");
6617 dest = null_check(dest, T_ARRAY);
6618
6619 if (!can_emit_guards) {
6620 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6621 // guards but the arraycopy node could still take advantage of a
6622 // tightly allocated allocation. tightly_coupled_allocation() is
6623 // called again to make sure it takes the null check above into
6624 // account: the null check is mandatory and if it caused an
6625 // uncommon trap to be emitted then the allocation can't be
6626 // considered tightly coupled in this context.
6627 alloc = tightly_coupled_allocation(dest);
6628 }
6629
6630 bool validated = false;
6631
6632 const Type* src_type = _gvn.type(src);
6633 const Type* dest_type = _gvn.type(dest);
6634 const TypeAryPtr* top_src = src_type->isa_aryptr();
6635 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6636
6637 // Do we have the type of src?
6638 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6639 // Do we have the type of dest?
6640 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6641 // Is the type for src from speculation?
6642 bool src_spec = false;
6643 // Is the type for dest from speculation?
6644 bool dest_spec = false;
6645
6646 if ((!has_src || !has_dest) && can_emit_guards) {
6647 // We don't have sufficient type information, let's see if
6648 // speculative types can help. We need to have types for both src
6649 // and dest so that it pays off.
6650
6651 // Do we already have or could we have type information for src
6652 bool could_have_src = has_src;
6653 // Do we already have or could we have type information for dest
6654 bool could_have_dest = has_dest;
6655
6656 ciKlass* src_k = nullptr;
6657 if (!has_src) {
6658 src_k = src_type->speculative_type_not_null();
6659 if (src_k != nullptr && src_k->is_array_klass()) {
6660 could_have_src = true;
6661 }
6662 }
6663
6664 ciKlass* dest_k = nullptr;
6665 if (!has_dest) {
6666 dest_k = dest_type->speculative_type_not_null();
6667 if (dest_k != nullptr && dest_k->is_array_klass()) {
6668 could_have_dest = true;
6669 }
6670 }
6671
6672 if (could_have_src && could_have_dest) {
6673 // This is going to pay off so emit the required guards
6674 if (!has_src) {
6675 src = maybe_cast_profiled_obj(src, src_k, true);
6676 src_type = _gvn.type(src);
6677 top_src = src_type->isa_aryptr();
6678 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6679 src_spec = true;
6680 }
6681 if (!has_dest) {
6682 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6683 dest_type = _gvn.type(dest);
6684 top_dest = dest_type->isa_aryptr();
6685 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6686 dest_spec = true;
6687 }
6688 }
6689 }
6690
6691 if (has_src && has_dest && can_emit_guards) {
6692 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6693 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6694 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6695 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6696
6697 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6698 // If both arrays are object arrays then having the exact types
6699 // for both will remove the need for a subtype check at runtime
6700 // before the call and may make it possible to pick a faster copy
6701 // routine (without a subtype check on every element)
6702 // Do we have the exact type of src?
6703 bool could_have_src = src_spec;
6704 // Do we have the exact type of dest?
6705 bool could_have_dest = dest_spec;
6706 ciKlass* src_k = nullptr;
6707 ciKlass* dest_k = nullptr;
6708 if (!src_spec) {
6709 src_k = src_type->speculative_type_not_null();
6710 if (src_k != nullptr && src_k->is_array_klass()) {
6711 could_have_src = true;
6712 }
6713 }
6714 if (!dest_spec) {
6715 dest_k = dest_type->speculative_type_not_null();
6716 if (dest_k != nullptr && dest_k->is_array_klass()) {
6717 could_have_dest = true;
6718 }
6719 }
6720 if (could_have_src && could_have_dest) {
6721 // If we can have both exact types, emit the missing guards
6722 if (could_have_src && !src_spec) {
6723 src = maybe_cast_profiled_obj(src, src_k, true);
6724 src_type = _gvn.type(src);
6725 top_src = src_type->isa_aryptr();
6726 }
6727 if (could_have_dest && !dest_spec) {
6728 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6729 dest_type = _gvn.type(dest);
6730 top_dest = dest_type->isa_aryptr();
6731 }
6732 }
6733 }
6734 }
6735
6736 ciMethod* trap_method = method();
6737 int trap_bci = bci();
6738 if (saved_jvms_before_guards != nullptr) {
6739 trap_method = alloc->jvms()->method();
6740 trap_bci = alloc->jvms()->bci();
6741 }
6742
6743 bool negative_length_guard_generated = false;
6744
6745 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6746 can_emit_guards && !src->is_top() && !dest->is_top()) {
6747 // validate arguments: enables transformation the ArrayCopyNode
6748 validated = true;
6749
6750 RegionNode* slow_region = new RegionNode(1);
6751 record_for_igvn(slow_region);
6752
6753 // (1) src and dest are arrays.
6754 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6755 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6756
6757 // (2) src and dest arrays must have elements of the same BasicType
6758 // done at macro expansion or at Ideal transformation time
6759
6760 // (4) src_offset must not be negative.
6761 generate_negative_guard(src_offset, slow_region);
6762
6763 // (5) dest_offset must not be negative.
6764 generate_negative_guard(dest_offset, slow_region);
6765
6766 // (7) src_offset + length must not exceed length of src.
6767 generate_limit_guard(src_offset, length,
6768 load_array_length(src),
6769 slow_region);
6770
6771 // (8) dest_offset + length must not exceed length of dest.
6772 generate_limit_guard(dest_offset, length,
6773 load_array_length(dest),
6774 slow_region);
6775
6776 // (6) length must not be negative.
6777 // This is also checked in generate_arraycopy() during macro expansion, but
6778 // we also have to check it here for the case where the ArrayCopyNode will
6779 // be eliminated by Escape Analysis.
6780 if (EliminateAllocations) {
6781 generate_negative_guard(length, slow_region);
6782 negative_length_guard_generated = true;
6783 }
6784
6785 // (9) each element of an oop array must be assignable
6786 Node* dest_klass = load_object_klass(dest);
6787 Node* refined_dest_klass = dest_klass;
6788 if (src != dest) {
6789 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6790 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6791 slow_region->add_req(not_subtype_ctrl);
6792 }
6793
6794 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6795 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6796 Node* src_klass = load_object_klass(src);
6797 Node* adr_prop_src = basic_plus_adr(src_klass, in_bytes(ArrayKlass::properties_offset()));
6798 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6799 Node* adr_prop_dest = basic_plus_adr(refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6800 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6801
6802 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6803 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6804 prop_src = _gvn.transform(new AndINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6805
6806 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6807 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6808 generate_fair_guard(tst, slow_region);
6809
6810 // TODO 8350865 This is too strong
6811 generate_fair_guard(flat_array_test(src), slow_region);
6812 generate_fair_guard(flat_array_test(dest), slow_region);
6813
6814 {
6815 PreserveJVMState pjvms(this);
6816 set_control(_gvn.transform(slow_region));
6817 uncommon_trap(Deoptimization::Reason_intrinsic,
6818 Deoptimization::Action_make_not_entrant);
6819 assert(stopped(), "Should be stopped");
6820 }
6821
6822 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->isa_klassptr();
6823 if (dest_klass_t == nullptr) {
6824 // refined_dest_klass may not be an array, which leads to dest_klass being top. This means we
6825 // are in a dead path.
6826 uncommon_trap(Deoptimization::Reason_intrinsic,
6827 Deoptimization::Action_make_not_entrant);
6828 return true;
6829 }
6830
6831 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6832 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6833 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6834 }
6835
6836 if (stopped()) {
6837 return true;
6838 }
6839
6840 Node* dest_klass = load_object_klass(dest);
6841 dest_klass = load_non_refined_array_klass(dest_klass);
6842
6843 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6844 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6845 // so the compiler has a chance to eliminate them: during macro expansion,
6846 // we have to set their control (CastPP nodes are eliminated).
6847 load_object_klass(src), dest_klass,
6848 load_array_length(src), load_array_length(dest));
6849
6850 ac->set_arraycopy(validated);
6851
6852 Node* n = _gvn.transform(ac);
6853 if (n == ac) {
6854 ac->connect_outputs(this);
6855 } else {
6856 assert(validated, "shouldn't transform if all arguments not validated");
6857 set_all_memory(n);
6858 }
6859 clear_upper_avx();
6860
6861
6862 return true;
6863 }
6864
6865
6866 // Helper function which determines if an arraycopy immediately follows
6867 // an allocation, with no intervening tests or other escapes for the object.
6868 AllocateArrayNode*
6869 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6870 if (stopped()) return nullptr; // no fast path
6871 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6872
6873 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6874 if (alloc == nullptr) return nullptr;
6875
6876 Node* rawmem = memory(Compile::AliasIdxRaw);
6877 // Is the allocation's memory state untouched?
6878 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6879 // Bail out if there have been raw-memory effects since the allocation.
6880 // (Example: There might have been a call or safepoint.)
6881 return nullptr;
6882 }
6883 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6884 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6885 return nullptr;
6886 }
6887
6888 // There must be no unexpected observers of this allocation.
6889 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6890 Node* obs = ptr->fast_out(i);
6891 if (obs != this->map()) {
6892 return nullptr;
6893 }
6894 }
6895
6896 // This arraycopy must unconditionally follow the allocation of the ptr.
6897 Node* alloc_ctl = ptr->in(0);
6898 Node* ctl = control();
6899 while (ctl != alloc_ctl) {
6900 // There may be guards which feed into the slow_region.
6901 // Any other control flow means that we might not get a chance
6902 // to finish initializing the allocated object.
6903 // Various low-level checks bottom out in uncommon traps. These
6904 // are considered safe since we've already checked above that
6905 // there is no unexpected observer of this allocation.
6906 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6907 assert(ctl->in(0)->is_If(), "must be If");
6908 ctl = ctl->in(0)->in(0);
6909 } else {
6910 return nullptr;
6911 }
6912 }
6913
6914 // If we get this far, we have an allocation which immediately
6915 // precedes the arraycopy, and we can take over zeroing the new object.
6916 // The arraycopy will finish the initialization, and provide
6917 // a new control state to which we will anchor the destination pointer.
6918
6919 return alloc;
6920 }
6921
6922 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6923 if (node->is_IfProj()) {
6924 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6925 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6926 Node* obs = other_proj->fast_out(j);
6927 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6928 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6929 return obs->as_CallStaticJava();
6930 }
6931 }
6932 }
6933 return nullptr;
6934 }
6935
6936 //-------------inline_encodeISOArray-----------------------------------
6937 // encode char[] to byte[] in ISO_8859_1 or ASCII
6938 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6939 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6940 // no receiver since it is static method
6941 Node *src = argument(0);
6942 Node *src_offset = argument(1);
6943 Node *dst = argument(2);
6944 Node *dst_offset = argument(3);
6945 Node *length = argument(4);
6946
6947 src = must_be_not_null(src, true);
6948 dst = must_be_not_null(dst, true);
6949
6950 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6951 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6952 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6953 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6954 // failed array check
6955 return false;
6956 }
6957
6958 // Figure out the size and type of the elements we will be copying.
6959 BasicType src_elem = src_type->elem()->array_element_basic_type();
6960 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6961 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6962 return false;
6963 }
6964
6965 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6966 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6967 // 'src_start' points to src array + scaled offset
6968 // 'dst_start' points to dst array + scaled offset
6969
6970 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6971 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6972 enc = _gvn.transform(enc);
6973 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6974 set_memory(res_mem, mtype);
6975 set_result(enc);
6976 clear_upper_avx();
6977
6978 return true;
6979 }
6980
6981 //-------------inline_multiplyToLen-----------------------------------
6982 bool LibraryCallKit::inline_multiplyToLen() {
6983 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6984
6985 address stubAddr = StubRoutines::multiplyToLen();
6986 if (stubAddr == nullptr) {
6987 return false; // Intrinsic's stub is not implemented on this platform
6988 }
6989 const char* stubName = "multiplyToLen";
6990
6991 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6992
6993 // no receiver because it is a static method
6994 Node* x = argument(0);
6995 Node* xlen = argument(1);
6996 Node* y = argument(2);
6997 Node* ylen = argument(3);
6998 Node* z = argument(4);
6999
7000 x = must_be_not_null(x, true);
7001 y = must_be_not_null(y, true);
7002
7003 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7004 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
7005 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7006 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
7007 // failed array check
7008 return false;
7009 }
7010
7011 BasicType x_elem = x_type->elem()->array_element_basic_type();
7012 BasicType y_elem = y_type->elem()->array_element_basic_type();
7013 if (x_elem != T_INT || y_elem != T_INT) {
7014 return false;
7015 }
7016
7017 Node* x_start = array_element_address(x, intcon(0), x_elem);
7018 Node* y_start = array_element_address(y, intcon(0), y_elem);
7019 // 'x_start' points to x array + scaled xlen
7020 // 'y_start' points to y array + scaled ylen
7021
7022 Node* z_start = array_element_address(z, intcon(0), T_INT);
7023
7024 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7025 OptoRuntime::multiplyToLen_Type(),
7026 stubAddr, stubName, TypePtr::BOTTOM,
7027 x_start, xlen, y_start, ylen, z_start);
7028
7029 C->set_has_split_ifs(true); // Has chance for split-if optimization
7030 set_result(z);
7031 return true;
7032 }
7033
7034 //-------------inline_squareToLen------------------------------------
7035 bool LibraryCallKit::inline_squareToLen() {
7036 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7037
7038 address stubAddr = StubRoutines::squareToLen();
7039 if (stubAddr == nullptr) {
7040 return false; // Intrinsic's stub is not implemented on this platform
7041 }
7042 const char* stubName = "squareToLen";
7043
7044 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7045
7046 Node* x = argument(0);
7047 Node* len = argument(1);
7048 Node* z = argument(2);
7049 Node* zlen = argument(3);
7050
7051 x = must_be_not_null(x, true);
7052 z = must_be_not_null(z, true);
7053
7054 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7055 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7056 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7057 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7058 // failed array check
7059 return false;
7060 }
7061
7062 BasicType x_elem = x_type->elem()->array_element_basic_type();
7063 BasicType z_elem = z_type->elem()->array_element_basic_type();
7064 if (x_elem != T_INT || z_elem != T_INT) {
7065 return false;
7066 }
7067
7068
7069 Node* x_start = array_element_address(x, intcon(0), x_elem);
7070 Node* z_start = array_element_address(z, intcon(0), z_elem);
7071
7072 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7073 OptoRuntime::squareToLen_Type(),
7074 stubAddr, stubName, TypePtr::BOTTOM,
7075 x_start, len, z_start, zlen);
7076
7077 set_result(z);
7078 return true;
7079 }
7080
7081 //-------------inline_mulAdd------------------------------------------
7082 bool LibraryCallKit::inline_mulAdd() {
7083 assert(UseMulAddIntrinsic, "not implemented on this platform");
7084
7085 address stubAddr = StubRoutines::mulAdd();
7086 if (stubAddr == nullptr) {
7087 return false; // Intrinsic's stub is not implemented on this platform
7088 }
7089 const char* stubName = "mulAdd";
7090
7091 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7092
7093 Node* out = argument(0);
7094 Node* in = argument(1);
7095 Node* offset = argument(2);
7096 Node* len = argument(3);
7097 Node* k = argument(4);
7098
7099 in = must_be_not_null(in, true);
7100 out = must_be_not_null(out, true);
7101
7102 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7103 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7104 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7105 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7106 // failed array check
7107 return false;
7108 }
7109
7110 BasicType out_elem = out_type->elem()->array_element_basic_type();
7111 BasicType in_elem = in_type->elem()->array_element_basic_type();
7112 if (out_elem != T_INT || in_elem != T_INT) {
7113 return false;
7114 }
7115
7116 Node* outlen = load_array_length(out);
7117 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7118 Node* out_start = array_element_address(out, intcon(0), out_elem);
7119 Node* in_start = array_element_address(in, intcon(0), in_elem);
7120
7121 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7122 OptoRuntime::mulAdd_Type(),
7123 stubAddr, stubName, TypePtr::BOTTOM,
7124 out_start,in_start, new_offset, len, k);
7125 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7126 set_result(result);
7127 return true;
7128 }
7129
7130 //-------------inline_montgomeryMultiply-----------------------------------
7131 bool LibraryCallKit::inline_montgomeryMultiply() {
7132 address stubAddr = StubRoutines::montgomeryMultiply();
7133 if (stubAddr == nullptr) {
7134 return false; // Intrinsic's stub is not implemented on this platform
7135 }
7136
7137 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7138 const char* stubName = "montgomery_multiply";
7139
7140 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7141
7142 Node* a = argument(0);
7143 Node* b = argument(1);
7144 Node* n = argument(2);
7145 Node* len = argument(3);
7146 Node* inv = argument(4);
7147 Node* m = argument(6);
7148
7149 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7150 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7151 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7152 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7153 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7154 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7155 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7156 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7157 // failed array check
7158 return false;
7159 }
7160
7161 BasicType a_elem = a_type->elem()->array_element_basic_type();
7162 BasicType b_elem = b_type->elem()->array_element_basic_type();
7163 BasicType n_elem = n_type->elem()->array_element_basic_type();
7164 BasicType m_elem = m_type->elem()->array_element_basic_type();
7165 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7166 return false;
7167 }
7168
7169 // Make the call
7170 {
7171 Node* a_start = array_element_address(a, intcon(0), a_elem);
7172 Node* b_start = array_element_address(b, intcon(0), b_elem);
7173 Node* n_start = array_element_address(n, intcon(0), n_elem);
7174 Node* m_start = array_element_address(m, intcon(0), m_elem);
7175
7176 Node* call = make_runtime_call(RC_LEAF,
7177 OptoRuntime::montgomeryMultiply_Type(),
7178 stubAddr, stubName, TypePtr::BOTTOM,
7179 a_start, b_start, n_start, len, inv, top(),
7180 m_start);
7181 set_result(m);
7182 }
7183
7184 return true;
7185 }
7186
7187 bool LibraryCallKit::inline_montgomerySquare() {
7188 address stubAddr = StubRoutines::montgomerySquare();
7189 if (stubAddr == nullptr) {
7190 return false; // Intrinsic's stub is not implemented on this platform
7191 }
7192
7193 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7194 const char* stubName = "montgomery_square";
7195
7196 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7197
7198 Node* a = argument(0);
7199 Node* n = argument(1);
7200 Node* len = argument(2);
7201 Node* inv = argument(3);
7202 Node* m = argument(5);
7203
7204 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7205 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7206 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7207 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7208 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7209 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7210 // failed array check
7211 return false;
7212 }
7213
7214 BasicType a_elem = a_type->elem()->array_element_basic_type();
7215 BasicType n_elem = n_type->elem()->array_element_basic_type();
7216 BasicType m_elem = m_type->elem()->array_element_basic_type();
7217 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7218 return false;
7219 }
7220
7221 // Make the call
7222 {
7223 Node* a_start = array_element_address(a, intcon(0), a_elem);
7224 Node* n_start = array_element_address(n, intcon(0), n_elem);
7225 Node* m_start = array_element_address(m, intcon(0), m_elem);
7226
7227 Node* call = make_runtime_call(RC_LEAF,
7228 OptoRuntime::montgomerySquare_Type(),
7229 stubAddr, stubName, TypePtr::BOTTOM,
7230 a_start, n_start, len, inv, top(),
7231 m_start);
7232 set_result(m);
7233 }
7234
7235 return true;
7236 }
7237
7238 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7239 address stubAddr = nullptr;
7240 const char* stubName = nullptr;
7241
7242 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7243 if (stubAddr == nullptr) {
7244 return false; // Intrinsic's stub is not implemented on this platform
7245 }
7246
7247 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7248
7249 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7250
7251 Node* newArr = argument(0);
7252 Node* oldArr = argument(1);
7253 Node* newIdx = argument(2);
7254 Node* shiftCount = argument(3);
7255 Node* numIter = argument(4);
7256
7257 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7258 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7259 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7260 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7261 return false;
7262 }
7263
7264 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7265 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7266 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7267 return false;
7268 }
7269
7270 // Make the call
7271 {
7272 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7273 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7274
7275 Node* call = make_runtime_call(RC_LEAF,
7276 OptoRuntime::bigIntegerShift_Type(),
7277 stubAddr,
7278 stubName,
7279 TypePtr::BOTTOM,
7280 newArr_start,
7281 oldArr_start,
7282 newIdx,
7283 shiftCount,
7284 numIter);
7285 }
7286
7287 return true;
7288 }
7289
7290 //-------------inline_vectorizedMismatch------------------------------
7291 bool LibraryCallKit::inline_vectorizedMismatch() {
7292 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7293
7294 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7295 Node* obja = argument(0); // Object
7296 Node* aoffset = argument(1); // long
7297 Node* objb = argument(3); // Object
7298 Node* boffset = argument(4); // long
7299 Node* length = argument(6); // int
7300 Node* scale = argument(7); // int
7301
7302 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7303 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7304 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7305 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7306 scale == top()) {
7307 return false; // failed input validation
7308 }
7309
7310 Node* obja_adr = make_unsafe_address(obja, aoffset);
7311 Node* objb_adr = make_unsafe_address(objb, boffset);
7312
7313 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7314 //
7315 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7316 // if (length <= inline_limit) {
7317 // inline_path:
7318 // vmask = VectorMaskGen length
7319 // vload1 = LoadVectorMasked obja, vmask
7320 // vload2 = LoadVectorMasked objb, vmask
7321 // result1 = VectorCmpMasked vload1, vload2, vmask
7322 // } else {
7323 // call_stub_path:
7324 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7325 // }
7326 // exit_block:
7327 // return Phi(result1, result2);
7328 //
7329 enum { inline_path = 1, // input is small enough to process it all at once
7330 stub_path = 2, // input is too large; call into the VM
7331 PATH_LIMIT = 3
7332 };
7333
7334 Node* exit_block = new RegionNode(PATH_LIMIT);
7335 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7336 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7337
7338 Node* call_stub_path = control();
7339
7340 BasicType elem_bt = T_ILLEGAL;
7341
7342 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7343 if (scale_t->is_con()) {
7344 switch (scale_t->get_con()) {
7345 case 0: elem_bt = T_BYTE; break;
7346 case 1: elem_bt = T_SHORT; break;
7347 case 2: elem_bt = T_INT; break;
7348 case 3: elem_bt = T_LONG; break;
7349
7350 default: elem_bt = T_ILLEGAL; break; // not supported
7351 }
7352 }
7353
7354 int inline_limit = 0;
7355 bool do_partial_inline = false;
7356
7357 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7358 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7359 do_partial_inline = inline_limit >= 16;
7360 }
7361
7362 if (do_partial_inline) {
7363 assert(elem_bt != T_ILLEGAL, "sanity");
7364
7365 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7366 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7367 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7368
7369 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7370 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7371 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7372
7373 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7374
7375 if (!stopped()) {
7376 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7377
7378 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7379 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7380 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7381 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7382
7383 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7384 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7385 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7386 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7387
7388 exit_block->init_req(inline_path, control());
7389 memory_phi->init_req(inline_path, map()->memory());
7390 result_phi->init_req(inline_path, result);
7391
7392 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7393 clear_upper_avx();
7394 }
7395 }
7396 }
7397
7398 if (call_stub_path != nullptr) {
7399 set_control(call_stub_path);
7400
7401 Node* call = make_runtime_call(RC_LEAF,
7402 OptoRuntime::vectorizedMismatch_Type(),
7403 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7404 obja_adr, objb_adr, length, scale);
7405
7406 exit_block->init_req(stub_path, control());
7407 memory_phi->init_req(stub_path, map()->memory());
7408 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7409 }
7410
7411 exit_block = _gvn.transform(exit_block);
7412 memory_phi = _gvn.transform(memory_phi);
7413 result_phi = _gvn.transform(result_phi);
7414
7415 record_for_igvn(exit_block);
7416 record_for_igvn(memory_phi);
7417 record_for_igvn(result_phi);
7418
7419 set_control(exit_block);
7420 set_all_memory(memory_phi);
7421 set_result(result_phi);
7422
7423 return true;
7424 }
7425
7426 //------------------------------inline_vectorizedHashcode----------------------------
7427 bool LibraryCallKit::inline_vectorizedHashCode() {
7428 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7429
7430 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7431 Node* array = argument(0);
7432 Node* offset = argument(1);
7433 Node* length = argument(2);
7434 Node* initialValue = argument(3);
7435 Node* basic_type = argument(4);
7436
7437 if (basic_type == top()) {
7438 return false; // failed input validation
7439 }
7440
7441 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7442 if (!basic_type_t->is_con()) {
7443 return false; // Only intrinsify if mode argument is constant
7444 }
7445
7446 array = must_be_not_null(array, true);
7447
7448 BasicType bt = (BasicType)basic_type_t->get_con();
7449
7450 // Resolve address of first element
7451 Node* array_start = array_element_address(array, offset, bt);
7452
7453 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7454 array_start, length, initialValue, basic_type)));
7455 clear_upper_avx();
7456
7457 return true;
7458 }
7459
7460 /**
7461 * Calculate CRC32 for byte.
7462 * int java.util.zip.CRC32.update(int crc, int b)
7463 */
7464 bool LibraryCallKit::inline_updateCRC32() {
7465 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7466 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7467 // no receiver since it is static method
7468 Node* crc = argument(0); // type: int
7469 Node* b = argument(1); // type: int
7470
7471 /*
7472 * int c = ~ crc;
7473 * b = timesXtoThe32[(b ^ c) & 0xFF];
7474 * b = b ^ (c >>> 8);
7475 * crc = ~b;
7476 */
7477
7478 Node* M1 = intcon(-1);
7479 crc = _gvn.transform(new XorINode(crc, M1));
7480 Node* result = _gvn.transform(new XorINode(crc, b));
7481 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7482
7483 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7484 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7485 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7486 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7487
7488 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7489 result = _gvn.transform(new XorINode(crc, result));
7490 result = _gvn.transform(new XorINode(result, M1));
7491 set_result(result);
7492 return true;
7493 }
7494
7495 /**
7496 * Calculate CRC32 for byte[] array.
7497 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7498 */
7499 bool LibraryCallKit::inline_updateBytesCRC32() {
7500 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7501 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7502 // no receiver since it is static method
7503 Node* crc = argument(0); // type: int
7504 Node* src = argument(1); // type: oop
7505 Node* offset = argument(2); // type: int
7506 Node* length = argument(3); // type: int
7507
7508 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7509 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7510 // failed array check
7511 return false;
7512 }
7513
7514 // Figure out the size and type of the elements we will be copying.
7515 BasicType src_elem = src_type->elem()->array_element_basic_type();
7516 if (src_elem != T_BYTE) {
7517 return false;
7518 }
7519
7520 // 'src_start' points to src array + scaled offset
7521 src = must_be_not_null(src, true);
7522 Node* src_start = array_element_address(src, offset, src_elem);
7523
7524 // We assume that range check is done by caller.
7525 // TODO: generate range check (offset+length < src.length) in debug VM.
7526
7527 // Call the stub.
7528 address stubAddr = StubRoutines::updateBytesCRC32();
7529 const char *stubName = "updateBytesCRC32";
7530
7531 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7532 stubAddr, stubName, TypePtr::BOTTOM,
7533 crc, src_start, length);
7534 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7535 set_result(result);
7536 return true;
7537 }
7538
7539 /**
7540 * Calculate CRC32 for ByteBuffer.
7541 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7542 */
7543 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7544 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7545 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7546 // no receiver since it is static method
7547 Node* crc = argument(0); // type: int
7548 Node* src = argument(1); // type: long
7549 Node* offset = argument(3); // type: int
7550 Node* length = argument(4); // type: int
7551
7552 src = ConvL2X(src); // adjust Java long to machine word
7553 Node* base = _gvn.transform(new CastX2PNode(src));
7554 offset = ConvI2X(offset);
7555
7556 // 'src_start' points to src array + scaled offset
7557 Node* src_start = basic_plus_adr(top(), base, offset);
7558
7559 // Call the stub.
7560 address stubAddr = StubRoutines::updateBytesCRC32();
7561 const char *stubName = "updateBytesCRC32";
7562
7563 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7564 stubAddr, stubName, TypePtr::BOTTOM,
7565 crc, src_start, length);
7566 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7567 set_result(result);
7568 return true;
7569 }
7570
7571 //------------------------------get_table_from_crc32c_class-----------------------
7572 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7573 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7574 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7575
7576 return table;
7577 }
7578
7579 //------------------------------inline_updateBytesCRC32C-----------------------
7580 //
7581 // Calculate CRC32C for byte[] array.
7582 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7583 //
7584 bool LibraryCallKit::inline_updateBytesCRC32C() {
7585 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7586 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7587 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7588 // no receiver since it is a static method
7589 Node* crc = argument(0); // type: int
7590 Node* src = argument(1); // type: oop
7591 Node* offset = argument(2); // type: int
7592 Node* end = argument(3); // type: int
7593
7594 Node* length = _gvn.transform(new SubINode(end, offset));
7595
7596 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7597 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7598 // failed array check
7599 return false;
7600 }
7601
7602 // Figure out the size and type of the elements we will be copying.
7603 BasicType src_elem = src_type->elem()->array_element_basic_type();
7604 if (src_elem != T_BYTE) {
7605 return false;
7606 }
7607
7608 // 'src_start' points to src array + scaled offset
7609 src = must_be_not_null(src, true);
7610 Node* src_start = array_element_address(src, offset, src_elem);
7611
7612 // static final int[] byteTable in class CRC32C
7613 Node* table = get_table_from_crc32c_class(callee()->holder());
7614 table = must_be_not_null(table, true);
7615 Node* table_start = array_element_address(table, intcon(0), T_INT);
7616
7617 // We assume that range check is done by caller.
7618 // TODO: generate range check (offset+length < src.length) in debug VM.
7619
7620 // Call the stub.
7621 address stubAddr = StubRoutines::updateBytesCRC32C();
7622 const char *stubName = "updateBytesCRC32C";
7623
7624 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7625 stubAddr, stubName, TypePtr::BOTTOM,
7626 crc, src_start, length, table_start);
7627 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7628 set_result(result);
7629 return true;
7630 }
7631
7632 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7633 //
7634 // Calculate CRC32C for DirectByteBuffer.
7635 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7636 //
7637 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7638 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7639 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7640 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7641 // no receiver since it is a static method
7642 Node* crc = argument(0); // type: int
7643 Node* src = argument(1); // type: long
7644 Node* offset = argument(3); // type: int
7645 Node* end = argument(4); // type: int
7646
7647 Node* length = _gvn.transform(new SubINode(end, offset));
7648
7649 src = ConvL2X(src); // adjust Java long to machine word
7650 Node* base = _gvn.transform(new CastX2PNode(src));
7651 offset = ConvI2X(offset);
7652
7653 // 'src_start' points to src array + scaled offset
7654 Node* src_start = basic_plus_adr(top(), base, offset);
7655
7656 // static final int[] byteTable in class CRC32C
7657 Node* table = get_table_from_crc32c_class(callee()->holder());
7658 table = must_be_not_null(table, true);
7659 Node* table_start = array_element_address(table, intcon(0), T_INT);
7660
7661 // Call the stub.
7662 address stubAddr = StubRoutines::updateBytesCRC32C();
7663 const char *stubName = "updateBytesCRC32C";
7664
7665 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7666 stubAddr, stubName, TypePtr::BOTTOM,
7667 crc, src_start, length, table_start);
7668 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7669 set_result(result);
7670 return true;
7671 }
7672
7673 //------------------------------inline_updateBytesAdler32----------------------
7674 //
7675 // Calculate Adler32 checksum for byte[] array.
7676 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7677 //
7678 bool LibraryCallKit::inline_updateBytesAdler32() {
7679 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7680 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7681 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7682 // no receiver since it is static method
7683 Node* crc = argument(0); // type: int
7684 Node* src = argument(1); // type: oop
7685 Node* offset = argument(2); // type: int
7686 Node* length = argument(3); // type: int
7687
7688 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7689 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7690 // failed array check
7691 return false;
7692 }
7693
7694 // Figure out the size and type of the elements we will be copying.
7695 BasicType src_elem = src_type->elem()->array_element_basic_type();
7696 if (src_elem != T_BYTE) {
7697 return false;
7698 }
7699
7700 // 'src_start' points to src array + scaled offset
7701 Node* src_start = array_element_address(src, offset, src_elem);
7702
7703 // We assume that range check is done by caller.
7704 // TODO: generate range check (offset+length < src.length) in debug VM.
7705
7706 // Call the stub.
7707 address stubAddr = StubRoutines::updateBytesAdler32();
7708 const char *stubName = "updateBytesAdler32";
7709
7710 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7711 stubAddr, stubName, TypePtr::BOTTOM,
7712 crc, src_start, length);
7713 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7714 set_result(result);
7715 return true;
7716 }
7717
7718 //------------------------------inline_updateByteBufferAdler32---------------
7719 //
7720 // Calculate Adler32 checksum for DirectByteBuffer.
7721 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7722 //
7723 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7724 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7725 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7726 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7727 // no receiver since it is static method
7728 Node* crc = argument(0); // type: int
7729 Node* src = argument(1); // type: long
7730 Node* offset = argument(3); // type: int
7731 Node* length = argument(4); // type: int
7732
7733 src = ConvL2X(src); // adjust Java long to machine word
7734 Node* base = _gvn.transform(new CastX2PNode(src));
7735 offset = ConvI2X(offset);
7736
7737 // 'src_start' points to src array + scaled offset
7738 Node* src_start = basic_plus_adr(top(), base, offset);
7739
7740 // Call the stub.
7741 address stubAddr = StubRoutines::updateBytesAdler32();
7742 const char *stubName = "updateBytesAdler32";
7743
7744 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7745 stubAddr, stubName, TypePtr::BOTTOM,
7746 crc, src_start, length);
7747
7748 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7749 set_result(result);
7750 return true;
7751 }
7752
7753 //----------------------------inline_reference_get0----------------------------
7754 // public T java.lang.ref.Reference.get();
7755 bool LibraryCallKit::inline_reference_get0() {
7756 const int referent_offset = java_lang_ref_Reference::referent_offset();
7757
7758 // Get the argument:
7759 Node* reference_obj = null_check_receiver();
7760 if (stopped()) return true;
7761
7762 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7763 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7764 decorators, /*is_static*/ false, nullptr);
7765 if (result == nullptr) return false;
7766
7767 // Add memory barrier to prevent commoning reads from this field
7768 // across safepoint since GC can change its value.
7769 insert_mem_bar(Op_MemBarCPUOrder);
7770
7771 set_result(result);
7772 return true;
7773 }
7774
7775 //----------------------------inline_reference_refersTo0----------------------------
7776 // bool java.lang.ref.Reference.refersTo0();
7777 // bool java.lang.ref.PhantomReference.refersTo0();
7778 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7779 // Get arguments:
7780 Node* reference_obj = null_check_receiver();
7781 Node* other_obj = argument(1);
7782 if (stopped()) return true;
7783
7784 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7785 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7786 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7787 decorators, /*is_static*/ false, nullptr);
7788 if (referent == nullptr) return false;
7789
7790 // Add memory barrier to prevent commoning reads from this field
7791 // across safepoint since GC can change its value.
7792 insert_mem_bar(Op_MemBarCPUOrder);
7793
7794 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7795 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7796 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7797
7798 RegionNode* region = new RegionNode(3);
7799 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7800
7801 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7802 region->init_req(1, if_true);
7803 phi->init_req(1, intcon(1));
7804
7805 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7806 region->init_req(2, if_false);
7807 phi->init_req(2, intcon(0));
7808
7809 set_control(_gvn.transform(region));
7810 record_for_igvn(region);
7811 set_result(_gvn.transform(phi));
7812 return true;
7813 }
7814
7815 //----------------------------inline_reference_clear0----------------------------
7816 // void java.lang.ref.Reference.clear0();
7817 // void java.lang.ref.PhantomReference.clear0();
7818 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7819 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7820
7821 // Get arguments
7822 Node* reference_obj = null_check_receiver();
7823 if (stopped()) return true;
7824
7825 // Common access parameters
7826 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7827 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7828 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7829 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7830 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7831
7832 Node* referent = access_load_at(reference_obj,
7833 referent_field_addr,
7834 referent_field_addr_type,
7835 val_type,
7836 T_OBJECT,
7837 decorators);
7838
7839 IdealKit ideal(this);
7840 #define __ ideal.
7841 __ if_then(referent, BoolTest::ne, null());
7842 sync_kit(ideal);
7843 access_store_at(reference_obj,
7844 referent_field_addr,
7845 referent_field_addr_type,
7846 null(),
7847 val_type,
7848 T_OBJECT,
7849 decorators);
7850 __ sync_kit(this);
7851 __ end_if();
7852 final_sync(ideal);
7853 #undef __
7854
7855 return true;
7856 }
7857
7858 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7859 DecoratorSet decorators, bool is_static,
7860 ciInstanceKlass* fromKls) {
7861 if (fromKls == nullptr) {
7862 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7863 assert(tinst != nullptr, "obj is null");
7864 assert(tinst->is_loaded(), "obj is not loaded");
7865 fromKls = tinst->instance_klass();
7866 } else {
7867 assert(is_static, "only for static field access");
7868 }
7869 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7870 ciSymbol::make(fieldTypeString),
7871 is_static);
7872
7873 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7874 if (field == nullptr) return (Node *) nullptr;
7875
7876 if (is_static) {
7877 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7878 fromObj = makecon(tip);
7879 }
7880
7881 // Next code copied from Parse::do_get_xxx():
7882
7883 // Compute address and memory type.
7884 int offset = field->offset_in_bytes();
7885 bool is_vol = field->is_volatile();
7886 ciType* field_klass = field->type();
7887 assert(field_klass->is_loaded(), "should be loaded");
7888 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7889 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7890 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7891 "slice of address and input slice don't match");
7892 BasicType bt = field->layout_type();
7893
7894 // Build the resultant type of the load
7895 const Type *type;
7896 if (bt == T_OBJECT) {
7897 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7898 } else {
7899 type = Type::get_const_basic_type(bt);
7900 }
7901
7902 if (is_vol) {
7903 decorators |= MO_SEQ_CST;
7904 }
7905
7906 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7907 }
7908
7909 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7910 bool is_exact /* true */, bool is_static /* false */,
7911 ciInstanceKlass * fromKls /* nullptr */) {
7912 if (fromKls == nullptr) {
7913 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7914 assert(tinst != nullptr, "obj is null");
7915 assert(tinst->is_loaded(), "obj is not loaded");
7916 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7917 fromKls = tinst->instance_klass();
7918 }
7919 else {
7920 assert(is_static, "only for static field access");
7921 }
7922 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7923 ciSymbol::make(fieldTypeString),
7924 is_static);
7925
7926 assert(field != nullptr, "undefined field");
7927 assert(!field->is_volatile(), "not defined for volatile fields");
7928
7929 if (is_static) {
7930 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7931 fromObj = makecon(tip);
7932 }
7933
7934 // Next code copied from Parse::do_get_xxx():
7935
7936 // Compute address and memory type.
7937 int offset = field->offset_in_bytes();
7938 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7939
7940 return adr;
7941 }
7942
7943 //------------------------------inline_aescrypt_Block-----------------------
7944 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7945 address stubAddr = nullptr;
7946 const char *stubName;
7947 bool is_decrypt = false;
7948 assert(UseAES, "need AES instruction support");
7949
7950 switch(id) {
7951 case vmIntrinsics::_aescrypt_encryptBlock:
7952 stubAddr = StubRoutines::aescrypt_encryptBlock();
7953 stubName = "aescrypt_encryptBlock";
7954 break;
7955 case vmIntrinsics::_aescrypt_decryptBlock:
7956 stubAddr = StubRoutines::aescrypt_decryptBlock();
7957 stubName = "aescrypt_decryptBlock";
7958 is_decrypt = true;
7959 break;
7960 default:
7961 break;
7962 }
7963 if (stubAddr == nullptr) return false;
7964
7965 Node* aescrypt_object = argument(0);
7966 Node* src = argument(1);
7967 Node* src_offset = argument(2);
7968 Node* dest = argument(3);
7969 Node* dest_offset = argument(4);
7970
7971 src = must_be_not_null(src, true);
7972 dest = must_be_not_null(dest, true);
7973
7974 // (1) src and dest are arrays.
7975 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7976 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7977 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7978 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7979
7980 // for the quick and dirty code we will skip all the checks.
7981 // we are just trying to get the call to be generated.
7982 Node* src_start = src;
7983 Node* dest_start = dest;
7984 if (src_offset != nullptr || dest_offset != nullptr) {
7985 assert(src_offset != nullptr && dest_offset != nullptr, "");
7986 src_start = array_element_address(src, src_offset, T_BYTE);
7987 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7988 }
7989
7990 // now need to get the start of its expanded key array
7991 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7992 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7993 if (k_start == nullptr) return false;
7994
7995 // Call the stub.
7996 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7997 stubAddr, stubName, TypePtr::BOTTOM,
7998 src_start, dest_start, k_start);
7999
8000 return true;
8001 }
8002
8003 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
8004 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
8005 address stubAddr = nullptr;
8006 const char *stubName = nullptr;
8007 bool is_decrypt = false;
8008 assert(UseAES, "need AES instruction support");
8009
8010 switch(id) {
8011 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8012 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8013 stubName = "cipherBlockChaining_encryptAESCrypt";
8014 break;
8015 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8016 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8017 stubName = "cipherBlockChaining_decryptAESCrypt";
8018 is_decrypt = true;
8019 break;
8020 default:
8021 break;
8022 }
8023 if (stubAddr == nullptr) return false;
8024
8025 Node* cipherBlockChaining_object = argument(0);
8026 Node* src = argument(1);
8027 Node* src_offset = argument(2);
8028 Node* len = argument(3);
8029 Node* dest = argument(4);
8030 Node* dest_offset = argument(5);
8031
8032 src = must_be_not_null(src, false);
8033 dest = must_be_not_null(dest, false);
8034
8035 // (1) src and dest are arrays.
8036 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8037 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8038 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8039 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8040
8041 // checks are the responsibility of the caller
8042 Node* src_start = src;
8043 Node* dest_start = dest;
8044 if (src_offset != nullptr || dest_offset != nullptr) {
8045 assert(src_offset != nullptr && dest_offset != nullptr, "");
8046 src_start = array_element_address(src, src_offset, T_BYTE);
8047 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8048 }
8049
8050 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8051 // (because of the predicated logic executed earlier).
8052 // so we cast it here safely.
8053 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8054
8055 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8056 if (embeddedCipherObj == nullptr) return false;
8057
8058 // cast it to what we know it will be at runtime
8059 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8060 assert(tinst != nullptr, "CBC obj is null");
8061 assert(tinst->is_loaded(), "CBC obj is not loaded");
8062 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8063 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8064
8065 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8066 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8067 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8068 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8069 aescrypt_object = _gvn.transform(aescrypt_object);
8070
8071 // we need to get the start of the aescrypt_object's expanded key array
8072 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8073 if (k_start == nullptr) return false;
8074
8075 // similarly, get the start address of the r vector
8076 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8077 if (objRvec == nullptr) return false;
8078 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8079
8080 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8081 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8082 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8083 stubAddr, stubName, TypePtr::BOTTOM,
8084 src_start, dest_start, k_start, r_start, len);
8085
8086 // return cipher length (int)
8087 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8088 set_result(retvalue);
8089 return true;
8090 }
8091
8092 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8093 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8094 address stubAddr = nullptr;
8095 const char *stubName = nullptr;
8096 bool is_decrypt = false;
8097 assert(UseAES, "need AES instruction support");
8098
8099 switch (id) {
8100 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8101 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8102 stubName = "electronicCodeBook_encryptAESCrypt";
8103 break;
8104 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8105 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8106 stubName = "electronicCodeBook_decryptAESCrypt";
8107 is_decrypt = true;
8108 break;
8109 default:
8110 break;
8111 }
8112
8113 if (stubAddr == nullptr) return false;
8114
8115 Node* electronicCodeBook_object = argument(0);
8116 Node* src = argument(1);
8117 Node* src_offset = argument(2);
8118 Node* len = argument(3);
8119 Node* dest = argument(4);
8120 Node* dest_offset = argument(5);
8121
8122 // (1) src and dest are arrays.
8123 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8124 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8125 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8126 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8127
8128 // checks are the responsibility of the caller
8129 Node* src_start = src;
8130 Node* dest_start = dest;
8131 if (src_offset != nullptr || dest_offset != nullptr) {
8132 assert(src_offset != nullptr && dest_offset != nullptr, "");
8133 src_start = array_element_address(src, src_offset, T_BYTE);
8134 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8135 }
8136
8137 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8138 // (because of the predicated logic executed earlier).
8139 // so we cast it here safely.
8140 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8141
8142 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8143 if (embeddedCipherObj == nullptr) return false;
8144
8145 // cast it to what we know it will be at runtime
8146 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8147 assert(tinst != nullptr, "ECB obj is null");
8148 assert(tinst->is_loaded(), "ECB obj is not loaded");
8149 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8150 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8151
8152 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8153 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8154 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8155 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8156 aescrypt_object = _gvn.transform(aescrypt_object);
8157
8158 // we need to get the start of the aescrypt_object's expanded key array
8159 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
8160 if (k_start == nullptr) return false;
8161
8162 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8163 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8164 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8165 stubAddr, stubName, TypePtr::BOTTOM,
8166 src_start, dest_start, k_start, len);
8167
8168 // return cipher length (int)
8169 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8170 set_result(retvalue);
8171 return true;
8172 }
8173
8174 //------------------------------inline_counterMode_AESCrypt-----------------------
8175 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8176 assert(UseAES, "need AES instruction support");
8177 if (!UseAESCTRIntrinsics) return false;
8178
8179 address stubAddr = nullptr;
8180 const char *stubName = nullptr;
8181 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8182 stubAddr = StubRoutines::counterMode_AESCrypt();
8183 stubName = "counterMode_AESCrypt";
8184 }
8185 if (stubAddr == nullptr) return false;
8186
8187 Node* counterMode_object = argument(0);
8188 Node* src = argument(1);
8189 Node* src_offset = argument(2);
8190 Node* len = argument(3);
8191 Node* dest = argument(4);
8192 Node* dest_offset = argument(5);
8193
8194 // (1) src and dest are arrays.
8195 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8196 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8197 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8198 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8199
8200 // checks are the responsibility of the caller
8201 Node* src_start = src;
8202 Node* dest_start = dest;
8203 if (src_offset != nullptr || dest_offset != nullptr) {
8204 assert(src_offset != nullptr && dest_offset != nullptr, "");
8205 src_start = array_element_address(src, src_offset, T_BYTE);
8206 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8207 }
8208
8209 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8210 // (because of the predicated logic executed earlier).
8211 // so we cast it here safely.
8212 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8213 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8214 if (embeddedCipherObj == nullptr) return false;
8215 // cast it to what we know it will be at runtime
8216 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8217 assert(tinst != nullptr, "CTR obj is null");
8218 assert(tinst->is_loaded(), "CTR obj is not loaded");
8219 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8220 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8221 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8222 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8223 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8224 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8225 aescrypt_object = _gvn.transform(aescrypt_object);
8226 // we need to get the start of the aescrypt_object's expanded key array
8227 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8228 if (k_start == nullptr) return false;
8229 // similarly, get the start address of the r vector
8230 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8231 if (obj_counter == nullptr) return false;
8232 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8233
8234 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8235 if (saved_encCounter == nullptr) return false;
8236 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8237 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8238
8239 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8240 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8241 OptoRuntime::counterMode_aescrypt_Type(),
8242 stubAddr, stubName, TypePtr::BOTTOM,
8243 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8244
8245 // return cipher length (int)
8246 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8247 set_result(retvalue);
8248 return true;
8249 }
8250
8251 //------------------------------get_key_start_from_aescrypt_object-----------------------
8252 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
8253 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8254 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8255 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8256 // The following platform specific stubs of encryption and decryption use the same round keys.
8257 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8258 bool use_decryption_key = false;
8259 #else
8260 bool use_decryption_key = is_decrypt;
8261 #endif
8262 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
8263 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8264 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8265
8266 // now have the array, need to get the start address of the selected key array
8267 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8268 return k_start;
8269 }
8270
8271 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8272 // Return node representing slow path of predicate check.
8273 // the pseudo code we want to emulate with this predicate is:
8274 // for encryption:
8275 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8276 // for decryption:
8277 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8278 // note cipher==plain is more conservative than the original java code but that's OK
8279 //
8280 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8281 // The receiver was checked for null already.
8282 Node* objCBC = argument(0);
8283
8284 Node* src = argument(1);
8285 Node* dest = argument(4);
8286
8287 // Load embeddedCipher field of CipherBlockChaining object.
8288 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8289
8290 // get AESCrypt klass for instanceOf check
8291 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8292 // will have same classloader as CipherBlockChaining object
8293 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8294 assert(tinst != nullptr, "CBCobj is null");
8295 assert(tinst->is_loaded(), "CBCobj is not loaded");
8296
8297 // we want to do an instanceof comparison against the AESCrypt class
8298 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8299 if (!klass_AESCrypt->is_loaded()) {
8300 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8301 Node* ctrl = control();
8302 set_control(top()); // no regular fast path
8303 return ctrl;
8304 }
8305
8306 src = must_be_not_null(src, true);
8307 dest = must_be_not_null(dest, true);
8308
8309 // Resolve oops to stable for CmpP below.
8310 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8311
8312 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8313 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8314 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8315
8316 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8317
8318 // for encryption, we are done
8319 if (!decrypting)
8320 return instof_false; // even if it is null
8321
8322 // for decryption, we need to add a further check to avoid
8323 // taking the intrinsic path when cipher and plain are the same
8324 // see the original java code for why.
8325 RegionNode* region = new RegionNode(3);
8326 region->init_req(1, instof_false);
8327
8328 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8329 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8330 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8331 region->init_req(2, src_dest_conjoint);
8332
8333 record_for_igvn(region);
8334 return _gvn.transform(region);
8335 }
8336
8337 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8338 // Return node representing slow path of predicate check.
8339 // the pseudo code we want to emulate with this predicate is:
8340 // for encryption:
8341 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8342 // for decryption:
8343 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8344 // note cipher==plain is more conservative than the original java code but that's OK
8345 //
8346 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8347 // The receiver was checked for null already.
8348 Node* objECB = argument(0);
8349
8350 // Load embeddedCipher field of ElectronicCodeBook object.
8351 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8352
8353 // get AESCrypt klass for instanceOf check
8354 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8355 // will have same classloader as ElectronicCodeBook object
8356 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8357 assert(tinst != nullptr, "ECBobj is null");
8358 assert(tinst->is_loaded(), "ECBobj is not loaded");
8359
8360 // we want to do an instanceof comparison against the AESCrypt class
8361 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8362 if (!klass_AESCrypt->is_loaded()) {
8363 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8364 Node* ctrl = control();
8365 set_control(top()); // no regular fast path
8366 return ctrl;
8367 }
8368 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8369
8370 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8371 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8372 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8373
8374 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8375
8376 // for encryption, we are done
8377 if (!decrypting)
8378 return instof_false; // even if it is null
8379
8380 // for decryption, we need to add a further check to avoid
8381 // taking the intrinsic path when cipher and plain are the same
8382 // see the original java code for why.
8383 RegionNode* region = new RegionNode(3);
8384 region->init_req(1, instof_false);
8385 Node* src = argument(1);
8386 Node* dest = argument(4);
8387 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8388 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8389 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8390 region->init_req(2, src_dest_conjoint);
8391
8392 record_for_igvn(region);
8393 return _gvn.transform(region);
8394 }
8395
8396 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8397 // Return node representing slow path of predicate check.
8398 // the pseudo code we want to emulate with this predicate is:
8399 // for encryption:
8400 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8401 // for decryption:
8402 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8403 // note cipher==plain is more conservative than the original java code but that's OK
8404 //
8405
8406 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8407 // The receiver was checked for null already.
8408 Node* objCTR = argument(0);
8409
8410 // Load embeddedCipher field of CipherBlockChaining object.
8411 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8412
8413 // get AESCrypt klass for instanceOf check
8414 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8415 // will have same classloader as CipherBlockChaining object
8416 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8417 assert(tinst != nullptr, "CTRobj is null");
8418 assert(tinst->is_loaded(), "CTRobj is not loaded");
8419
8420 // we want to do an instanceof comparison against the AESCrypt class
8421 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8422 if (!klass_AESCrypt->is_loaded()) {
8423 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8424 Node* ctrl = control();
8425 set_control(top()); // no regular fast path
8426 return ctrl;
8427 }
8428
8429 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8430 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8431 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8432 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8433 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8434
8435 return instof_false; // even if it is null
8436 }
8437
8438 //------------------------------inline_ghash_processBlocks
8439 bool LibraryCallKit::inline_ghash_processBlocks() {
8440 address stubAddr;
8441 const char *stubName;
8442 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8443
8444 stubAddr = StubRoutines::ghash_processBlocks();
8445 stubName = "ghash_processBlocks";
8446
8447 Node* data = argument(0);
8448 Node* offset = argument(1);
8449 Node* len = argument(2);
8450 Node* state = argument(3);
8451 Node* subkeyH = argument(4);
8452
8453 state = must_be_not_null(state, true);
8454 subkeyH = must_be_not_null(subkeyH, true);
8455 data = must_be_not_null(data, true);
8456
8457 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8458 assert(state_start, "state is null");
8459 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8460 assert(subkeyH_start, "subkeyH is null");
8461 Node* data_start = array_element_address(data, offset, T_BYTE);
8462 assert(data_start, "data is null");
8463
8464 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8465 OptoRuntime::ghash_processBlocks_Type(),
8466 stubAddr, stubName, TypePtr::BOTTOM,
8467 state_start, subkeyH_start, data_start, len);
8468 return true;
8469 }
8470
8471 //------------------------------inline_chacha20Block
8472 bool LibraryCallKit::inline_chacha20Block() {
8473 address stubAddr;
8474 const char *stubName;
8475 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8476
8477 stubAddr = StubRoutines::chacha20Block();
8478 stubName = "chacha20Block";
8479
8480 Node* state = argument(0);
8481 Node* result = argument(1);
8482
8483 state = must_be_not_null(state, true);
8484 result = must_be_not_null(result, true);
8485
8486 Node* state_start = array_element_address(state, intcon(0), T_INT);
8487 assert(state_start, "state is null");
8488 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8489 assert(result_start, "result is null");
8490
8491 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8492 OptoRuntime::chacha20Block_Type(),
8493 stubAddr, stubName, TypePtr::BOTTOM,
8494 state_start, result_start);
8495 // return key stream length (int)
8496 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8497 set_result(retvalue);
8498 return true;
8499 }
8500
8501 //------------------------------inline_kyberNtt
8502 bool LibraryCallKit::inline_kyberNtt() {
8503 address stubAddr;
8504 const char *stubName;
8505 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8506 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8507
8508 stubAddr = StubRoutines::kyberNtt();
8509 stubName = "kyberNtt";
8510 if (!stubAddr) return false;
8511
8512 Node* coeffs = argument(0);
8513 Node* ntt_zetas = argument(1);
8514
8515 coeffs = must_be_not_null(coeffs, true);
8516 ntt_zetas = must_be_not_null(ntt_zetas, true);
8517
8518 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8519 assert(coeffs_start, "coeffs is null");
8520 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8521 assert(ntt_zetas_start, "ntt_zetas is null");
8522 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8523 OptoRuntime::kyberNtt_Type(),
8524 stubAddr, stubName, TypePtr::BOTTOM,
8525 coeffs_start, ntt_zetas_start);
8526 // return an int
8527 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8528 set_result(retvalue);
8529 return true;
8530 }
8531
8532 //------------------------------inline_kyberInverseNtt
8533 bool LibraryCallKit::inline_kyberInverseNtt() {
8534 address stubAddr;
8535 const char *stubName;
8536 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8537 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8538
8539 stubAddr = StubRoutines::kyberInverseNtt();
8540 stubName = "kyberInverseNtt";
8541 if (!stubAddr) return false;
8542
8543 Node* coeffs = argument(0);
8544 Node* zetas = argument(1);
8545
8546 coeffs = must_be_not_null(coeffs, true);
8547 zetas = must_be_not_null(zetas, true);
8548
8549 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8550 assert(coeffs_start, "coeffs is null");
8551 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8552 assert(zetas_start, "inverseNtt_zetas is null");
8553 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8554 OptoRuntime::kyberInverseNtt_Type(),
8555 stubAddr, stubName, TypePtr::BOTTOM,
8556 coeffs_start, zetas_start);
8557
8558 // return an int
8559 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8560 set_result(retvalue);
8561 return true;
8562 }
8563
8564 //------------------------------inline_kyberNttMult
8565 bool LibraryCallKit::inline_kyberNttMult() {
8566 address stubAddr;
8567 const char *stubName;
8568 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8569 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8570
8571 stubAddr = StubRoutines::kyberNttMult();
8572 stubName = "kyberNttMult";
8573 if (!stubAddr) return false;
8574
8575 Node* result = argument(0);
8576 Node* ntta = argument(1);
8577 Node* nttb = argument(2);
8578 Node* zetas = argument(3);
8579
8580 result = must_be_not_null(result, true);
8581 ntta = must_be_not_null(ntta, true);
8582 nttb = must_be_not_null(nttb, true);
8583 zetas = must_be_not_null(zetas, true);
8584
8585 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8586 assert(result_start, "result is null");
8587 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8588 assert(ntta_start, "ntta is null");
8589 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8590 assert(nttb_start, "nttb is null");
8591 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8592 assert(zetas_start, "nttMult_zetas is null");
8593 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8594 OptoRuntime::kyberNttMult_Type(),
8595 stubAddr, stubName, TypePtr::BOTTOM,
8596 result_start, ntta_start, nttb_start,
8597 zetas_start);
8598
8599 // return an int
8600 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8601 set_result(retvalue);
8602
8603 return true;
8604 }
8605
8606 //------------------------------inline_kyberAddPoly_2
8607 bool LibraryCallKit::inline_kyberAddPoly_2() {
8608 address stubAddr;
8609 const char *stubName;
8610 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8611 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8612
8613 stubAddr = StubRoutines::kyberAddPoly_2();
8614 stubName = "kyberAddPoly_2";
8615 if (!stubAddr) return false;
8616
8617 Node* result = argument(0);
8618 Node* a = argument(1);
8619 Node* b = argument(2);
8620
8621 result = must_be_not_null(result, true);
8622 a = must_be_not_null(a, true);
8623 b = must_be_not_null(b, true);
8624
8625 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8626 assert(result_start, "result is null");
8627 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8628 assert(a_start, "a is null");
8629 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8630 assert(b_start, "b is null");
8631 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8632 OptoRuntime::kyberAddPoly_2_Type(),
8633 stubAddr, stubName, TypePtr::BOTTOM,
8634 result_start, a_start, b_start);
8635 // return an int
8636 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8637 set_result(retvalue);
8638 return true;
8639 }
8640
8641 //------------------------------inline_kyberAddPoly_3
8642 bool LibraryCallKit::inline_kyberAddPoly_3() {
8643 address stubAddr;
8644 const char *stubName;
8645 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8646 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8647
8648 stubAddr = StubRoutines::kyberAddPoly_3();
8649 stubName = "kyberAddPoly_3";
8650 if (!stubAddr) return false;
8651
8652 Node* result = argument(0);
8653 Node* a = argument(1);
8654 Node* b = argument(2);
8655 Node* c = argument(3);
8656
8657 result = must_be_not_null(result, true);
8658 a = must_be_not_null(a, true);
8659 b = must_be_not_null(b, true);
8660 c = must_be_not_null(c, true);
8661
8662 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8663 assert(result_start, "result is null");
8664 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8665 assert(a_start, "a is null");
8666 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8667 assert(b_start, "b is null");
8668 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8669 assert(c_start, "c is null");
8670 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8671 OptoRuntime::kyberAddPoly_3_Type(),
8672 stubAddr, stubName, TypePtr::BOTTOM,
8673 result_start, a_start, b_start, c_start);
8674 // return an int
8675 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8676 set_result(retvalue);
8677 return true;
8678 }
8679
8680 //------------------------------inline_kyber12To16
8681 bool LibraryCallKit::inline_kyber12To16() {
8682 address stubAddr;
8683 const char *stubName;
8684 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8685 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8686
8687 stubAddr = StubRoutines::kyber12To16();
8688 stubName = "kyber12To16";
8689 if (!stubAddr) return false;
8690
8691 Node* condensed = argument(0);
8692 Node* condensedOffs = argument(1);
8693 Node* parsed = argument(2);
8694 Node* parsedLength = argument(3);
8695
8696 condensed = must_be_not_null(condensed, true);
8697 parsed = must_be_not_null(parsed, true);
8698
8699 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8700 assert(condensed_start, "condensed is null");
8701 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8702 assert(parsed_start, "parsed is null");
8703 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8704 OptoRuntime::kyber12To16_Type(),
8705 stubAddr, stubName, TypePtr::BOTTOM,
8706 condensed_start, condensedOffs, parsed_start, parsedLength);
8707 // return an int
8708 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8709 set_result(retvalue);
8710 return true;
8711
8712 }
8713
8714 //------------------------------inline_kyberBarrettReduce
8715 bool LibraryCallKit::inline_kyberBarrettReduce() {
8716 address stubAddr;
8717 const char *stubName;
8718 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8719 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8720
8721 stubAddr = StubRoutines::kyberBarrettReduce();
8722 stubName = "kyberBarrettReduce";
8723 if (!stubAddr) return false;
8724
8725 Node* coeffs = argument(0);
8726
8727 coeffs = must_be_not_null(coeffs, true);
8728
8729 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8730 assert(coeffs_start, "coeffs is null");
8731 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8732 OptoRuntime::kyberBarrettReduce_Type(),
8733 stubAddr, stubName, TypePtr::BOTTOM,
8734 coeffs_start);
8735 // return an int
8736 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8737 set_result(retvalue);
8738 return true;
8739 }
8740
8741 //------------------------------inline_dilithiumAlmostNtt
8742 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8743 address stubAddr;
8744 const char *stubName;
8745 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8746 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8747
8748 stubAddr = StubRoutines::dilithiumAlmostNtt();
8749 stubName = "dilithiumAlmostNtt";
8750 if (!stubAddr) return false;
8751
8752 Node* coeffs = argument(0);
8753 Node* ntt_zetas = argument(1);
8754
8755 coeffs = must_be_not_null(coeffs, true);
8756 ntt_zetas = must_be_not_null(ntt_zetas, true);
8757
8758 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8759 assert(coeffs_start, "coeffs is null");
8760 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8761 assert(ntt_zetas_start, "ntt_zetas is null");
8762 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8763 OptoRuntime::dilithiumAlmostNtt_Type(),
8764 stubAddr, stubName, TypePtr::BOTTOM,
8765 coeffs_start, ntt_zetas_start);
8766 // return an int
8767 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8768 set_result(retvalue);
8769 return true;
8770 }
8771
8772 //------------------------------inline_dilithiumAlmostInverseNtt
8773 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8774 address stubAddr;
8775 const char *stubName;
8776 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8777 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8778
8779 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8780 stubName = "dilithiumAlmostInverseNtt";
8781 if (!stubAddr) return false;
8782
8783 Node* coeffs = argument(0);
8784 Node* zetas = argument(1);
8785
8786 coeffs = must_be_not_null(coeffs, true);
8787 zetas = must_be_not_null(zetas, true);
8788
8789 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8790 assert(coeffs_start, "coeffs is null");
8791 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8792 assert(zetas_start, "inverseNtt_zetas is null");
8793 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8794 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8795 stubAddr, stubName, TypePtr::BOTTOM,
8796 coeffs_start, zetas_start);
8797 // return an int
8798 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8799 set_result(retvalue);
8800 return true;
8801 }
8802
8803 //------------------------------inline_dilithiumNttMult
8804 bool LibraryCallKit::inline_dilithiumNttMult() {
8805 address stubAddr;
8806 const char *stubName;
8807 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8808 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8809
8810 stubAddr = StubRoutines::dilithiumNttMult();
8811 stubName = "dilithiumNttMult";
8812 if (!stubAddr) return false;
8813
8814 Node* result = argument(0);
8815 Node* ntta = argument(1);
8816 Node* nttb = argument(2);
8817 Node* zetas = argument(3);
8818
8819 result = must_be_not_null(result, true);
8820 ntta = must_be_not_null(ntta, true);
8821 nttb = must_be_not_null(nttb, true);
8822 zetas = must_be_not_null(zetas, true);
8823
8824 Node* result_start = array_element_address(result, intcon(0), T_INT);
8825 assert(result_start, "result is null");
8826 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8827 assert(ntta_start, "ntta is null");
8828 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8829 assert(nttb_start, "nttb is null");
8830 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8831 OptoRuntime::dilithiumNttMult_Type(),
8832 stubAddr, stubName, TypePtr::BOTTOM,
8833 result_start, ntta_start, nttb_start);
8834
8835 // return an int
8836 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8837 set_result(retvalue);
8838
8839 return true;
8840 }
8841
8842 //------------------------------inline_dilithiumMontMulByConstant
8843 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8844 address stubAddr;
8845 const char *stubName;
8846 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8847 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8848
8849 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8850 stubName = "dilithiumMontMulByConstant";
8851 if (!stubAddr) return false;
8852
8853 Node* coeffs = argument(0);
8854 Node* constant = argument(1);
8855
8856 coeffs = must_be_not_null(coeffs, true);
8857
8858 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8859 assert(coeffs_start, "coeffs is null");
8860 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8861 OptoRuntime::dilithiumMontMulByConstant_Type(),
8862 stubAddr, stubName, TypePtr::BOTTOM,
8863 coeffs_start, constant);
8864
8865 // return an int
8866 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8867 set_result(retvalue);
8868 return true;
8869 }
8870
8871
8872 //------------------------------inline_dilithiumDecomposePoly
8873 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8874 address stubAddr;
8875 const char *stubName;
8876 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8877 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8878
8879 stubAddr = StubRoutines::dilithiumDecomposePoly();
8880 stubName = "dilithiumDecomposePoly";
8881 if (!stubAddr) return false;
8882
8883 Node* input = argument(0);
8884 Node* lowPart = argument(1);
8885 Node* highPart = argument(2);
8886 Node* twoGamma2 = argument(3);
8887 Node* multiplier = argument(4);
8888
8889 input = must_be_not_null(input, true);
8890 lowPart = must_be_not_null(lowPart, true);
8891 highPart = must_be_not_null(highPart, true);
8892
8893 Node* input_start = array_element_address(input, intcon(0), T_INT);
8894 assert(input_start, "input is null");
8895 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8896 assert(lowPart_start, "lowPart is null");
8897 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8898 assert(highPart_start, "highPart is null");
8899
8900 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8901 OptoRuntime::dilithiumDecomposePoly_Type(),
8902 stubAddr, stubName, TypePtr::BOTTOM,
8903 input_start, lowPart_start, highPart_start,
8904 twoGamma2, multiplier);
8905
8906 // return an int
8907 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8908 set_result(retvalue);
8909 return true;
8910 }
8911
8912 bool LibraryCallKit::inline_base64_encodeBlock() {
8913 address stubAddr;
8914 const char *stubName;
8915 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8916 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8917 stubAddr = StubRoutines::base64_encodeBlock();
8918 stubName = "encodeBlock";
8919
8920 if (!stubAddr) return false;
8921 Node* base64obj = argument(0);
8922 Node* src = argument(1);
8923 Node* offset = argument(2);
8924 Node* len = argument(3);
8925 Node* dest = argument(4);
8926 Node* dp = argument(5);
8927 Node* isURL = argument(6);
8928
8929 src = must_be_not_null(src, true);
8930 dest = must_be_not_null(dest, true);
8931
8932 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8933 assert(src_start, "source array is null");
8934 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8935 assert(dest_start, "destination array is null");
8936
8937 Node* base64 = make_runtime_call(RC_LEAF,
8938 OptoRuntime::base64_encodeBlock_Type(),
8939 stubAddr, stubName, TypePtr::BOTTOM,
8940 src_start, offset, len, dest_start, dp, isURL);
8941 return true;
8942 }
8943
8944 bool LibraryCallKit::inline_base64_decodeBlock() {
8945 address stubAddr;
8946 const char *stubName;
8947 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8948 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8949 stubAddr = StubRoutines::base64_decodeBlock();
8950 stubName = "decodeBlock";
8951
8952 if (!stubAddr) return false;
8953 Node* base64obj = argument(0);
8954 Node* src = argument(1);
8955 Node* src_offset = argument(2);
8956 Node* len = argument(3);
8957 Node* dest = argument(4);
8958 Node* dest_offset = argument(5);
8959 Node* isURL = argument(6);
8960 Node* isMIME = argument(7);
8961
8962 src = must_be_not_null(src, true);
8963 dest = must_be_not_null(dest, true);
8964
8965 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8966 assert(src_start, "source array is null");
8967 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8968 assert(dest_start, "destination array is null");
8969
8970 Node* call = make_runtime_call(RC_LEAF,
8971 OptoRuntime::base64_decodeBlock_Type(),
8972 stubAddr, stubName, TypePtr::BOTTOM,
8973 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8974 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8975 set_result(result);
8976 return true;
8977 }
8978
8979 bool LibraryCallKit::inline_poly1305_processBlocks() {
8980 address stubAddr;
8981 const char *stubName;
8982 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8983 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8984 stubAddr = StubRoutines::poly1305_processBlocks();
8985 stubName = "poly1305_processBlocks";
8986
8987 if (!stubAddr) return false;
8988 null_check_receiver(); // null-check receiver
8989 if (stopped()) return true;
8990
8991 Node* input = argument(1);
8992 Node* input_offset = argument(2);
8993 Node* len = argument(3);
8994 Node* alimbs = argument(4);
8995 Node* rlimbs = argument(5);
8996
8997 input = must_be_not_null(input, true);
8998 alimbs = must_be_not_null(alimbs, true);
8999 rlimbs = must_be_not_null(rlimbs, true);
9000
9001 Node* input_start = array_element_address(input, input_offset, T_BYTE);
9002 assert(input_start, "input array is null");
9003 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
9004 assert(acc_start, "acc array is null");
9005 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
9006 assert(r_start, "r array is null");
9007
9008 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9009 OptoRuntime::poly1305_processBlocks_Type(),
9010 stubAddr, stubName, TypePtr::BOTTOM,
9011 input_start, len, acc_start, r_start);
9012 return true;
9013 }
9014
9015 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9016 address stubAddr;
9017 const char *stubName;
9018 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9019 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9020 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9021 stubName = "intpoly_montgomeryMult_P256";
9022
9023 if (!stubAddr) return false;
9024 null_check_receiver(); // null-check receiver
9025 if (stopped()) return true;
9026
9027 Node* a = argument(1);
9028 Node* b = argument(2);
9029 Node* r = argument(3);
9030
9031 a = must_be_not_null(a, true);
9032 b = must_be_not_null(b, true);
9033 r = must_be_not_null(r, true);
9034
9035 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9036 assert(a_start, "a array is null");
9037 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9038 assert(b_start, "b array is null");
9039 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9040 assert(r_start, "r array is null");
9041
9042 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9043 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9044 stubAddr, stubName, TypePtr::BOTTOM,
9045 a_start, b_start, r_start);
9046 return true;
9047 }
9048
9049 bool LibraryCallKit::inline_intpoly_assign() {
9050 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9051 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9052 const char *stubName = "intpoly_assign";
9053 address stubAddr = StubRoutines::intpoly_assign();
9054 if (!stubAddr) return false;
9055
9056 Node* set = argument(0);
9057 Node* a = argument(1);
9058 Node* b = argument(2);
9059 Node* arr_length = load_array_length(a);
9060
9061 a = must_be_not_null(a, true);
9062 b = must_be_not_null(b, true);
9063
9064 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9065 assert(a_start, "a array is null");
9066 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9067 assert(b_start, "b array is null");
9068
9069 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9070 OptoRuntime::intpoly_assign_Type(),
9071 stubAddr, stubName, TypePtr::BOTTOM,
9072 set, a_start, b_start, arr_length);
9073 return true;
9074 }
9075
9076 //------------------------------inline_digestBase_implCompress-----------------------
9077 //
9078 // Calculate MD5 for single-block byte[] array.
9079 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9080 //
9081 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9082 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9083 //
9084 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9085 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9086 //
9087 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9088 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9089 //
9090 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9091 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9092 //
9093 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9094 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9095
9096 Node* digestBase_obj = argument(0);
9097 Node* src = argument(1); // type oop
9098 Node* ofs = argument(2); // type int
9099
9100 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9101 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9102 // failed array check
9103 return false;
9104 }
9105 // Figure out the size and type of the elements we will be copying.
9106 BasicType src_elem = src_type->elem()->array_element_basic_type();
9107 if (src_elem != T_BYTE) {
9108 return false;
9109 }
9110 // 'src_start' points to src array + offset
9111 src = must_be_not_null(src, true);
9112 Node* src_start = array_element_address(src, ofs, src_elem);
9113 Node* state = nullptr;
9114 Node* block_size = nullptr;
9115 address stubAddr;
9116 const char *stubName;
9117
9118 switch(id) {
9119 case vmIntrinsics::_md5_implCompress:
9120 assert(UseMD5Intrinsics, "need MD5 instruction support");
9121 state = get_state_from_digest_object(digestBase_obj, T_INT);
9122 stubAddr = StubRoutines::md5_implCompress();
9123 stubName = "md5_implCompress";
9124 break;
9125 case vmIntrinsics::_sha_implCompress:
9126 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9127 state = get_state_from_digest_object(digestBase_obj, T_INT);
9128 stubAddr = StubRoutines::sha1_implCompress();
9129 stubName = "sha1_implCompress";
9130 break;
9131 case vmIntrinsics::_sha2_implCompress:
9132 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9133 state = get_state_from_digest_object(digestBase_obj, T_INT);
9134 stubAddr = StubRoutines::sha256_implCompress();
9135 stubName = "sha256_implCompress";
9136 break;
9137 case vmIntrinsics::_sha5_implCompress:
9138 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9139 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9140 stubAddr = StubRoutines::sha512_implCompress();
9141 stubName = "sha512_implCompress";
9142 break;
9143 case vmIntrinsics::_sha3_implCompress:
9144 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9145 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9146 stubAddr = StubRoutines::sha3_implCompress();
9147 stubName = "sha3_implCompress";
9148 block_size = get_block_size_from_digest_object(digestBase_obj);
9149 if (block_size == nullptr) return false;
9150 break;
9151 default:
9152 fatal_unexpected_iid(id);
9153 return false;
9154 }
9155 if (state == nullptr) return false;
9156
9157 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9158 if (stubAddr == nullptr) return false;
9159
9160 // Call the stub.
9161 Node* call;
9162 if (block_size == nullptr) {
9163 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9164 stubAddr, stubName, TypePtr::BOTTOM,
9165 src_start, state);
9166 } else {
9167 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9168 stubAddr, stubName, TypePtr::BOTTOM,
9169 src_start, state, block_size);
9170 }
9171
9172 return true;
9173 }
9174
9175 //------------------------------inline_double_keccak
9176 bool LibraryCallKit::inline_double_keccak() {
9177 address stubAddr;
9178 const char *stubName;
9179 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9180 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9181
9182 stubAddr = StubRoutines::double_keccak();
9183 stubName = "double_keccak";
9184 if (!stubAddr) return false;
9185
9186 Node* status0 = argument(0);
9187 Node* status1 = argument(1);
9188
9189 status0 = must_be_not_null(status0, true);
9190 status1 = must_be_not_null(status1, true);
9191
9192 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9193 assert(status0_start, "status0 is null");
9194 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9195 assert(status1_start, "status1 is null");
9196 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9197 OptoRuntime::double_keccak_Type(),
9198 stubAddr, stubName, TypePtr::BOTTOM,
9199 status0_start, status1_start);
9200 // return an int
9201 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9202 set_result(retvalue);
9203 return true;
9204 }
9205
9206
9207 //------------------------------inline_digestBase_implCompressMB-----------------------
9208 //
9209 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9210 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9211 //
9212 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9213 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9214 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9215 assert((uint)predicate < 5, "sanity");
9216 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9217
9218 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9219 Node* src = argument(1); // byte[] array
9220 Node* ofs = argument(2); // type int
9221 Node* limit = argument(3); // type int
9222
9223 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9224 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9225 // failed array check
9226 return false;
9227 }
9228 // Figure out the size and type of the elements we will be copying.
9229 BasicType src_elem = src_type->elem()->array_element_basic_type();
9230 if (src_elem != T_BYTE) {
9231 return false;
9232 }
9233 // 'src_start' points to src array + offset
9234 src = must_be_not_null(src, false);
9235 Node* src_start = array_element_address(src, ofs, src_elem);
9236
9237 const char* klass_digestBase_name = nullptr;
9238 const char* stub_name = nullptr;
9239 address stub_addr = nullptr;
9240 BasicType elem_type = T_INT;
9241
9242 switch (predicate) {
9243 case 0:
9244 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9245 klass_digestBase_name = "sun/security/provider/MD5";
9246 stub_name = "md5_implCompressMB";
9247 stub_addr = StubRoutines::md5_implCompressMB();
9248 }
9249 break;
9250 case 1:
9251 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9252 klass_digestBase_name = "sun/security/provider/SHA";
9253 stub_name = "sha1_implCompressMB";
9254 stub_addr = StubRoutines::sha1_implCompressMB();
9255 }
9256 break;
9257 case 2:
9258 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9259 klass_digestBase_name = "sun/security/provider/SHA2";
9260 stub_name = "sha256_implCompressMB";
9261 stub_addr = StubRoutines::sha256_implCompressMB();
9262 }
9263 break;
9264 case 3:
9265 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9266 klass_digestBase_name = "sun/security/provider/SHA5";
9267 stub_name = "sha512_implCompressMB";
9268 stub_addr = StubRoutines::sha512_implCompressMB();
9269 elem_type = T_LONG;
9270 }
9271 break;
9272 case 4:
9273 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9274 klass_digestBase_name = "sun/security/provider/SHA3";
9275 stub_name = "sha3_implCompressMB";
9276 stub_addr = StubRoutines::sha3_implCompressMB();
9277 elem_type = T_LONG;
9278 }
9279 break;
9280 default:
9281 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9282 }
9283 if (klass_digestBase_name != nullptr) {
9284 assert(stub_addr != nullptr, "Stub is generated");
9285 if (stub_addr == nullptr) return false;
9286
9287 // get DigestBase klass to lookup for SHA klass
9288 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9289 assert(tinst != nullptr, "digestBase_obj is not instance???");
9290 assert(tinst->is_loaded(), "DigestBase is not loaded");
9291
9292 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9293 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9294 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9295 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9296 }
9297 return false;
9298 }
9299
9300 //------------------------------inline_digestBase_implCompressMB-----------------------
9301 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9302 BasicType elem_type, address stubAddr, const char *stubName,
9303 Node* src_start, Node* ofs, Node* limit) {
9304 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9305 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9306 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9307 digest_obj = _gvn.transform(digest_obj);
9308
9309 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9310 if (state == nullptr) return false;
9311
9312 Node* block_size = nullptr;
9313 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9314 block_size = get_block_size_from_digest_object(digest_obj);
9315 if (block_size == nullptr) return false;
9316 }
9317
9318 // Call the stub.
9319 Node* call;
9320 if (block_size == nullptr) {
9321 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9322 OptoRuntime::digestBase_implCompressMB_Type(false),
9323 stubAddr, stubName, TypePtr::BOTTOM,
9324 src_start, state, ofs, limit);
9325 } else {
9326 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9327 OptoRuntime::digestBase_implCompressMB_Type(true),
9328 stubAddr, stubName, TypePtr::BOTTOM,
9329 src_start, state, block_size, ofs, limit);
9330 }
9331
9332 // return ofs (int)
9333 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9334 set_result(result);
9335
9336 return true;
9337 }
9338
9339 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9340 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9341 assert(UseAES, "need AES instruction support");
9342 address stubAddr = nullptr;
9343 const char *stubName = nullptr;
9344 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9345 stubName = "galoisCounterMode_AESCrypt";
9346
9347 if (stubAddr == nullptr) return false;
9348
9349 Node* in = argument(0);
9350 Node* inOfs = argument(1);
9351 Node* len = argument(2);
9352 Node* ct = argument(3);
9353 Node* ctOfs = argument(4);
9354 Node* out = argument(5);
9355 Node* outOfs = argument(6);
9356 Node* gctr_object = argument(7);
9357 Node* ghash_object = argument(8);
9358
9359 // (1) in, ct and out are arrays.
9360 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9361 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9362 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9363 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9364 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9365 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9366
9367 // checks are the responsibility of the caller
9368 Node* in_start = in;
9369 Node* ct_start = ct;
9370 Node* out_start = out;
9371 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9372 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9373 in_start = array_element_address(in, inOfs, T_BYTE);
9374 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9375 out_start = array_element_address(out, outOfs, T_BYTE);
9376 }
9377
9378 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9379 // (because of the predicated logic executed earlier).
9380 // so we cast it here safely.
9381 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9382 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9383 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9384 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9385 Node* state = load_field_from_object(ghash_object, "state", "[J");
9386
9387 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9388 return false;
9389 }
9390 // cast it to what we know it will be at runtime
9391 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9392 assert(tinst != nullptr, "GCTR obj is null");
9393 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9394 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9395 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9396 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9397 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9398 const TypeOopPtr* xtype = aklass->as_instance_type();
9399 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9400 aescrypt_object = _gvn.transform(aescrypt_object);
9401 // we need to get the start of the aescrypt_object's expanded key array
9402 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
9403 if (k_start == nullptr) return false;
9404 // similarly, get the start address of the r vector
9405 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9406 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9407 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9408
9409
9410 // Call the stub, passing params
9411 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9412 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9413 stubAddr, stubName, TypePtr::BOTTOM,
9414 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9415
9416 // return cipher length (int)
9417 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9418 set_result(retvalue);
9419
9420 return true;
9421 }
9422
9423 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9424 // Return node representing slow path of predicate check.
9425 // the pseudo code we want to emulate with this predicate is:
9426 // for encryption:
9427 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9428 // for decryption:
9429 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9430 // note cipher==plain is more conservative than the original java code but that's OK
9431 //
9432
9433 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9434 // The receiver was checked for null already.
9435 Node* objGCTR = argument(7);
9436 // Load embeddedCipher field of GCTR object.
9437 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9438 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9439
9440 // get AESCrypt klass for instanceOf check
9441 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9442 // will have same classloader as CipherBlockChaining object
9443 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9444 assert(tinst != nullptr, "GCTR obj is null");
9445 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9446
9447 // we want to do an instanceof comparison against the AESCrypt class
9448 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9449 if (!klass_AESCrypt->is_loaded()) {
9450 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9451 Node* ctrl = control();
9452 set_control(top()); // no regular fast path
9453 return ctrl;
9454 }
9455
9456 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9457 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9458 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9459 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9460 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9461
9462 return instof_false; // even if it is null
9463 }
9464
9465 //------------------------------get_state_from_digest_object-----------------------
9466 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9467 const char* state_type;
9468 switch (elem_type) {
9469 case T_BYTE: state_type = "[B"; break;
9470 case T_INT: state_type = "[I"; break;
9471 case T_LONG: state_type = "[J"; break;
9472 default: ShouldNotReachHere();
9473 }
9474 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9475 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9476 if (digest_state == nullptr) return (Node *) nullptr;
9477
9478 // now have the array, need to get the start address of the state array
9479 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9480 return state;
9481 }
9482
9483 //------------------------------get_block_size_from_sha3_object----------------------------------
9484 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9485 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9486 assert (block_size != nullptr, "sanity");
9487 return block_size;
9488 }
9489
9490 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9491 // Return node representing slow path of predicate check.
9492 // the pseudo code we want to emulate with this predicate is:
9493 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9494 //
9495 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9496 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9497 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9498 assert((uint)predicate < 5, "sanity");
9499
9500 // The receiver was checked for null already.
9501 Node* digestBaseObj = argument(0);
9502
9503 // get DigestBase klass for instanceOf check
9504 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9505 assert(tinst != nullptr, "digestBaseObj is null");
9506 assert(tinst->is_loaded(), "DigestBase is not loaded");
9507
9508 const char* klass_name = nullptr;
9509 switch (predicate) {
9510 case 0:
9511 if (UseMD5Intrinsics) {
9512 // we want to do an instanceof comparison against the MD5 class
9513 klass_name = "sun/security/provider/MD5";
9514 }
9515 break;
9516 case 1:
9517 if (UseSHA1Intrinsics) {
9518 // we want to do an instanceof comparison against the SHA class
9519 klass_name = "sun/security/provider/SHA";
9520 }
9521 break;
9522 case 2:
9523 if (UseSHA256Intrinsics) {
9524 // we want to do an instanceof comparison against the SHA2 class
9525 klass_name = "sun/security/provider/SHA2";
9526 }
9527 break;
9528 case 3:
9529 if (UseSHA512Intrinsics) {
9530 // we want to do an instanceof comparison against the SHA5 class
9531 klass_name = "sun/security/provider/SHA5";
9532 }
9533 break;
9534 case 4:
9535 if (UseSHA3Intrinsics) {
9536 // we want to do an instanceof comparison against the SHA3 class
9537 klass_name = "sun/security/provider/SHA3";
9538 }
9539 break;
9540 default:
9541 fatal("unknown SHA intrinsic predicate: %d", predicate);
9542 }
9543
9544 ciKlass* klass = nullptr;
9545 if (klass_name != nullptr) {
9546 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9547 }
9548 if ((klass == nullptr) || !klass->is_loaded()) {
9549 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9550 Node* ctrl = control();
9551 set_control(top()); // no intrinsic path
9552 return ctrl;
9553 }
9554 ciInstanceKlass* instklass = klass->as_instance_klass();
9555
9556 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9557 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9558 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9559 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9560
9561 return instof_false; // even if it is null
9562 }
9563
9564 //-------------inline_fma-----------------------------------
9565 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9566 Node *a = nullptr;
9567 Node *b = nullptr;
9568 Node *c = nullptr;
9569 Node* result = nullptr;
9570 switch (id) {
9571 case vmIntrinsics::_fmaD:
9572 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9573 // no receiver since it is static method
9574 a = argument(0);
9575 b = argument(2);
9576 c = argument(4);
9577 result = _gvn.transform(new FmaDNode(a, b, c));
9578 break;
9579 case vmIntrinsics::_fmaF:
9580 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9581 a = argument(0);
9582 b = argument(1);
9583 c = argument(2);
9584 result = _gvn.transform(new FmaFNode(a, b, c));
9585 break;
9586 default:
9587 fatal_unexpected_iid(id); break;
9588 }
9589 set_result(result);
9590 return true;
9591 }
9592
9593 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9594 // argument(0) is receiver
9595 Node* codePoint = argument(1);
9596 Node* n = nullptr;
9597
9598 switch (id) {
9599 case vmIntrinsics::_isDigit :
9600 n = new DigitNode(control(), codePoint);
9601 break;
9602 case vmIntrinsics::_isLowerCase :
9603 n = new LowerCaseNode(control(), codePoint);
9604 break;
9605 case vmIntrinsics::_isUpperCase :
9606 n = new UpperCaseNode(control(), codePoint);
9607 break;
9608 case vmIntrinsics::_isWhitespace :
9609 n = new WhitespaceNode(control(), codePoint);
9610 break;
9611 default:
9612 fatal_unexpected_iid(id);
9613 }
9614
9615 set_result(_gvn.transform(n));
9616 return true;
9617 }
9618
9619 bool LibraryCallKit::inline_profileBoolean() {
9620 Node* counts = argument(1);
9621 const TypeAryPtr* ary = nullptr;
9622 ciArray* aobj = nullptr;
9623 if (counts->is_Con()
9624 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9625 && (aobj = ary->const_oop()->as_array()) != nullptr
9626 && (aobj->length() == 2)) {
9627 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9628 jint false_cnt = aobj->element_value(0).as_int();
9629 jint true_cnt = aobj->element_value(1).as_int();
9630
9631 if (C->log() != nullptr) {
9632 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9633 false_cnt, true_cnt);
9634 }
9635
9636 if (false_cnt + true_cnt == 0) {
9637 // According to profile, never executed.
9638 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9639 Deoptimization::Action_reinterpret);
9640 return true;
9641 }
9642
9643 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9644 // is a number of each value occurrences.
9645 Node* result = argument(0);
9646 if (false_cnt == 0 || true_cnt == 0) {
9647 // According to profile, one value has been never seen.
9648 int expected_val = (false_cnt == 0) ? 1 : 0;
9649
9650 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9651 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9652
9653 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9654 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9655 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9656
9657 { // Slow path: uncommon trap for never seen value and then reexecute
9658 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9659 // the value has been seen at least once.
9660 PreserveJVMState pjvms(this);
9661 PreserveReexecuteState preexecs(this);
9662 jvms()->set_should_reexecute(true);
9663
9664 set_control(slow_path);
9665 set_i_o(i_o());
9666
9667 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9668 Deoptimization::Action_reinterpret);
9669 }
9670 // The guard for never seen value enables sharpening of the result and
9671 // returning a constant. It allows to eliminate branches on the same value
9672 // later on.
9673 set_control(fast_path);
9674 result = intcon(expected_val);
9675 }
9676 // Stop profiling.
9677 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9678 // By replacing method body with profile data (represented as ProfileBooleanNode
9679 // on IR level) we effectively disable profiling.
9680 // It enables full speed execution once optimized code is generated.
9681 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9682 C->record_for_igvn(profile);
9683 set_result(profile);
9684 return true;
9685 } else {
9686 // Continue profiling.
9687 // Profile data isn't available at the moment. So, execute method's bytecode version.
9688 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9689 // is compiled and counters aren't available since corresponding MethodHandle
9690 // isn't a compile-time constant.
9691 return false;
9692 }
9693 }
9694
9695 bool LibraryCallKit::inline_isCompileConstant() {
9696 Node* n = argument(0);
9697 set_result(n->is_Con() ? intcon(1) : intcon(0));
9698 return true;
9699 }
9700
9701 //------------------------------- inline_getObjectSize --------------------------------------
9702 //
9703 // Calculate the runtime size of the object/array.
9704 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9705 //
9706 bool LibraryCallKit::inline_getObjectSize() {
9707 Node* obj = argument(3);
9708 Node* klass_node = load_object_klass(obj);
9709
9710 jint layout_con = Klass::_lh_neutral_value;
9711 Node* layout_val = get_layout_helper(klass_node, layout_con);
9712 int layout_is_con = (layout_val == nullptr);
9713
9714 if (layout_is_con) {
9715 // Layout helper is constant, can figure out things at compile time.
9716
9717 if (Klass::layout_helper_is_instance(layout_con)) {
9718 // Instance case: layout_con contains the size itself.
9719 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9720 set_result(size);
9721 } else {
9722 // Array case: size is round(header + element_size*arraylength).
9723 // Since arraylength is different for every array instance, we have to
9724 // compute the whole thing at runtime.
9725
9726 Node* arr_length = load_array_length(obj);
9727
9728 int round_mask = MinObjAlignmentInBytes - 1;
9729 int hsize = Klass::layout_helper_header_size(layout_con);
9730 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9731
9732 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9733 round_mask = 0; // strength-reduce it if it goes away completely
9734 }
9735 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9736 Node* header_size = intcon(hsize + round_mask);
9737
9738 Node* lengthx = ConvI2X(arr_length);
9739 Node* headerx = ConvI2X(header_size);
9740
9741 Node* abody = lengthx;
9742 if (eshift != 0) {
9743 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9744 }
9745 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9746 if (round_mask != 0) {
9747 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9748 }
9749 size = ConvX2L(size);
9750 set_result(size);
9751 }
9752 } else {
9753 // Layout helper is not constant, need to test for array-ness at runtime.
9754
9755 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9756 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9757 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9758 record_for_igvn(result_reg);
9759
9760 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9761 if (array_ctl != nullptr) {
9762 // Array case: size is round(header + element_size*arraylength).
9763 // Since arraylength is different for every array instance, we have to
9764 // compute the whole thing at runtime.
9765
9766 PreserveJVMState pjvms(this);
9767 set_control(array_ctl);
9768 Node* arr_length = load_array_length(obj);
9769
9770 int round_mask = MinObjAlignmentInBytes - 1;
9771 Node* mask = intcon(round_mask);
9772
9773 Node* hss = intcon(Klass::_lh_header_size_shift);
9774 Node* hsm = intcon(Klass::_lh_header_size_mask);
9775 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9776 header_size = _gvn.transform(new AndINode(header_size, hsm));
9777 header_size = _gvn.transform(new AddINode(header_size, mask));
9778
9779 // There is no need to mask or shift this value.
9780 // The semantics of LShiftINode include an implicit mask to 0x1F.
9781 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9782 Node* elem_shift = layout_val;
9783
9784 Node* lengthx = ConvI2X(arr_length);
9785 Node* headerx = ConvI2X(header_size);
9786
9787 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9788 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9789 if (round_mask != 0) {
9790 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9791 }
9792 size = ConvX2L(size);
9793
9794 result_reg->init_req(_array_path, control());
9795 result_val->init_req(_array_path, size);
9796 }
9797
9798 if (!stopped()) {
9799 // Instance case: the layout helper gives us instance size almost directly,
9800 // but we need to mask out the _lh_instance_slow_path_bit.
9801 Node* size = ConvI2X(layout_val);
9802 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9803 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9804 size = _gvn.transform(new AndXNode(size, mask));
9805 size = ConvX2L(size);
9806
9807 result_reg->init_req(_instance_path, control());
9808 result_val->init_req(_instance_path, size);
9809 }
9810
9811 set_result(result_reg, result_val);
9812 }
9813
9814 return true;
9815 }
9816
9817 //------------------------------- inline_blackhole --------------------------------------
9818 //
9819 // Make sure all arguments to this node are alive.
9820 // This matches methods that were requested to be blackholed through compile commands.
9821 //
9822 bool LibraryCallKit::inline_blackhole() {
9823 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9824 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9825 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9826
9827 // Blackhole node pinches only the control, not memory. This allows
9828 // the blackhole to be pinned in the loop that computes blackholed
9829 // values, but have no other side effects, like breaking the optimizations
9830 // across the blackhole.
9831
9832 Node* bh = _gvn.transform(new BlackholeNode(control()));
9833 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9834
9835 // Bind call arguments as blackhole arguments to keep them alive
9836 uint nargs = callee()->arg_size();
9837 for (uint i = 0; i < nargs; i++) {
9838 bh->add_req(argument(i));
9839 }
9840
9841 return true;
9842 }
9843
9844 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9845 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9846 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9847 return nullptr; // box klass is not Float16
9848 }
9849
9850 // Null check; get notnull casted pointer
9851 Node* null_ctl = top();
9852 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9853 // If not_null_box is dead, only null-path is taken
9854 if (stopped()) {
9855 set_control(null_ctl);
9856 return nullptr;
9857 }
9858 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9859 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9860 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9861 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9862 }
9863
9864 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9865 PreserveReexecuteState preexecs(this);
9866 jvms()->set_should_reexecute(true);
9867
9868 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9869 Node* klass_node = makecon(klass_type);
9870 Node* box = new_instance(klass_node);
9871
9872 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9873 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9874
9875 Node* field_store = _gvn.transform(access_store_at(box,
9876 value_field,
9877 value_adr_type,
9878 value,
9879 TypeInt::SHORT,
9880 T_SHORT,
9881 IN_HEAP));
9882 set_memory(field_store, value_adr_type);
9883 return box;
9884 }
9885
9886 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9887 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9888 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9889 return false;
9890 }
9891
9892 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9893 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9894 return false;
9895 }
9896
9897 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9898 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9899 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9900 ciSymbols::short_signature(),
9901 false);
9902 assert(field != nullptr, "");
9903
9904 // Transformed nodes
9905 Node* fld1 = nullptr;
9906 Node* fld2 = nullptr;
9907 Node* fld3 = nullptr;
9908 switch(num_args) {
9909 case 3:
9910 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9911 if (fld3 == nullptr) {
9912 return false;
9913 }
9914 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9915 // fall-through
9916 case 2:
9917 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9918 if (fld2 == nullptr) {
9919 return false;
9920 }
9921 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9922 // fall-through
9923 case 1:
9924 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9925 if (fld1 == nullptr) {
9926 return false;
9927 }
9928 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9929 break;
9930 default: fatal("Unsupported number of arguments %d", num_args);
9931 }
9932
9933 Node* result = nullptr;
9934 switch (id) {
9935 // Unary operations
9936 case vmIntrinsics::_sqrt_float16:
9937 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9938 break;
9939 // Ternary operations
9940 case vmIntrinsics::_fma_float16:
9941 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9942 break;
9943 default:
9944 fatal_unexpected_iid(id);
9945 break;
9946 }
9947 result = _gvn.transform(new ReinterpretHF2SNode(result));
9948 set_result(box_fp16_value(float16_box_type, field, result));
9949 return true;
9950 }
9951