1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/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/jniHandles.inline.hpp"
68 #include "runtime/objectMonitor.hpp"
69 #include "runtime/sharedRuntime.hpp"
70 #include "runtime/stubRoutines.hpp"
71 #include "utilities/globalDefinitions.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/powerOfTwo.hpp"
74
75 //---------------------------make_vm_intrinsic----------------------------
76 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
77 vmIntrinsicID id = m->intrinsic_id();
78 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
79
80 if (!m->is_loaded()) {
81 // Do not attempt to inline unloaded methods.
82 return nullptr;
83 }
84
85 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
86 bool is_available = false;
87
88 {
89 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
90 // the compiler must transition to '_thread_in_vm' state because both
91 // methods access VM-internal data.
92 VM_ENTRY_MARK;
93 methodHandle mh(THREAD, m->get_Method());
94 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
95 if (is_available && is_virtual) {
96 is_available = vmIntrinsics::does_virtual_dispatch(id);
97 }
98 }
99
100 if (is_available) {
101 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
102 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
103 return new LibraryIntrinsic(m, is_virtual,
104 vmIntrinsics::predicates_needed(id),
105 vmIntrinsics::does_virtual_dispatch(id),
106 id);
107 } else {
108 return nullptr;
109 }
110 }
111
112 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
113 LibraryCallKit kit(jvms, this);
114 Compile* C = kit.C;
115 int nodes = C->unique();
116 #ifndef PRODUCT
117 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
118 char buf[1000];
119 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
120 tty->print_cr("Intrinsic %s", str);
121 }
122 #endif
123 ciMethod* callee = kit.callee();
124 const int bci = kit.bci();
125 #ifdef ASSERT
126 Node* ctrl = kit.control();
127 #endif
128 // Try to inline the intrinsic.
129 if (callee->check_intrinsic_candidate() &&
130 kit.try_to_inline(_last_predicate)) {
131 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
132 : "(intrinsic)";
133 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
134 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
135 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
136 if (C->log()) {
137 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
138 vmIntrinsics::name_at(intrinsic_id()),
139 (is_virtual() ? " virtual='1'" : ""),
140 C->unique() - nodes);
141 }
142 // Push the result from the inlined method onto the stack.
143 kit.push_result();
144 return kit.transfer_exceptions_into_jvms();
145 }
146
147 // The intrinsic bailed out
148 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
149 if (jvms->has_method()) {
150 // Not a root compile.
151 const char* msg;
152 if (callee->intrinsic_candidate()) {
153 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
154 } else {
155 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
156 : "failed to inline (intrinsic), method not annotated";
157 }
158 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
159 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
160 } else {
161 // Root compile
162 ResourceMark rm;
163 stringStream msg_stream;
164 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
165 vmIntrinsics::name_at(intrinsic_id()),
166 is_virtual() ? " (virtual)" : "", bci);
167 const char *msg = msg_stream.freeze();
168 log_debug(jit, inlining)("%s", msg);
169 if (C->print_intrinsics() || C->print_inlining()) {
170 tty->print("%s", msg);
171 }
172 }
173 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
174
175 return nullptr;
176 }
177
178 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
179 LibraryCallKit kit(jvms, this);
180 Compile* C = kit.C;
181 int nodes = C->unique();
182 _last_predicate = predicate;
183 #ifndef PRODUCT
184 assert(is_predicated() && predicate < predicates_count(), "sanity");
185 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
186 char buf[1000];
187 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
188 tty->print_cr("Predicate for intrinsic %s", str);
189 }
190 #endif
191 ciMethod* callee = kit.callee();
192 const int bci = kit.bci();
193
194 Node* slow_ctl = kit.try_to_predicate(predicate);
195 if (!kit.failing()) {
196 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
197 : "(intrinsic, predicate)";
198 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
199 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
200
201 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
202 if (C->log()) {
203 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
204 vmIntrinsics::name_at(intrinsic_id()),
205 (is_virtual() ? " virtual='1'" : ""),
206 C->unique() - nodes);
207 }
208 return slow_ctl; // Could be null if the check folds.
209 }
210
211 // The intrinsic bailed out
212 if (jvms->has_method()) {
213 // Not a root compile.
214 const char* msg = "failed to generate predicate for intrinsic";
215 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
216 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
217 } else {
218 // Root compile
219 ResourceMark rm;
220 stringStream msg_stream;
221 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
222 vmIntrinsics::name_at(intrinsic_id()),
223 is_virtual() ? " (virtual)" : "", bci);
224 const char *msg = msg_stream.freeze();
225 log_debug(jit, inlining)("%s", msg);
226 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
227 }
228 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
229 return nullptr;
230 }
231
232 bool LibraryCallKit::try_to_inline(int predicate) {
233 // Handle symbolic names for otherwise undistinguished boolean switches:
234 const bool is_store = true;
235 const bool is_compress = true;
236 const bool is_static = true;
237 const bool is_volatile = true;
238
239 if (!jvms()->has_method()) {
240 // Root JVMState has a null method.
241 assert(map()->memory()->Opcode() == Op_Parm, "");
242 // Insert the memory aliasing node
243 set_all_memory(reset_memory());
244 }
245 assert(merged_memory(), "");
246
247 switch (intrinsic_id()) {
248 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
249 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
250 case vmIntrinsics::_getClass: return inline_native_getClass();
251
252 case vmIntrinsics::_ceil:
253 case vmIntrinsics::_floor:
254 case vmIntrinsics::_rint:
255 case vmIntrinsics::_dsin:
256 case vmIntrinsics::_dcos:
257 case vmIntrinsics::_dtan:
258 case vmIntrinsics::_dtanh:
259 case vmIntrinsics::_dcbrt:
260 case vmIntrinsics::_dabs:
261 case vmIntrinsics::_fabs:
262 case vmIntrinsics::_iabs:
263 case vmIntrinsics::_labs:
264 case vmIntrinsics::_datan2:
265 case vmIntrinsics::_dsqrt:
266 case vmIntrinsics::_dsqrt_strict:
267 case vmIntrinsics::_dexp:
268 case vmIntrinsics::_dlog:
269 case vmIntrinsics::_dlog10:
270 case vmIntrinsics::_dpow:
271 case vmIntrinsics::_dcopySign:
272 case vmIntrinsics::_fcopySign:
273 case vmIntrinsics::_dsignum:
274 case vmIntrinsics::_roundF:
275 case vmIntrinsics::_roundD:
276 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
277
278 case vmIntrinsics::_notify:
279 case vmIntrinsics::_notifyAll:
280 return inline_notify(intrinsic_id());
281
282 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
283 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
284 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
285 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
286 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
287 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
288 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
289 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
290 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
291 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
292 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
293 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
294 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
295 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
296
297 case vmIntrinsics::_arraycopy: return inline_arraycopy();
298
299 case vmIntrinsics::_arraySort: return inline_array_sort();
300 case vmIntrinsics::_arrayPartition: return inline_array_partition();
301
302 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
303 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
304 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
305 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
306
307 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
308 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
309 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
310 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
311 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
312 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
313 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
314 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
315
316 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
317
318 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
319
320 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
321 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
322 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
323 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
324
325 case vmIntrinsics::_compressStringC:
326 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
327 case vmIntrinsics::_inflateStringC:
328 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
329
330 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
331 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
332 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
341 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
342
343 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
344 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
345 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
346 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
347 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
348 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
349 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
350 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
351 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
352 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
353
354 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
355 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
356 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
357 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
358 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
359 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
360 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
361 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
362 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
363
364 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
365 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
366 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
367 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
368 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
369 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
370 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
371 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
372 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
373
374 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
375 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
376 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
377 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
378
379 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
380 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
381 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
382 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
383
384 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
385 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
386 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
387 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
388 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
389 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
390 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
391 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
392 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
393
394 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
395 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
396 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
397 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
398 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
399 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
400 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
401 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
402 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
403
404 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
405 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
406 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
407 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
408 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
409 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
410 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
411 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
412 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
413
414 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
415 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
416 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
417 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
418 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
419 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
420 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
421 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
422 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
423
424 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
425 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
426
427 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
428 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
429 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
432
433 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
445 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
446 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
447 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
448 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
449 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
450 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
451 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
452 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
453
454 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
460 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
461 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
462 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
463 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
464 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
465 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
466 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
467 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
468 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
469
470 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
471 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
472 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
474
475 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
476 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
477 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
480
481 case vmIntrinsics::_loadFence:
482 case vmIntrinsics::_storeFence:
483 case vmIntrinsics::_storeStoreFence:
484 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
485
486 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
487
488 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
489 case vmIntrinsics::_currentThread: return inline_native_currentThread();
490 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
491
492 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
493 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
494
495 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
496 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
497
498 #if INCLUDE_JVMTI
499 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
500 "notifyJvmtiStart", true, false);
501 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
502 "notifyJvmtiEnd", false, true);
503 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
504 "notifyJvmtiMount", false, false);
505 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
506 "notifyJvmtiUnmount", false, false);
507 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
508 #endif
509
510 #ifdef JFR_HAVE_INTRINSICS
511 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
512 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
513 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
514 #endif
515 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
516 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
517 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
518 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
519 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
520 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
521 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
522 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
523 case vmIntrinsics::_getLength: return inline_native_getLength();
524 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
525 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
526 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
527 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
528 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
529 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
530 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
531
532 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
533 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
534 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
535 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
536 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
537
538 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
539
540 case vmIntrinsics::_isInstance:
541 case vmIntrinsics::_isHidden:
542 case vmIntrinsics::_getSuperclass:
543 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
544
545 case vmIntrinsics::_floatToRawIntBits:
546 case vmIntrinsics::_floatToIntBits:
547 case vmIntrinsics::_intBitsToFloat:
548 case vmIntrinsics::_doubleToRawLongBits:
549 case vmIntrinsics::_doubleToLongBits:
550 case vmIntrinsics::_longBitsToDouble:
551 case vmIntrinsics::_floatToFloat16:
552 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
553 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
554 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
555 case vmIntrinsics::_floatIsFinite:
556 case vmIntrinsics::_floatIsInfinite:
557 case vmIntrinsics::_doubleIsFinite:
558 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
559
560 case vmIntrinsics::_numberOfLeadingZeros_i:
561 case vmIntrinsics::_numberOfLeadingZeros_l:
562 case vmIntrinsics::_numberOfTrailingZeros_i:
563 case vmIntrinsics::_numberOfTrailingZeros_l:
564 case vmIntrinsics::_bitCount_i:
565 case vmIntrinsics::_bitCount_l:
566 case vmIntrinsics::_reverse_i:
567 case vmIntrinsics::_reverse_l:
568 case vmIntrinsics::_reverseBytes_i:
569 case vmIntrinsics::_reverseBytes_l:
570 case vmIntrinsics::_reverseBytes_s:
571 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
572
573 case vmIntrinsics::_compress_i:
574 case vmIntrinsics::_compress_l:
575 case vmIntrinsics::_expand_i:
576 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
577
578 case vmIntrinsics::_compareUnsigned_i:
579 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
580
581 case vmIntrinsics::_divideUnsigned_i:
582 case vmIntrinsics::_divideUnsigned_l:
583 case vmIntrinsics::_remainderUnsigned_i:
584 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
585
586 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
587
588 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
589 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
590 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
591 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
592 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
593
594 case vmIntrinsics::_Class_cast: return inline_Class_cast();
595
596 case vmIntrinsics::_aescrypt_encryptBlock:
597 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
598
599 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
600 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
601 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
602
603 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
604 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
605 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
606
607 case vmIntrinsics::_counterMode_AESCrypt:
608 return inline_counterMode_AESCrypt(intrinsic_id());
609
610 case vmIntrinsics::_galoisCounterMode_AESCrypt:
611 return inline_galoisCounterMode_AESCrypt();
612
613 case vmIntrinsics::_md5_implCompress:
614 case vmIntrinsics::_sha_implCompress:
615 case vmIntrinsics::_sha2_implCompress:
616 case vmIntrinsics::_sha5_implCompress:
617 case vmIntrinsics::_sha3_implCompress:
618 return inline_digestBase_implCompress(intrinsic_id());
619 case vmIntrinsics::_double_keccak:
620 return inline_double_keccak();
621
622 case vmIntrinsics::_digestBase_implCompressMB:
623 return inline_digestBase_implCompressMB(predicate);
624
625 case vmIntrinsics::_multiplyToLen:
626 return inline_multiplyToLen();
627
628 case vmIntrinsics::_squareToLen:
629 return inline_squareToLen();
630
631 case vmIntrinsics::_mulAdd:
632 return inline_mulAdd();
633
634 case vmIntrinsics::_montgomeryMultiply:
635 return inline_montgomeryMultiply();
636 case vmIntrinsics::_montgomerySquare:
637 return inline_montgomerySquare();
638
639 case vmIntrinsics::_bigIntegerRightShiftWorker:
640 return inline_bigIntegerShift(true);
641 case vmIntrinsics::_bigIntegerLeftShiftWorker:
642 return inline_bigIntegerShift(false);
643
644 case vmIntrinsics::_vectorizedMismatch:
645 return inline_vectorizedMismatch();
646
647 case vmIntrinsics::_ghash_processBlocks:
648 return inline_ghash_processBlocks();
649 case vmIntrinsics::_chacha20Block:
650 return inline_chacha20Block();
651 case vmIntrinsics::_kyberNtt:
652 return inline_kyberNtt();
653 case vmIntrinsics::_kyberInverseNtt:
654 return inline_kyberInverseNtt();
655 case vmIntrinsics::_kyberNttMult:
656 return inline_kyberNttMult();
657 case vmIntrinsics::_kyberAddPoly_2:
658 return inline_kyberAddPoly_2();
659 case vmIntrinsics::_kyberAddPoly_3:
660 return inline_kyberAddPoly_3();
661 case vmIntrinsics::_kyber12To16:
662 return inline_kyber12To16();
663 case vmIntrinsics::_kyberBarrettReduce:
664 return inline_kyberBarrettReduce();
665 case vmIntrinsics::_dilithiumAlmostNtt:
666 return inline_dilithiumAlmostNtt();
667 case vmIntrinsics::_dilithiumAlmostInverseNtt:
668 return inline_dilithiumAlmostInverseNtt();
669 case vmIntrinsics::_dilithiumNttMult:
670 return inline_dilithiumNttMult();
671 case vmIntrinsics::_dilithiumMontMulByConstant:
672 return inline_dilithiumMontMulByConstant();
673 case vmIntrinsics::_dilithiumDecomposePoly:
674 return inline_dilithiumDecomposePoly();
675 case vmIntrinsics::_base64_encodeBlock:
676 return inline_base64_encodeBlock();
677 case vmIntrinsics::_base64_decodeBlock:
678 return inline_base64_decodeBlock();
679 case vmIntrinsics::_poly1305_processBlocks:
680 return inline_poly1305_processBlocks();
681 case vmIntrinsics::_intpoly_montgomeryMult_P256:
682 return inline_intpoly_montgomeryMult_P256();
683 case vmIntrinsics::_intpoly_assign:
684 return inline_intpoly_assign();
685 case vmIntrinsics::_encodeISOArray:
686 case vmIntrinsics::_encodeByteISOArray:
687 return inline_encodeISOArray(false);
688 case vmIntrinsics::_encodeAsciiArray:
689 return inline_encodeISOArray(true);
690
691 case vmIntrinsics::_updateCRC32:
692 return inline_updateCRC32();
693 case vmIntrinsics::_updateBytesCRC32:
694 return inline_updateBytesCRC32();
695 case vmIntrinsics::_updateByteBufferCRC32:
696 return inline_updateByteBufferCRC32();
697
698 case vmIntrinsics::_updateBytesCRC32C:
699 return inline_updateBytesCRC32C();
700 case vmIntrinsics::_updateDirectByteBufferCRC32C:
701 return inline_updateDirectByteBufferCRC32C();
702
703 case vmIntrinsics::_updateBytesAdler32:
704 return inline_updateBytesAdler32();
705 case vmIntrinsics::_updateByteBufferAdler32:
706 return inline_updateByteBufferAdler32();
707
708 case vmIntrinsics::_profileBoolean:
709 return inline_profileBoolean();
710 case vmIntrinsics::_isCompileConstant:
711 return inline_isCompileConstant();
712
713 case vmIntrinsics::_countPositives:
714 return inline_countPositives();
715
716 case vmIntrinsics::_fmaD:
717 case vmIntrinsics::_fmaF:
718 return inline_fma(intrinsic_id());
719
720 case vmIntrinsics::_isDigit:
721 case vmIntrinsics::_isLowerCase:
722 case vmIntrinsics::_isUpperCase:
723 case vmIntrinsics::_isWhitespace:
724 return inline_character_compare(intrinsic_id());
725
726 case vmIntrinsics::_min:
727 case vmIntrinsics::_max:
728 case vmIntrinsics::_min_strict:
729 case vmIntrinsics::_max_strict:
730 case vmIntrinsics::_minL:
731 case vmIntrinsics::_maxL:
732 case vmIntrinsics::_minF:
733 case vmIntrinsics::_maxF:
734 case vmIntrinsics::_minD:
735 case vmIntrinsics::_maxD:
736 case vmIntrinsics::_minF_strict:
737 case vmIntrinsics::_maxF_strict:
738 case vmIntrinsics::_minD_strict:
739 case vmIntrinsics::_maxD_strict:
740 return inline_min_max(intrinsic_id());
741
742 case vmIntrinsics::_VectorUnaryOp:
743 return inline_vector_nary_operation(1);
744 case vmIntrinsics::_VectorBinaryOp:
745 return inline_vector_nary_operation(2);
746 case vmIntrinsics::_VectorUnaryLibOp:
747 return inline_vector_call(1);
748 case vmIntrinsics::_VectorBinaryLibOp:
749 return inline_vector_call(2);
750 case vmIntrinsics::_VectorTernaryOp:
751 return inline_vector_nary_operation(3);
752 case vmIntrinsics::_VectorFromBitsCoerced:
753 return inline_vector_frombits_coerced();
754 case vmIntrinsics::_VectorMaskOp:
755 return inline_vector_mask_operation();
756 case vmIntrinsics::_VectorLoadOp:
757 return inline_vector_mem_operation(/*is_store=*/false);
758 case vmIntrinsics::_VectorLoadMaskedOp:
759 return inline_vector_mem_masked_operation(/*is_store*/false);
760 case vmIntrinsics::_VectorStoreOp:
761 return inline_vector_mem_operation(/*is_store=*/true);
762 case vmIntrinsics::_VectorStoreMaskedOp:
763 return inline_vector_mem_masked_operation(/*is_store=*/true);
764 case vmIntrinsics::_VectorGatherOp:
765 return inline_vector_gather_scatter(/*is_scatter*/ false);
766 case vmIntrinsics::_VectorScatterOp:
767 return inline_vector_gather_scatter(/*is_scatter*/ true);
768 case vmIntrinsics::_VectorReductionCoerced:
769 return inline_vector_reduction();
770 case vmIntrinsics::_VectorTest:
771 return inline_vector_test();
772 case vmIntrinsics::_VectorBlend:
773 return inline_vector_blend();
774 case vmIntrinsics::_VectorRearrange:
775 return inline_vector_rearrange();
776 case vmIntrinsics::_VectorSelectFrom:
777 return inline_vector_select_from();
778 case vmIntrinsics::_VectorCompare:
779 return inline_vector_compare();
780 case vmIntrinsics::_VectorBroadcastInt:
781 return inline_vector_broadcast_int();
782 case vmIntrinsics::_VectorConvert:
783 return inline_vector_convert();
784 case vmIntrinsics::_VectorInsert:
785 return inline_vector_insert();
786 case vmIntrinsics::_VectorExtract:
787 return inline_vector_extract();
788 case vmIntrinsics::_VectorCompressExpand:
789 return inline_vector_compress_expand();
790 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
791 return inline_vector_select_from_two_vectors();
792 case vmIntrinsics::_IndexVector:
793 return inline_index_vector();
794 case vmIntrinsics::_IndexPartiallyInUpperRange:
795 return inline_index_partially_in_upper_range();
796
797 case vmIntrinsics::_getObjectSize:
798 return inline_getObjectSize();
799
800 case vmIntrinsics::_blackhole:
801 return inline_blackhole();
802
803 default:
804 // If you get here, it may be that someone has added a new intrinsic
805 // to the list in vmIntrinsics.hpp without implementing it here.
806 #ifndef PRODUCT
807 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
808 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
809 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
810 }
811 #endif
812 return false;
813 }
814 }
815
816 Node* LibraryCallKit::try_to_predicate(int predicate) {
817 if (!jvms()->has_method()) {
818 // Root JVMState has a null method.
819 assert(map()->memory()->Opcode() == Op_Parm, "");
820 // Insert the memory aliasing node
821 set_all_memory(reset_memory());
822 }
823 assert(merged_memory(), "");
824
825 switch (intrinsic_id()) {
826 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
827 return inline_cipherBlockChaining_AESCrypt_predicate(false);
828 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
829 return inline_cipherBlockChaining_AESCrypt_predicate(true);
830 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
831 return inline_electronicCodeBook_AESCrypt_predicate(false);
832 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
833 return inline_electronicCodeBook_AESCrypt_predicate(true);
834 case vmIntrinsics::_counterMode_AESCrypt:
835 return inline_counterMode_AESCrypt_predicate();
836 case vmIntrinsics::_digestBase_implCompressMB:
837 return inline_digestBase_implCompressMB_predicate(predicate);
838 case vmIntrinsics::_galoisCounterMode_AESCrypt:
839 return inline_galoisCounterMode_AESCrypt_predicate();
840
841 default:
842 // If you get here, it may be that someone has added a new intrinsic
843 // to the list in vmIntrinsics.hpp without implementing it here.
844 #ifndef PRODUCT
845 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
846 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
847 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
848 }
849 #endif
850 Node* slow_ctl = control();
851 set_control(top()); // No fast path intrinsic
852 return slow_ctl;
853 }
854 }
855
856 //------------------------------set_result-------------------------------
857 // Helper function for finishing intrinsics.
858 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
859 record_for_igvn(region);
860 set_control(_gvn.transform(region));
861 set_result( _gvn.transform(value));
862 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
863 }
864
865 //------------------------------generate_guard---------------------------
866 // Helper function for generating guarded fast-slow graph structures.
867 // The given 'test', if true, guards a slow path. If the test fails
868 // then a fast path can be taken. (We generally hope it fails.)
869 // In all cases, GraphKit::control() is updated to the fast path.
870 // The returned value represents the control for the slow path.
871 // The return value is never 'top'; it is either a valid control
872 // or null if it is obvious that the slow path can never be taken.
873 // Also, if region and the slow control are not null, the slow edge
874 // is appended to the region.
875 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
876 if (stopped()) {
877 // Already short circuited.
878 return nullptr;
879 }
880
881 // Build an if node and its projections.
882 // If test is true we take the slow path, which we assume is uncommon.
883 if (_gvn.type(test) == TypeInt::ZERO) {
884 // The slow branch is never taken. No need to build this guard.
885 return nullptr;
886 }
887
888 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
889
890 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
891 if (if_slow == top()) {
892 // The slow branch is never taken. No need to build this guard.
893 return nullptr;
894 }
895
896 if (region != nullptr)
897 region->add_req(if_slow);
898
899 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
900 set_control(if_fast);
901
902 return if_slow;
903 }
904
905 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
906 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
907 }
908 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
909 return generate_guard(test, region, PROB_FAIR);
910 }
911
912 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
913 Node* *pos_index) {
914 if (stopped())
915 return nullptr; // already stopped
916 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
917 return nullptr; // index is already adequately typed
918 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
919 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
920 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
921 if (is_neg != nullptr && pos_index != nullptr) {
922 // Emulate effect of Parse::adjust_map_after_if.
923 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
924 (*pos_index) = _gvn.transform(ccast);
925 }
926 return is_neg;
927 }
928
929 // Make sure that 'position' is a valid limit index, in [0..length].
930 // There are two equivalent plans for checking this:
931 // A. (offset + copyLength) unsigned<= arrayLength
932 // B. offset <= (arrayLength - copyLength)
933 // We require that all of the values above, except for the sum and
934 // difference, are already known to be non-negative.
935 // Plan A is robust in the face of overflow, if offset and copyLength
936 // are both hugely positive.
937 //
938 // Plan B is less direct and intuitive, but it does not overflow at
939 // all, since the difference of two non-negatives is always
940 // representable. Whenever Java methods must perform the equivalent
941 // check they generally use Plan B instead of Plan A.
942 // For the moment we use Plan A.
943 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
944 Node* subseq_length,
945 Node* array_length,
946 RegionNode* region) {
947 if (stopped())
948 return nullptr; // already stopped
949 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
950 if (zero_offset && subseq_length->eqv_uncast(array_length))
951 return nullptr; // common case of whole-array copy
952 Node* last = subseq_length;
953 if (!zero_offset) // last += offset
954 last = _gvn.transform(new AddINode(last, offset));
955 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
956 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
957 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
958 return is_over;
959 }
960
961 // Emit range checks for the given String.value byte array
962 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
963 if (stopped()) {
964 return; // already stopped
965 }
966 RegionNode* bailout = new RegionNode(1);
967 record_for_igvn(bailout);
968 if (char_count) {
969 // Convert char count to byte count
970 count = _gvn.transform(new LShiftINode(count, intcon(1)));
971 }
972
973 // Offset and count must not be negative
974 generate_negative_guard(offset, bailout);
975 generate_negative_guard(count, bailout);
976 // Offset + count must not exceed length of array
977 generate_limit_guard(offset, count, load_array_length(array), bailout);
978
979 if (bailout->req() > 1) {
980 PreserveJVMState pjvms(this);
981 set_control(_gvn.transform(bailout));
982 uncommon_trap(Deoptimization::Reason_intrinsic,
983 Deoptimization::Action_maybe_recompile);
984 }
985 }
986
987 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
988 bool is_immutable) {
989 ciKlass* thread_klass = env()->Thread_klass();
990 const Type* thread_type
991 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
992
993 Node* thread = _gvn.transform(new ThreadLocalNode());
994 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
995 tls_output = thread;
996
997 Node* thread_obj_handle
998 = (is_immutable
999 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1000 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1001 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1002 thread_obj_handle = _gvn.transform(thread_obj_handle);
1003
1004 DecoratorSet decorators = IN_NATIVE;
1005 if (is_immutable) {
1006 decorators |= C2_IMMUTABLE_MEMORY;
1007 }
1008 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1009 }
1010
1011 //--------------------------generate_current_thread--------------------
1012 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1013 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1014 /*is_immutable*/false);
1015 }
1016
1017 //--------------------------generate_virtual_thread--------------------
1018 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1019 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1020 !C->method()->changes_current_thread());
1021 }
1022
1023 //------------------------------make_string_method_node------------------------
1024 // Helper method for String intrinsic functions. This version is called with
1025 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1026 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1027 // containing the lengths of str1 and str2.
1028 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1029 Node* result = nullptr;
1030 switch (opcode) {
1031 case Op_StrIndexOf:
1032 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1033 str1_start, cnt1, str2_start, cnt2, ae);
1034 break;
1035 case Op_StrComp:
1036 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1037 str1_start, cnt1, str2_start, cnt2, ae);
1038 break;
1039 case Op_StrEquals:
1040 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1041 // Use the constant length if there is one because optimized match rule may exist.
1042 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1043 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1044 break;
1045 default:
1046 ShouldNotReachHere();
1047 return nullptr;
1048 }
1049
1050 // All these intrinsics have checks.
1051 C->set_has_split_ifs(true); // Has chance for split-if optimization
1052 clear_upper_avx();
1053
1054 return _gvn.transform(result);
1055 }
1056
1057 //------------------------------inline_string_compareTo------------------------
1058 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1059 Node* arg1 = argument(0);
1060 Node* arg2 = argument(1);
1061
1062 arg1 = must_be_not_null(arg1, true);
1063 arg2 = must_be_not_null(arg2, true);
1064
1065 // Get start addr and length of first argument
1066 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1067 Node* arg1_cnt = load_array_length(arg1);
1068
1069 // Get start addr and length of second argument
1070 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1071 Node* arg2_cnt = load_array_length(arg2);
1072
1073 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1074 set_result(result);
1075 return true;
1076 }
1077
1078 //------------------------------inline_string_equals------------------------
1079 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1080 Node* arg1 = argument(0);
1081 Node* arg2 = argument(1);
1082
1083 // paths (plus control) merge
1084 RegionNode* region = new RegionNode(3);
1085 Node* phi = new PhiNode(region, TypeInt::BOOL);
1086
1087 if (!stopped()) {
1088
1089 arg1 = must_be_not_null(arg1, true);
1090 arg2 = must_be_not_null(arg2, true);
1091
1092 // Get start addr and length of first argument
1093 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1094 Node* arg1_cnt = load_array_length(arg1);
1095
1096 // Get start addr and length of second argument
1097 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1098 Node* arg2_cnt = load_array_length(arg2);
1099
1100 // Check for arg1_cnt != arg2_cnt
1101 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1102 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1103 Node* if_ne = generate_slow_guard(bol, nullptr);
1104 if (if_ne != nullptr) {
1105 phi->init_req(2, intcon(0));
1106 region->init_req(2, if_ne);
1107 }
1108
1109 // Check for count == 0 is done by assembler code for StrEquals.
1110
1111 if (!stopped()) {
1112 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1113 phi->init_req(1, equals);
1114 region->init_req(1, control());
1115 }
1116 }
1117
1118 // post merge
1119 set_control(_gvn.transform(region));
1120 record_for_igvn(region);
1121
1122 set_result(_gvn.transform(phi));
1123 return true;
1124 }
1125
1126 //------------------------------inline_array_equals----------------------------
1127 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1128 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1129 Node* arg1 = argument(0);
1130 Node* arg2 = argument(1);
1131
1132 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1133 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1134 clear_upper_avx();
1135
1136 return true;
1137 }
1138
1139
1140 //------------------------------inline_countPositives------------------------------
1141 bool LibraryCallKit::inline_countPositives() {
1142 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1143 return false;
1144 }
1145
1146 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1147 // no receiver since it is static method
1148 Node* ba = argument(0);
1149 Node* offset = argument(1);
1150 Node* len = argument(2);
1151
1152 ba = must_be_not_null(ba, true);
1153
1154 // Range checks
1155 generate_string_range_check(ba, offset, len, false);
1156 if (stopped()) {
1157 return true;
1158 }
1159 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1160 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1161 set_result(_gvn.transform(result));
1162 clear_upper_avx();
1163 return true;
1164 }
1165
1166 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1167 Node* index = argument(0);
1168 Node* length = bt == T_INT ? argument(1) : argument(2);
1169 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1170 return false;
1171 }
1172
1173 // check that length is positive
1174 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1175 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1176
1177 {
1178 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1179 uncommon_trap(Deoptimization::Reason_intrinsic,
1180 Deoptimization::Action_make_not_entrant);
1181 }
1182
1183 if (stopped()) {
1184 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1185 return true;
1186 }
1187
1188 // length is now known positive, add a cast node to make this explicit
1189 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1190 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1191 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1192 ConstraintCastNode::RegularDependency, bt);
1193 casted_length = _gvn.transform(casted_length);
1194 replace_in_map(length, casted_length);
1195 length = casted_length;
1196
1197 // Use an unsigned comparison for the range check itself
1198 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1199 BoolTest::mask btest = BoolTest::lt;
1200 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1201 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1202 _gvn.set_type(rc, rc->Value(&_gvn));
1203 if (!rc_bool->is_Con()) {
1204 record_for_igvn(rc);
1205 }
1206 set_control(_gvn.transform(new IfTrueNode(rc)));
1207 {
1208 PreserveJVMState pjvms(this);
1209 set_control(_gvn.transform(new IfFalseNode(rc)));
1210 uncommon_trap(Deoptimization::Reason_range_check,
1211 Deoptimization::Action_make_not_entrant);
1212 }
1213
1214 if (stopped()) {
1215 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1216 return true;
1217 }
1218
1219 // index is now known to be >= 0 and < length, cast it
1220 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1221 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1222 ConstraintCastNode::RegularDependency, bt);
1223 result = _gvn.transform(result);
1224 set_result(result);
1225 replace_in_map(index, result);
1226 return true;
1227 }
1228
1229 //------------------------------inline_string_indexOf------------------------
1230 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1231 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1232 return false;
1233 }
1234 Node* src = argument(0);
1235 Node* tgt = argument(1);
1236
1237 // Make the merge point
1238 RegionNode* result_rgn = new RegionNode(4);
1239 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1240
1241 src = must_be_not_null(src, true);
1242 tgt = must_be_not_null(tgt, true);
1243
1244 // Get start addr and length of source string
1245 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1246 Node* src_count = load_array_length(src);
1247
1248 // Get start addr and length of substring
1249 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1250 Node* tgt_count = load_array_length(tgt);
1251
1252 Node* result = nullptr;
1253 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1254
1255 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1256 // Divide src size by 2 if String is UTF16 encoded
1257 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1258 }
1259 if (ae == StrIntrinsicNode::UU) {
1260 // Divide substring size by 2 if String is UTF16 encoded
1261 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1262 }
1263
1264 if (call_opt_stub) {
1265 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1266 StubRoutines::_string_indexof_array[ae],
1267 "stringIndexOf", TypePtr::BOTTOM, src_start,
1268 src_count, tgt_start, tgt_count);
1269 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1270 } else {
1271 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1272 result_rgn, result_phi, ae);
1273 }
1274 if (result != nullptr) {
1275 result_phi->init_req(3, result);
1276 result_rgn->init_req(3, control());
1277 }
1278 set_control(_gvn.transform(result_rgn));
1279 record_for_igvn(result_rgn);
1280 set_result(_gvn.transform(result_phi));
1281
1282 return true;
1283 }
1284
1285 //-----------------------------inline_string_indexOfI-----------------------
1286 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1287 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1288 return false;
1289 }
1290 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1291 return false;
1292 }
1293
1294 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1295 Node* src = argument(0); // byte[]
1296 Node* src_count = argument(1); // char count
1297 Node* tgt = argument(2); // byte[]
1298 Node* tgt_count = argument(3); // char count
1299 Node* from_index = argument(4); // char index
1300
1301 src = must_be_not_null(src, true);
1302 tgt = must_be_not_null(tgt, true);
1303
1304 // Multiply byte array index by 2 if String is UTF16 encoded
1305 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1306 src_count = _gvn.transform(new SubINode(src_count, from_index));
1307 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1308 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1309
1310 // Range checks
1311 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1312 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1313 if (stopped()) {
1314 return true;
1315 }
1316
1317 RegionNode* region = new RegionNode(5);
1318 Node* phi = new PhiNode(region, TypeInt::INT);
1319 Node* result = nullptr;
1320
1321 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1322
1323 if (call_opt_stub) {
1324 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1325 StubRoutines::_string_indexof_array[ae],
1326 "stringIndexOf", TypePtr::BOTTOM, src_start,
1327 src_count, tgt_start, tgt_count);
1328 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1329 } else {
1330 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1331 region, phi, ae);
1332 }
1333 if (result != nullptr) {
1334 // The result is index relative to from_index if substring was found, -1 otherwise.
1335 // Generate code which will fold into cmove.
1336 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1337 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1338
1339 Node* if_lt = generate_slow_guard(bol, nullptr);
1340 if (if_lt != nullptr) {
1341 // result == -1
1342 phi->init_req(3, result);
1343 region->init_req(3, if_lt);
1344 }
1345 if (!stopped()) {
1346 result = _gvn.transform(new AddINode(result, from_index));
1347 phi->init_req(4, result);
1348 region->init_req(4, control());
1349 }
1350 }
1351
1352 set_control(_gvn.transform(region));
1353 record_for_igvn(region);
1354 set_result(_gvn.transform(phi));
1355 clear_upper_avx();
1356
1357 return true;
1358 }
1359
1360 // Create StrIndexOfNode with fast path checks
1361 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1362 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1363 // Check for substr count > string count
1364 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1365 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1366 Node* if_gt = generate_slow_guard(bol, nullptr);
1367 if (if_gt != nullptr) {
1368 phi->init_req(1, intcon(-1));
1369 region->init_req(1, if_gt);
1370 }
1371 if (!stopped()) {
1372 // Check for substr count == 0
1373 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1374 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1375 Node* if_zero = generate_slow_guard(bol, nullptr);
1376 if (if_zero != nullptr) {
1377 phi->init_req(2, intcon(0));
1378 region->init_req(2, if_zero);
1379 }
1380 }
1381 if (!stopped()) {
1382 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1383 }
1384 return nullptr;
1385 }
1386
1387 //-----------------------------inline_string_indexOfChar-----------------------
1388 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1389 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1390 return false;
1391 }
1392 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1393 return false;
1394 }
1395 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1396 Node* src = argument(0); // byte[]
1397 Node* int_ch = argument(1);
1398 Node* from_index = argument(2);
1399 Node* max = argument(3);
1400
1401 src = must_be_not_null(src, true);
1402
1403 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1404 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1405 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1406
1407 // Range checks
1408 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1409
1410 // Check for int_ch >= 0
1411 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1412 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1413 {
1414 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1415 uncommon_trap(Deoptimization::Reason_intrinsic,
1416 Deoptimization::Action_maybe_recompile);
1417 }
1418 if (stopped()) {
1419 return true;
1420 }
1421
1422 RegionNode* region = new RegionNode(3);
1423 Node* phi = new PhiNode(region, TypeInt::INT);
1424
1425 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1426 C->set_has_split_ifs(true); // Has chance for split-if optimization
1427 _gvn.transform(result);
1428
1429 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1430 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1431
1432 Node* if_lt = generate_slow_guard(bol, nullptr);
1433 if (if_lt != nullptr) {
1434 // result == -1
1435 phi->init_req(2, result);
1436 region->init_req(2, if_lt);
1437 }
1438 if (!stopped()) {
1439 result = _gvn.transform(new AddINode(result, from_index));
1440 phi->init_req(1, result);
1441 region->init_req(1, control());
1442 }
1443 set_control(_gvn.transform(region));
1444 record_for_igvn(region);
1445 set_result(_gvn.transform(phi));
1446 clear_upper_avx();
1447
1448 return true;
1449 }
1450 //---------------------------inline_string_copy---------------------
1451 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1452 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1453 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1454 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1455 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1456 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1457 bool LibraryCallKit::inline_string_copy(bool compress) {
1458 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1459 return false;
1460 }
1461 int nargs = 5; // 2 oops, 3 ints
1462 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1463
1464 Node* src = argument(0);
1465 Node* src_offset = argument(1);
1466 Node* dst = argument(2);
1467 Node* dst_offset = argument(3);
1468 Node* length = argument(4);
1469
1470 // Check for allocation before we add nodes that would confuse
1471 // tightly_coupled_allocation()
1472 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1473
1474 // Figure out the size and type of the elements we will be copying.
1475 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1476 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1477 if (src_type == nullptr || dst_type == nullptr) {
1478 return false;
1479 }
1480 BasicType src_elem = src_type->elem()->array_element_basic_type();
1481 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1482 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1483 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1484 "Unsupported array types for inline_string_copy");
1485
1486 src = must_be_not_null(src, true);
1487 dst = must_be_not_null(dst, true);
1488
1489 // Convert char[] offsets to byte[] offsets
1490 bool convert_src = (compress && src_elem == T_BYTE);
1491 bool convert_dst = (!compress && dst_elem == T_BYTE);
1492 if (convert_src) {
1493 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1494 } else if (convert_dst) {
1495 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1496 }
1497
1498 // Range checks
1499 generate_string_range_check(src, src_offset, length, convert_src);
1500 generate_string_range_check(dst, dst_offset, length, convert_dst);
1501 if (stopped()) {
1502 return true;
1503 }
1504
1505 Node* src_start = array_element_address(src, src_offset, src_elem);
1506 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1507 // 'src_start' points to src array + scaled offset
1508 // 'dst_start' points to dst array + scaled offset
1509 Node* count = nullptr;
1510 if (compress) {
1511 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1512 } else {
1513 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1514 }
1515
1516 if (alloc != nullptr) {
1517 if (alloc->maybe_set_complete(&_gvn)) {
1518 // "You break it, you buy it."
1519 InitializeNode* init = alloc->initialization();
1520 assert(init->is_complete(), "we just did this");
1521 init->set_complete_with_arraycopy();
1522 assert(dst->is_CheckCastPP(), "sanity");
1523 assert(dst->in(0)->in(0) == init, "dest pinned");
1524 }
1525 // Do not let stores that initialize this object be reordered with
1526 // a subsequent store that would make this object accessible by
1527 // other threads.
1528 // Record what AllocateNode this StoreStore protects so that
1529 // escape analysis can go from the MemBarStoreStoreNode to the
1530 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1531 // based on the escape status of the AllocateNode.
1532 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1533 }
1534 if (compress) {
1535 set_result(_gvn.transform(count));
1536 }
1537 clear_upper_avx();
1538
1539 return true;
1540 }
1541
1542 #ifdef _LP64
1543 #define XTOP ,top() /*additional argument*/
1544 #else //_LP64
1545 #define XTOP /*no additional argument*/
1546 #endif //_LP64
1547
1548 //------------------------inline_string_toBytesU--------------------------
1549 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1550 bool LibraryCallKit::inline_string_toBytesU() {
1551 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1552 return false;
1553 }
1554 // Get the arguments.
1555 Node* value = argument(0);
1556 Node* offset = argument(1);
1557 Node* length = argument(2);
1558
1559 Node* newcopy = nullptr;
1560
1561 // Set the original stack and the reexecute bit for the interpreter to reexecute
1562 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1563 { PreserveReexecuteState preexecs(this);
1564 jvms()->set_should_reexecute(true);
1565
1566 // Check if a null path was taken unconditionally.
1567 value = null_check(value);
1568
1569 RegionNode* bailout = new RegionNode(1);
1570 record_for_igvn(bailout);
1571
1572 // Range checks
1573 generate_negative_guard(offset, bailout);
1574 generate_negative_guard(length, bailout);
1575 generate_limit_guard(offset, length, load_array_length(value), bailout);
1576 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1577 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1578
1579 if (bailout->req() > 1) {
1580 PreserveJVMState pjvms(this);
1581 set_control(_gvn.transform(bailout));
1582 uncommon_trap(Deoptimization::Reason_intrinsic,
1583 Deoptimization::Action_maybe_recompile);
1584 }
1585 if (stopped()) {
1586 return true;
1587 }
1588
1589 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1590 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1591 newcopy = new_array(klass_node, size, 0); // no arguments to push
1592 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1593 guarantee(alloc != nullptr, "created above");
1594
1595 // Calculate starting addresses.
1596 Node* src_start = array_element_address(value, offset, T_CHAR);
1597 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1598
1599 // Check if dst array address is aligned to HeapWordSize
1600 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1601 // If true, then check if src array address is aligned to HeapWordSize
1602 if (aligned) {
1603 const TypeInt* toffset = gvn().type(offset)->is_int();
1604 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1605 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1606 }
1607
1608 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1609 const char* copyfunc_name = "arraycopy";
1610 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1611 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1612 OptoRuntime::fast_arraycopy_Type(),
1613 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1614 src_start, dst_start, ConvI2X(length) XTOP);
1615 // Do not let reads from the cloned object float above the arraycopy.
1616 if (alloc->maybe_set_complete(&_gvn)) {
1617 // "You break it, you buy it."
1618 InitializeNode* init = alloc->initialization();
1619 assert(init->is_complete(), "we just did this");
1620 init->set_complete_with_arraycopy();
1621 assert(newcopy->is_CheckCastPP(), "sanity");
1622 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1623 }
1624 // Do not let stores that initialize this object be reordered with
1625 // a subsequent store that would make this object accessible by
1626 // other threads.
1627 // Record what AllocateNode this StoreStore protects so that
1628 // escape analysis can go from the MemBarStoreStoreNode to the
1629 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1630 // based on the escape status of the AllocateNode.
1631 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1632 } // original reexecute is set back here
1633
1634 C->set_has_split_ifs(true); // Has chance for split-if optimization
1635 if (!stopped()) {
1636 set_result(newcopy);
1637 }
1638 clear_upper_avx();
1639
1640 return true;
1641 }
1642
1643 //------------------------inline_string_getCharsU--------------------------
1644 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1645 bool LibraryCallKit::inline_string_getCharsU() {
1646 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1647 return false;
1648 }
1649
1650 // Get the arguments.
1651 Node* src = argument(0);
1652 Node* src_begin = argument(1);
1653 Node* src_end = argument(2); // exclusive offset (i < src_end)
1654 Node* dst = argument(3);
1655 Node* dst_begin = argument(4);
1656
1657 // Check for allocation before we add nodes that would confuse
1658 // tightly_coupled_allocation()
1659 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1660
1661 // Check if a null path was taken unconditionally.
1662 src = null_check(src);
1663 dst = null_check(dst);
1664 if (stopped()) {
1665 return true;
1666 }
1667
1668 // Get length and convert char[] offset to byte[] offset
1669 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1670 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1671
1672 // Range checks
1673 generate_string_range_check(src, src_begin, length, true);
1674 generate_string_range_check(dst, dst_begin, length, false);
1675 if (stopped()) {
1676 return true;
1677 }
1678
1679 if (!stopped()) {
1680 // Calculate starting addresses.
1681 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1682 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1683
1684 // Check if array addresses are aligned to HeapWordSize
1685 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1686 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1687 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1688 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1689
1690 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1691 const char* copyfunc_name = "arraycopy";
1692 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1693 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1694 OptoRuntime::fast_arraycopy_Type(),
1695 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1696 src_start, dst_start, ConvI2X(length) XTOP);
1697 // Do not let reads from the cloned object float above the arraycopy.
1698 if (alloc != nullptr) {
1699 if (alloc->maybe_set_complete(&_gvn)) {
1700 // "You break it, you buy it."
1701 InitializeNode* init = alloc->initialization();
1702 assert(init->is_complete(), "we just did this");
1703 init->set_complete_with_arraycopy();
1704 assert(dst->is_CheckCastPP(), "sanity");
1705 assert(dst->in(0)->in(0) == init, "dest pinned");
1706 }
1707 // Do not let stores that initialize this object be reordered with
1708 // a subsequent store that would make this object accessible by
1709 // other threads.
1710 // Record what AllocateNode this StoreStore protects so that
1711 // escape analysis can go from the MemBarStoreStoreNode to the
1712 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1713 // based on the escape status of the AllocateNode.
1714 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1715 } else {
1716 insert_mem_bar(Op_MemBarCPUOrder);
1717 }
1718 }
1719
1720 C->set_has_split_ifs(true); // Has chance for split-if optimization
1721 return true;
1722 }
1723
1724 //----------------------inline_string_char_access----------------------------
1725 // Store/Load char to/from byte[] array.
1726 // static void StringUTF16.putChar(byte[] val, int index, int c)
1727 // static char StringUTF16.getChar(byte[] val, int index)
1728 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1729 Node* value = argument(0);
1730 Node* index = argument(1);
1731 Node* ch = is_store ? argument(2) : nullptr;
1732
1733 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1734 // correctly requires matched array shapes.
1735 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1736 "sanity: byte[] and char[] bases agree");
1737 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1738 "sanity: byte[] and char[] scales agree");
1739
1740 // Bail when getChar over constants is requested: constant folding would
1741 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1742 // Java method would constant fold nicely instead.
1743 if (!is_store && value->is_Con() && index->is_Con()) {
1744 return false;
1745 }
1746
1747 // Save state and restore on bailout
1748 uint old_sp = sp();
1749 SafePointNode* old_map = clone_map();
1750
1751 value = must_be_not_null(value, true);
1752
1753 Node* adr = array_element_address(value, index, T_CHAR);
1754 if (adr->is_top()) {
1755 set_map(old_map);
1756 set_sp(old_sp);
1757 return false;
1758 }
1759 destruct_map_clone(old_map);
1760 if (is_store) {
1761 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1762 } else {
1763 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);
1764 set_result(ch);
1765 }
1766 return true;
1767 }
1768
1769
1770 //------------------------------inline_math-----------------------------------
1771 // public static double Math.abs(double)
1772 // public static double Math.sqrt(double)
1773 // public static double Math.log(double)
1774 // public static double Math.log10(double)
1775 // public static double Math.round(double)
1776 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1777 Node* arg = argument(0);
1778 Node* n = nullptr;
1779 switch (id) {
1780 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1781 case vmIntrinsics::_dsqrt:
1782 case vmIntrinsics::_dsqrt_strict:
1783 n = new SqrtDNode(C, control(), arg); break;
1784 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1785 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1786 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1787 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1788 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1789 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1790 default: fatal_unexpected_iid(id); break;
1791 }
1792 set_result(_gvn.transform(n));
1793 return true;
1794 }
1795
1796 //------------------------------inline_math-----------------------------------
1797 // public static float Math.abs(float)
1798 // public static int Math.abs(int)
1799 // public static long Math.abs(long)
1800 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1801 Node* arg = argument(0);
1802 Node* n = nullptr;
1803 switch (id) {
1804 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1805 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1806 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1807 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1808 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1809 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1810 default: fatal_unexpected_iid(id); break;
1811 }
1812 set_result(_gvn.transform(n));
1813 return true;
1814 }
1815
1816 //------------------------------runtime_math-----------------------------
1817 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1818 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1819 "must be (DD)D or (D)D type");
1820
1821 // Inputs
1822 Node* a = argument(0);
1823 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1824
1825 const TypePtr* no_memory_effects = nullptr;
1826 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1827 no_memory_effects,
1828 a, top(), b, b ? top() : nullptr);
1829 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1830 #ifdef ASSERT
1831 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1832 assert(value_top == top(), "second value must be top");
1833 #endif
1834
1835 set_result(value);
1836 return true;
1837 }
1838
1839 //------------------------------inline_math_pow-----------------------------
1840 bool LibraryCallKit::inline_math_pow() {
1841 Node* exp = argument(2);
1842 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1843 if (d != nullptr) {
1844 if (d->getd() == 2.0) {
1845 // Special case: pow(x, 2.0) => x * x
1846 Node* base = argument(0);
1847 set_result(_gvn.transform(new MulDNode(base, base)));
1848 return true;
1849 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1850 // Special case: pow(x, 0.5) => sqrt(x)
1851 Node* base = argument(0);
1852 Node* zero = _gvn.zerocon(T_DOUBLE);
1853
1854 RegionNode* region = new RegionNode(3);
1855 Node* phi = new PhiNode(region, Type::DOUBLE);
1856
1857 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1858 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1859 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1860 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1861 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1862
1863 Node* if_pow = generate_slow_guard(test, nullptr);
1864 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1865 phi->init_req(1, value_sqrt);
1866 region->init_req(1, control());
1867
1868 if (if_pow != nullptr) {
1869 set_control(if_pow);
1870 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1871 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1872 const TypePtr* no_memory_effects = nullptr;
1873 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1874 no_memory_effects, base, top(), exp, top());
1875 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1876 #ifdef ASSERT
1877 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1878 assert(value_top == top(), "second value must be top");
1879 #endif
1880 phi->init_req(2, value_pow);
1881 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1882 }
1883
1884 C->set_has_split_ifs(true); // Has chance for split-if optimization
1885 set_control(_gvn.transform(region));
1886 record_for_igvn(region);
1887 set_result(_gvn.transform(phi));
1888
1889 return true;
1890 }
1891 }
1892
1893 return StubRoutines::dpow() != nullptr ?
1894 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1895 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1896 }
1897
1898 //------------------------------inline_math_native-----------------------------
1899 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1900 switch (id) {
1901 case vmIntrinsics::_dsin:
1902 return StubRoutines::dsin() != nullptr ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1905 case vmIntrinsics::_dcos:
1906 return StubRoutines::dcos() != nullptr ?
1907 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1908 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1909 case vmIntrinsics::_dtan:
1910 return StubRoutines::dtan() != nullptr ?
1911 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1912 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1913 case vmIntrinsics::_dtanh:
1914 return StubRoutines::dtanh() != nullptr ?
1915 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1916 case vmIntrinsics::_dcbrt:
1917 return StubRoutines::dcbrt() != nullptr ?
1918 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1919 case vmIntrinsics::_dexp:
1920 return StubRoutines::dexp() != nullptr ?
1921 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1922 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1923 case vmIntrinsics::_dlog:
1924 return StubRoutines::dlog() != nullptr ?
1925 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1926 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1927 case vmIntrinsics::_dlog10:
1928 return StubRoutines::dlog10() != nullptr ?
1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1930 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1931
1932 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1933 case vmIntrinsics::_ceil:
1934 case vmIntrinsics::_floor:
1935 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1936
1937 case vmIntrinsics::_dsqrt:
1938 case vmIntrinsics::_dsqrt_strict:
1939 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1940 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1941 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1942 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1943 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1944
1945 case vmIntrinsics::_dpow: return inline_math_pow();
1946 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1947 case vmIntrinsics::_fcopySign: return inline_math(id);
1948 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1949 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1950 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1951
1952 // These intrinsics are not yet correctly implemented
1953 case vmIntrinsics::_datan2:
1954 return false;
1955
1956 default:
1957 fatal_unexpected_iid(id);
1958 return false;
1959 }
1960 }
1961
1962 //----------------------------inline_notify-----------------------------------*
1963 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1964 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1965 address func;
1966 if (id == vmIntrinsics::_notify) {
1967 func = OptoRuntime::monitor_notify_Java();
1968 } else {
1969 func = OptoRuntime::monitor_notifyAll_Java();
1970 }
1971 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1972 make_slow_call_ex(call, env()->Throwable_klass(), false);
1973 return true;
1974 }
1975
1976
1977 //----------------------------inline_min_max-----------------------------------
1978 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1979 Node* a = nullptr;
1980 Node* b = nullptr;
1981 Node* n = nullptr;
1982 switch (id) {
1983 case vmIntrinsics::_min:
1984 case vmIntrinsics::_max:
1985 case vmIntrinsics::_minF:
1986 case vmIntrinsics::_maxF:
1987 case vmIntrinsics::_minF_strict:
1988 case vmIntrinsics::_maxF_strict:
1989 case vmIntrinsics::_min_strict:
1990 case vmIntrinsics::_max_strict:
1991 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1992 a = argument(0);
1993 b = argument(1);
1994 break;
1995 case vmIntrinsics::_minD:
1996 case vmIntrinsics::_maxD:
1997 case vmIntrinsics::_minD_strict:
1998 case vmIntrinsics::_maxD_strict:
1999 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2000 a = argument(0);
2001 b = argument(2);
2002 break;
2003 case vmIntrinsics::_minL:
2004 case vmIntrinsics::_maxL:
2005 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2006 a = argument(0);
2007 b = argument(2);
2008 break;
2009 default:
2010 fatal_unexpected_iid(id);
2011 break;
2012 }
2013
2014 switch (id) {
2015 case vmIntrinsics::_min:
2016 case vmIntrinsics::_min_strict:
2017 n = new MinINode(a, b);
2018 break;
2019 case vmIntrinsics::_max:
2020 case vmIntrinsics::_max_strict:
2021 n = new MaxINode(a, b);
2022 break;
2023 case vmIntrinsics::_minF:
2024 case vmIntrinsics::_minF_strict:
2025 n = new MinFNode(a, b);
2026 break;
2027 case vmIntrinsics::_maxF:
2028 case vmIntrinsics::_maxF_strict:
2029 n = new MaxFNode(a, b);
2030 break;
2031 case vmIntrinsics::_minD:
2032 case vmIntrinsics::_minD_strict:
2033 n = new MinDNode(a, b);
2034 break;
2035 case vmIntrinsics::_maxD:
2036 case vmIntrinsics::_maxD_strict:
2037 n = new MaxDNode(a, b);
2038 break;
2039 case vmIntrinsics::_minL:
2040 n = new MinLNode(_gvn.C, a, b);
2041 break;
2042 case vmIntrinsics::_maxL:
2043 n = new MaxLNode(_gvn.C, a, b);
2044 break;
2045 default:
2046 fatal_unexpected_iid(id);
2047 break;
2048 }
2049
2050 set_result(_gvn.transform(n));
2051 return true;
2052 }
2053
2054 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2055 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2056 env()->ArithmeticException_instance())) {
2057 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2058 // so let's bail out intrinsic rather than risking deopting again.
2059 return false;
2060 }
2061
2062 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2063 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2064 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2065 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2066
2067 {
2068 PreserveJVMState pjvms(this);
2069 PreserveReexecuteState preexecs(this);
2070 jvms()->set_should_reexecute(true);
2071
2072 set_control(slow_path);
2073 set_i_o(i_o());
2074
2075 builtin_throw(Deoptimization::Reason_intrinsic,
2076 env()->ArithmeticException_instance(),
2077 /*allow_too_many_traps*/ false);
2078 }
2079
2080 set_control(fast_path);
2081 set_result(math);
2082 return true;
2083 }
2084
2085 template <typename OverflowOp>
2086 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2087 typedef typename OverflowOp::MathOp MathOp;
2088
2089 MathOp* mathOp = new MathOp(arg1, arg2);
2090 Node* operation = _gvn.transform( mathOp );
2091 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2092 return inline_math_mathExact(operation, ofcheck);
2093 }
2094
2095 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2096 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2097 }
2098
2099 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2100 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2101 }
2102
2103 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2104 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2105 }
2106
2107 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2108 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2109 }
2110
2111 bool LibraryCallKit::inline_math_negateExactI() {
2112 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2113 }
2114
2115 bool LibraryCallKit::inline_math_negateExactL() {
2116 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2117 }
2118
2119 bool LibraryCallKit::inline_math_multiplyExactI() {
2120 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2121 }
2122
2123 bool LibraryCallKit::inline_math_multiplyExactL() {
2124 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2125 }
2126
2127 bool LibraryCallKit::inline_math_multiplyHigh() {
2128 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2129 return true;
2130 }
2131
2132 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2133 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2134 return true;
2135 }
2136
2137 inline int
2138 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2139 const TypePtr* base_type = TypePtr::NULL_PTR;
2140 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2141 if (base_type == nullptr) {
2142 // Unknown type.
2143 return Type::AnyPtr;
2144 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2145 // Since this is a null+long form, we have to switch to a rawptr.
2146 base = _gvn.transform(new CastX2PNode(offset));
2147 offset = MakeConX(0);
2148 return Type::RawPtr;
2149 } else if (base_type->base() == Type::RawPtr) {
2150 return Type::RawPtr;
2151 } else if (base_type->isa_oopptr()) {
2152 // Base is never null => always a heap address.
2153 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2154 return Type::OopPtr;
2155 }
2156 // Offset is small => always a heap address.
2157 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2158 if (offset_type != nullptr &&
2159 base_type->offset() == 0 && // (should always be?)
2160 offset_type->_lo >= 0 &&
2161 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2162 return Type::OopPtr;
2163 } else if (type == T_OBJECT) {
2164 // off heap access to an oop doesn't make any sense. Has to be on
2165 // heap.
2166 return Type::OopPtr;
2167 }
2168 // Otherwise, it might either be oop+off or null+addr.
2169 return Type::AnyPtr;
2170 } else {
2171 // No information:
2172 return Type::AnyPtr;
2173 }
2174 }
2175
2176 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2177 Node* uncasted_base = base;
2178 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2179 if (kind == Type::RawPtr) {
2180 return basic_plus_adr(top(), uncasted_base, offset);
2181 } else if (kind == Type::AnyPtr) {
2182 assert(base == uncasted_base, "unexpected base change");
2183 if (can_cast) {
2184 if (!_gvn.type(base)->speculative_maybe_null() &&
2185 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2186 // According to profiling, this access is always on
2187 // heap. Casting the base to not null and thus avoiding membars
2188 // around the access should allow better optimizations
2189 Node* null_ctl = top();
2190 base = null_check_oop(base, &null_ctl, true, true, true);
2191 assert(null_ctl->is_top(), "no null control here");
2192 return basic_plus_adr(base, offset);
2193 } else if (_gvn.type(base)->speculative_always_null() &&
2194 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2195 // According to profiling, this access is always off
2196 // heap.
2197 base = null_assert(base);
2198 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2199 offset = MakeConX(0);
2200 return basic_plus_adr(top(), raw_base, offset);
2201 }
2202 }
2203 // We don't know if it's an on heap or off heap access. Fall back
2204 // to raw memory access.
2205 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2206 return basic_plus_adr(top(), raw, offset);
2207 } else {
2208 assert(base == uncasted_base, "unexpected base change");
2209 // We know it's an on heap access so base can't be null
2210 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2211 base = must_be_not_null(base, true);
2212 }
2213 return basic_plus_adr(base, offset);
2214 }
2215 }
2216
2217 //--------------------------inline_number_methods-----------------------------
2218 // inline int Integer.numberOfLeadingZeros(int)
2219 // inline int Long.numberOfLeadingZeros(long)
2220 //
2221 // inline int Integer.numberOfTrailingZeros(int)
2222 // inline int Long.numberOfTrailingZeros(long)
2223 //
2224 // inline int Integer.bitCount(int)
2225 // inline int Long.bitCount(long)
2226 //
2227 // inline char Character.reverseBytes(char)
2228 // inline short Short.reverseBytes(short)
2229 // inline int Integer.reverseBytes(int)
2230 // inline long Long.reverseBytes(long)
2231 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2232 Node* arg = argument(0);
2233 Node* n = nullptr;
2234 switch (id) {
2235 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2236 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2237 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2238 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2239 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2240 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2241 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2242 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2243 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2244 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2245 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2246 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2247 default: fatal_unexpected_iid(id); break;
2248 }
2249 set_result(_gvn.transform(n));
2250 return true;
2251 }
2252
2253 //--------------------------inline_bitshuffle_methods-----------------------------
2254 // inline int Integer.compress(int, int)
2255 // inline int Integer.expand(int, int)
2256 // inline long Long.compress(long, long)
2257 // inline long Long.expand(long, long)
2258 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2259 Node* n = nullptr;
2260 switch (id) {
2261 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2262 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2263 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2264 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2265 default: fatal_unexpected_iid(id); break;
2266 }
2267 set_result(_gvn.transform(n));
2268 return true;
2269 }
2270
2271 //--------------------------inline_number_methods-----------------------------
2272 // inline int Integer.compareUnsigned(int, int)
2273 // inline int Long.compareUnsigned(long, long)
2274 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2275 Node* arg1 = argument(0);
2276 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2277 Node* n = nullptr;
2278 switch (id) {
2279 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2280 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2281 default: fatal_unexpected_iid(id); break;
2282 }
2283 set_result(_gvn.transform(n));
2284 return true;
2285 }
2286
2287 //--------------------------inline_unsigned_divmod_methods-----------------------------
2288 // inline int Integer.divideUnsigned(int, int)
2289 // inline int Integer.remainderUnsigned(int, int)
2290 // inline long Long.divideUnsigned(long, long)
2291 // inline long Long.remainderUnsigned(long, long)
2292 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2293 Node* n = nullptr;
2294 switch (id) {
2295 case vmIntrinsics::_divideUnsigned_i: {
2296 zero_check_int(argument(1));
2297 // Compile-time detect of null-exception
2298 if (stopped()) {
2299 return true; // keep the graph constructed so far
2300 }
2301 n = new UDivINode(control(), argument(0), argument(1));
2302 break;
2303 }
2304 case vmIntrinsics::_divideUnsigned_l: {
2305 zero_check_long(argument(2));
2306 // Compile-time detect of null-exception
2307 if (stopped()) {
2308 return true; // keep the graph constructed so far
2309 }
2310 n = new UDivLNode(control(), argument(0), argument(2));
2311 break;
2312 }
2313 case vmIntrinsics::_remainderUnsigned_i: {
2314 zero_check_int(argument(1));
2315 // Compile-time detect of null-exception
2316 if (stopped()) {
2317 return true; // keep the graph constructed so far
2318 }
2319 n = new UModINode(control(), argument(0), argument(1));
2320 break;
2321 }
2322 case vmIntrinsics::_remainderUnsigned_l: {
2323 zero_check_long(argument(2));
2324 // Compile-time detect of null-exception
2325 if (stopped()) {
2326 return true; // keep the graph constructed so far
2327 }
2328 n = new UModLNode(control(), argument(0), argument(2));
2329 break;
2330 }
2331 default: fatal_unexpected_iid(id); break;
2332 }
2333 set_result(_gvn.transform(n));
2334 return true;
2335 }
2336
2337 //----------------------------inline_unsafe_access----------------------------
2338
2339 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2340 // Attempt to infer a sharper value type from the offset and base type.
2341 ciKlass* sharpened_klass = nullptr;
2342 bool null_free = false;
2343
2344 // See if it is an instance field, with an object type.
2345 if (alias_type->field() != nullptr) {
2346 if (alias_type->field()->type()->is_klass()) {
2347 sharpened_klass = alias_type->field()->type()->as_klass();
2348 null_free = alias_type->field()->is_null_free();
2349 }
2350 }
2351
2352 const TypeOopPtr* result = nullptr;
2353 // See if it is a narrow oop array.
2354 if (adr_type->isa_aryptr()) {
2355 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2356 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2357 null_free = adr_type->is_aryptr()->is_null_free();
2358 if (elem_type != nullptr && elem_type->is_loaded()) {
2359 // Sharpen the value type.
2360 result = elem_type;
2361 }
2362 }
2363 }
2364
2365 // The sharpened class might be unloaded if there is no class loader
2366 // contraint in place.
2367 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2368 // Sharpen the value type.
2369 result = TypeOopPtr::make_from_klass(sharpened_klass);
2370 if (null_free) {
2371 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2372 }
2373 }
2374 if (result != nullptr) {
2375 #ifndef PRODUCT
2376 if (C->print_intrinsics() || C->print_inlining()) {
2377 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2378 tty->print(" sharpened value: "); result->dump(); tty->cr();
2379 }
2380 #endif
2381 }
2382 return result;
2383 }
2384
2385 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2386 switch (kind) {
2387 case Relaxed:
2388 return MO_UNORDERED;
2389 case Opaque:
2390 return MO_RELAXED;
2391 case Acquire:
2392 return MO_ACQUIRE;
2393 case Release:
2394 return MO_RELEASE;
2395 case Volatile:
2396 return MO_SEQ_CST;
2397 default:
2398 ShouldNotReachHere();
2399 return 0;
2400 }
2401 }
2402
2403 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2404 if (callee()->is_static()) return false; // caller must have the capability!
2405 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2406 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2407 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2408 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2409
2410 if (is_reference_type(type)) {
2411 decorators |= ON_UNKNOWN_OOP_REF;
2412 }
2413
2414 if (unaligned) {
2415 decorators |= C2_UNALIGNED;
2416 }
2417
2418 #ifndef PRODUCT
2419 {
2420 ResourceMark rm;
2421 // Check the signatures.
2422 ciSignature* sig = callee()->signature();
2423 #ifdef ASSERT
2424 if (!is_store) {
2425 // Object getReference(Object base, int/long offset), etc.
2426 BasicType rtype = sig->return_type()->basic_type();
2427 assert(rtype == type, "getter must return the expected value");
2428 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2429 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2430 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2431 } else {
2432 // void putReference(Object base, int/long offset, Object x), etc.
2433 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2434 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2435 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2436 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2437 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2438 assert(vtype == type, "putter must accept the expected value");
2439 }
2440 #endif // ASSERT
2441 }
2442 #endif //PRODUCT
2443
2444 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2445
2446 Node* receiver = argument(0); // type: oop
2447
2448 // Build address expression.
2449 Node* heap_base_oop = top();
2450
2451 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2452 Node* base = argument(1); // type: oop
2453 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2454 Node* offset = argument(2); // type: long
2455 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2456 // to be plain byte offsets, which are also the same as those accepted
2457 // by oopDesc::field_addr.
2458 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2459 "fieldOffset must be byte-scaled");
2460
2461 ciInlineKlass* inline_klass = nullptr;
2462 if (is_flat) {
2463 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2464 if (cls == nullptr || cls->const_oop() == nullptr) {
2465 return false;
2466 }
2467 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2468 if (!mirror_type->is_inlinetype()) {
2469 return false;
2470 }
2471 inline_klass = mirror_type->as_inline_klass();
2472 }
2473
2474 if (base->is_InlineType()) {
2475 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2476 InlineTypeNode* vt = base->as_InlineType();
2477 if (offset->is_Con()) {
2478 long off = find_long_con(offset, 0);
2479 ciInlineKlass* vk = vt->type()->inline_klass();
2480 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2481 return false;
2482 }
2483
2484 ciField* field = vk->get_non_flat_field_by_offset(off);
2485 if (field != nullptr) {
2486 BasicType bt = type2field[field->type()->basic_type()];
2487 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2488 bt = T_OBJECT;
2489 }
2490 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2491 Node* value = vt->field_value_by_offset(off, false);
2492 if (value->is_InlineType()) {
2493 value = value->as_InlineType()->adjust_scalarization_depth(this);
2494 }
2495 set_result(value);
2496 return true;
2497 }
2498 }
2499 }
2500 {
2501 // Re-execute the unsafe access if allocation triggers deoptimization.
2502 PreserveReexecuteState preexecs(this);
2503 jvms()->set_should_reexecute(true);
2504 vt = vt->buffer(this);
2505 }
2506 base = vt->get_oop();
2507 }
2508
2509 // 32-bit machines ignore the high half!
2510 offset = ConvL2X(offset);
2511
2512 // Save state and restore on bailout
2513 uint old_sp = sp();
2514 SafePointNode* old_map = clone_map();
2515
2516 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2517 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2518
2519 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2520 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2521 decorators |= IN_NATIVE; // off-heap primitive access
2522 } else {
2523 set_map(old_map);
2524 set_sp(old_sp);
2525 return false; // off-heap oop accesses are not supported
2526 }
2527 } else {
2528 heap_base_oop = base; // on-heap or mixed access
2529 }
2530
2531 // Can base be null? Otherwise, always on-heap access.
2532 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2533
2534 if (!can_access_non_heap) {
2535 decorators |= IN_HEAP;
2536 }
2537
2538 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2539
2540 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2541 if (adr_type == TypePtr::NULL_PTR) {
2542 set_map(old_map);
2543 set_sp(old_sp);
2544 return false; // off-heap access with zero address
2545 }
2546
2547 // Try to categorize the address.
2548 Compile::AliasType* alias_type = C->alias_type(adr_type);
2549 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2550
2551 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2552 alias_type->adr_type() == TypeAryPtr::RANGE) {
2553 set_map(old_map);
2554 set_sp(old_sp);
2555 return false; // not supported
2556 }
2557
2558 bool mismatched = false;
2559 BasicType bt = T_ILLEGAL;
2560 ciField* field = nullptr;
2561 if (adr_type->isa_instptr()) {
2562 const TypeInstPtr* instptr = adr_type->is_instptr();
2563 ciInstanceKlass* k = instptr->instance_klass();
2564 int off = instptr->offset();
2565 if (instptr->const_oop() != nullptr &&
2566 k == ciEnv::current()->Class_klass() &&
2567 instptr->offset() >= (k->size_helper() * wordSize)) {
2568 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2569 field = k->get_field_by_offset(off, true);
2570 } else {
2571 field = k->get_non_flat_field_by_offset(off);
2572 }
2573 if (field != nullptr) {
2574 bt = type2field[field->type()->basic_type()];
2575 }
2576 if (bt != alias_type->basic_type()) {
2577 // Type mismatch. Is it an access to a nested flat field?
2578 field = k->get_field_by_offset(off, false);
2579 if (field != nullptr) {
2580 bt = type2field[field->type()->basic_type()];
2581 }
2582 }
2583 assert(bt == alias_type->basic_type() || is_flat, "should match");
2584 } else {
2585 bt = alias_type->basic_type();
2586 }
2587
2588 if (bt != T_ILLEGAL) {
2589 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2590 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2591 // Alias type doesn't differentiate between byte[] and boolean[]).
2592 // Use address type to get the element type.
2593 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2594 }
2595 if (is_reference_type(bt, true)) {
2596 // accessing an array field with getReference is not a mismatch
2597 bt = T_OBJECT;
2598 }
2599 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2600 // Don't intrinsify mismatched object accesses
2601 set_map(old_map);
2602 set_sp(old_sp);
2603 return false;
2604 }
2605 mismatched = (bt != type);
2606 } else if (alias_type->adr_type()->isa_oopptr()) {
2607 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2608 }
2609
2610 if (is_flat) {
2611 if (adr_type->isa_instptr()) {
2612 if (field == nullptr || field->type() != inline_klass) {
2613 mismatched = true;
2614 }
2615 } else if (adr_type->isa_aryptr()) {
2616 const Type* elem = adr_type->is_aryptr()->elem();
2617 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2618 mismatched = true;
2619 }
2620 } else {
2621 mismatched = true;
2622 }
2623 if (is_store) {
2624 const Type* val_t = _gvn.type(val);
2625 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2626 set_map(old_map);
2627 set_sp(old_sp);
2628 return false;
2629 }
2630 }
2631 }
2632
2633 destruct_map_clone(old_map);
2634 assert(!mismatched || is_flat || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2635
2636 if (mismatched) {
2637 decorators |= C2_MISMATCHED;
2638 }
2639
2640 // First guess at the value type.
2641 const Type *value_type = Type::get_const_basic_type(type);
2642
2643 // Figure out the memory ordering.
2644 decorators |= mo_decorator_for_access_kind(kind);
2645
2646 if (!is_store) {
2647 if (type == T_OBJECT && !is_flat) {
2648 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2649 if (tjp != nullptr) {
2650 value_type = tjp;
2651 }
2652 }
2653 }
2654
2655 receiver = null_check(receiver);
2656 if (stopped()) {
2657 return true;
2658 }
2659 // Heap pointers get a null-check from the interpreter,
2660 // as a courtesy. However, this is not guaranteed by Unsafe,
2661 // and it is not possible to fully distinguish unintended nulls
2662 // from intended ones in this API.
2663
2664 if (!is_store) {
2665 Node* p = nullptr;
2666 // Try to constant fold a load from a constant field
2667
2668 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2669 // final or stable field
2670 p = make_constant_from_field(field, heap_base_oop);
2671 }
2672
2673 if (p == nullptr) { // Could not constant fold the load
2674 if (is_flat) {
2675 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, adr_type, false, false, true);
2676 } else {
2677 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2678 const TypeOopPtr* ptr = value_type->make_oopptr();
2679 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2680 // Load a non-flattened inline type from memory
2681 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2682 }
2683 }
2684 // Normalize the value returned by getBoolean in the following cases
2685 if (type == T_BOOLEAN &&
2686 (mismatched ||
2687 heap_base_oop == top() || // - heap_base_oop is null or
2688 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2689 // and the unsafe access is made to large offset
2690 // (i.e., larger than the maximum offset necessary for any
2691 // field access)
2692 ) {
2693 IdealKit ideal = IdealKit(this);
2694 #define __ ideal.
2695 IdealVariable normalized_result(ideal);
2696 __ declarations_done();
2697 __ set(normalized_result, p);
2698 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2699 __ set(normalized_result, ideal.ConI(1));
2700 ideal.end_if();
2701 final_sync(ideal);
2702 p = __ value(normalized_result);
2703 #undef __
2704 }
2705 }
2706 if (type == T_ADDRESS) {
2707 p = gvn().transform(new CastP2XNode(nullptr, p));
2708 p = ConvX2UL(p);
2709 }
2710 // The load node has the control of the preceding MemBarCPUOrder. All
2711 // following nodes will have the control of the MemBarCPUOrder inserted at
2712 // the end of this method. So, pushing the load onto the stack at a later
2713 // point is fine.
2714 set_result(p);
2715 } else {
2716 if (bt == T_ADDRESS) {
2717 // Repackage the long as a pointer.
2718 val = ConvL2X(val);
2719 val = gvn().transform(new CastX2PNode(val));
2720 }
2721 if (is_flat) {
2722 val->as_InlineType()->store_flat(this, base, adr, false, false, true, decorators);
2723 } else {
2724 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2725 }
2726 }
2727
2728 return true;
2729 }
2730
2731 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2732 #ifdef ASSERT
2733 {
2734 ResourceMark rm;
2735 // Check the signatures.
2736 ciSignature* sig = callee()->signature();
2737 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2738 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2739 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2740 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2741 if (is_store) {
2742 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2743 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2744 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2745 } else {
2746 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2747 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2748 }
2749 }
2750 #endif // ASSERT
2751
2752 assert(kind == Relaxed, "Only plain accesses for now");
2753 if (callee()->is_static()) {
2754 // caller must have the capability!
2755 return false;
2756 }
2757 C->set_has_unsafe_access(true);
2758
2759 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2760 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2761 // parameter valueType is not a constant
2762 return false;
2763 }
2764 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2765 if (!mirror_type->is_inlinetype()) {
2766 // Dead code
2767 return false;
2768 }
2769 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2770
2771 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2772 if (layout_type == nullptr || !layout_type->is_con()) {
2773 // parameter layoutKind is not a constant
2774 return false;
2775 }
2776 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2777 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2778 "invalid layoutKind %d", layout_type->get_con());
2779 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2780 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NON_ATOMIC_FLAT ||
2781 layout == LayoutKind::ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2782 "unexpected layoutKind %d", layout_type->get_con());
2783
2784 null_check(argument(0));
2785 if (stopped()) {
2786 return true;
2787 }
2788
2789 Node* base = must_be_not_null(argument(1), true);
2790 Node* offset = argument(2);
2791 const Type* base_type = _gvn.type(base);
2792
2793 Node* ptr;
2794 bool immutable_memory = false;
2795 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2796 if (base_type->isa_instptr()) {
2797 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2798 if (offset_type == nullptr || !offset_type->is_con()) {
2799 // Offset into a non-array should be a constant
2800 decorators |= C2_MISMATCHED;
2801 } else {
2802 int offset_con = checked_cast<int>(offset_type->get_con());
2803 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2804 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2805 if (field == nullptr) {
2806 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2807 decorators |= C2_MISMATCHED;
2808 } else {
2809 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2810 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2811 immutable_memory = field->is_strict() && field->is_final();
2812
2813 if (base->is_InlineType()) {
2814 assert(!is_store, "Cannot store into a non-larval value object");
2815 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2816 return true;
2817 }
2818 }
2819 }
2820
2821 if (base->is_InlineType()) {
2822 assert(!is_store, "Cannot store into a non-larval value object");
2823 base = base->as_InlineType()->buffer(this, true);
2824 }
2825 ptr = basic_plus_adr(base, ConvL2X(offset));
2826 } else if (base_type->isa_aryptr()) {
2827 decorators |= IS_ARRAY;
2828 if (layout == LayoutKind::REFERENCE) {
2829 if (!base_type->is_aryptr()->is_not_flat()) {
2830 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2831 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::StrongDependency));
2832 replace_in_map(base, new_base);
2833 base = new_base;
2834 }
2835 ptr = basic_plus_adr(base, ConvL2X(offset));
2836 } else {
2837 if (UseArrayFlattening) {
2838 // Flat array must have an exact type
2839 bool is_null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2840 bool is_atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2841 Node* new_base = cast_to_flat_array(base, value_klass, is_null_free, !is_null_free, is_atomic);
2842 replace_in_map(base, new_base);
2843 base = new_base;
2844 ptr = basic_plus_adr(base, ConvL2X(offset));
2845 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2846 if (ptr_type->field_offset().get() != 0) {
2847 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::StrongDependency));
2848 }
2849 } else {
2850 uncommon_trap(Deoptimization::Reason_intrinsic,
2851 Deoptimization::Action_none);
2852 return true;
2853 }
2854 }
2855 } else {
2856 decorators |= C2_MISMATCHED;
2857 ptr = basic_plus_adr(base, ConvL2X(offset));
2858 }
2859
2860 if (is_store) {
2861 Node* value = argument(6);
2862 const Type* value_type = _gvn.type(value);
2863 if (!value_type->is_inlinetypeptr()) {
2864 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2865 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::StrongDependency));
2866 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2867 replace_in_map(value, new_value);
2868 value = new_value;
2869 }
2870
2871 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());
2872 if (layout == LayoutKind::REFERENCE) {
2873 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2874 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2875 } else {
2876 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2877 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2878 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2879 }
2880
2881 return true;
2882 } else {
2883 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2884 InlineTypeNode* result;
2885 if (layout == LayoutKind::REFERENCE) {
2886 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2887 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2888 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2889 } else {
2890 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2891 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2892 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2893 }
2894
2895 set_result(result);
2896 return true;
2897 }
2898 }
2899
2900 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2901 Node* receiver = argument(0);
2902 Node* value = argument(1);
2903
2904 const Type* type = gvn().type(value);
2905 if (!type->is_inlinetypeptr()) {
2906 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2907 return false;
2908 }
2909
2910 null_check(receiver);
2911 if (stopped()) {
2912 return true;
2913 }
2914
2915 value = null_check(value);
2916 if (stopped()) {
2917 return true;
2918 }
2919
2920 ciInlineKlass* vk = type->inline_klass();
2921 Node* klass = makecon(TypeKlassPtr::make(vk));
2922 Node* obj = new_instance(klass);
2923 AllocateNode::Ideal_allocation(obj)->_larval = true;
2924
2925 assert(value->is_InlineType(), "must be an InlineTypeNode");
2926 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2927 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2928
2929 set_result(obj);
2930 return true;
2931 }
2932
2933 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2934 Node* receiver = argument(0);
2935 Node* buffer = argument(1);
2936
2937 const Type* type = gvn().type(buffer);
2938 if (!type->is_inlinetypeptr()) {
2939 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2940 return false;
2941 }
2942
2943 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2944 if (alloc == nullptr) {
2945 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2946 return false;
2947 }
2948
2949 null_check(receiver);
2950 if (stopped()) {
2951 return true;
2952 }
2953
2954 // Unset the larval bit in the object header
2955 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
2956 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
2957 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
2958
2959 // We must ensure that the buffer is properly published
2960 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
2961 assert(!type->maybe_null(), "result of an allocation should not be null");
2962 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
2963 return true;
2964 }
2965
2966 //----------------------------inline_unsafe_load_store----------------------------
2967 // This method serves a couple of different customers (depending on LoadStoreKind):
2968 //
2969 // LS_cmp_swap:
2970 //
2971 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2972 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2973 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2974 //
2975 // LS_cmp_swap_weak:
2976 //
2977 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2978 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2979 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2980 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2981 //
2982 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2983 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2984 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2985 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2986 //
2987 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2988 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2989 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2990 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2991 //
2992 // LS_cmp_exchange:
2993 //
2994 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2995 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2996 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2997 //
2998 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2999 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
3000 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
3001 //
3002 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
3003 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
3004 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
3005 //
3006 // LS_get_add:
3007 //
3008 // int getAndAddInt( Object o, long offset, int delta)
3009 // long getAndAddLong(Object o, long offset, long delta)
3010 //
3011 // LS_get_set:
3012 //
3013 // int getAndSet(Object o, long offset, int newValue)
3014 // long getAndSet(Object o, long offset, long newValue)
3015 // Object getAndSet(Object o, long offset, Object newValue)
3016 //
3017 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
3018 // This basic scheme here is the same as inline_unsafe_access, but
3019 // differs in enough details that combining them would make the code
3020 // overly confusing. (This is a true fact! I originally combined
3021 // them, but even I was confused by it!) As much code/comments as
3022 // possible are retained from inline_unsafe_access though to make
3023 // the correspondences clearer. - dl
3024
3025 if (callee()->is_static()) return false; // caller must have the capability!
3026
3027 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3028 decorators |= mo_decorator_for_access_kind(access_kind);
3029
3030 #ifndef PRODUCT
3031 BasicType rtype;
3032 {
3033 ResourceMark rm;
3034 // Check the signatures.
3035 ciSignature* sig = callee()->signature();
3036 rtype = sig->return_type()->basic_type();
3037 switch(kind) {
3038 case LS_get_add:
3039 case LS_get_set: {
3040 // Check the signatures.
3041 #ifdef ASSERT
3042 assert(rtype == type, "get and set must return the expected type");
3043 assert(sig->count() == 3, "get and set has 3 arguments");
3044 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3045 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3046 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3047 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3048 #endif // ASSERT
3049 break;
3050 }
3051 case LS_cmp_swap:
3052 case LS_cmp_swap_weak: {
3053 // Check the signatures.
3054 #ifdef ASSERT
3055 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3056 assert(sig->count() == 4, "CAS has 4 arguments");
3057 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3058 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3059 #endif // ASSERT
3060 break;
3061 }
3062 case LS_cmp_exchange: {
3063 // Check the signatures.
3064 #ifdef ASSERT
3065 assert(rtype == type, "CAS must return the expected type");
3066 assert(sig->count() == 4, "CAS has 4 arguments");
3067 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3068 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3069 #endif // ASSERT
3070 break;
3071 }
3072 default:
3073 ShouldNotReachHere();
3074 }
3075 }
3076 #endif //PRODUCT
3077
3078 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3079
3080 // Get arguments:
3081 Node* receiver = nullptr;
3082 Node* base = nullptr;
3083 Node* offset = nullptr;
3084 Node* oldval = nullptr;
3085 Node* newval = nullptr;
3086 switch(kind) {
3087 case LS_cmp_swap:
3088 case LS_cmp_swap_weak:
3089 case LS_cmp_exchange: {
3090 const bool two_slot_type = type2size[type] == 2;
3091 receiver = argument(0); // type: oop
3092 base = argument(1); // type: oop
3093 offset = argument(2); // type: long
3094 oldval = argument(4); // type: oop, int, or long
3095 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3096 break;
3097 }
3098 case LS_get_add:
3099 case LS_get_set: {
3100 receiver = argument(0); // type: oop
3101 base = argument(1); // type: oop
3102 offset = argument(2); // type: long
3103 oldval = nullptr;
3104 newval = argument(4); // type: oop, int, or long
3105 break;
3106 }
3107 default:
3108 ShouldNotReachHere();
3109 }
3110
3111 // Build field offset expression.
3112 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3113 // to be plain byte offsets, which are also the same as those accepted
3114 // by oopDesc::field_addr.
3115 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3116 // 32-bit machines ignore the high half of long offsets
3117 offset = ConvL2X(offset);
3118 // Save state and restore on bailout
3119 uint old_sp = sp();
3120 SafePointNode* old_map = clone_map();
3121 Node* adr = make_unsafe_address(base, offset,type, false);
3122 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3123
3124 Compile::AliasType* alias_type = C->alias_type(adr_type);
3125 BasicType bt = alias_type->basic_type();
3126 if (bt != T_ILLEGAL &&
3127 (is_reference_type(bt) != (type == T_OBJECT))) {
3128 // Don't intrinsify mismatched object accesses.
3129 set_map(old_map);
3130 set_sp(old_sp);
3131 return false;
3132 }
3133
3134 destruct_map_clone(old_map);
3135
3136 // For CAS, unlike inline_unsafe_access, there seems no point in
3137 // trying to refine types. Just use the coarse types here.
3138 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3139 const Type *value_type = Type::get_const_basic_type(type);
3140
3141 switch (kind) {
3142 case LS_get_set:
3143 case LS_cmp_exchange: {
3144 if (type == T_OBJECT) {
3145 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3146 if (tjp != nullptr) {
3147 value_type = tjp;
3148 }
3149 }
3150 break;
3151 }
3152 case LS_cmp_swap:
3153 case LS_cmp_swap_weak:
3154 case LS_get_add:
3155 break;
3156 default:
3157 ShouldNotReachHere();
3158 }
3159
3160 // Null check receiver.
3161 receiver = null_check(receiver);
3162 if (stopped()) {
3163 return true;
3164 }
3165
3166 int alias_idx = C->get_alias_index(adr_type);
3167
3168 if (is_reference_type(type)) {
3169 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3170
3171 if (oldval != nullptr && oldval->is_InlineType()) {
3172 // Re-execute the unsafe access if allocation triggers deoptimization.
3173 PreserveReexecuteState preexecs(this);
3174 jvms()->set_should_reexecute(true);
3175 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3176 }
3177 if (newval != nullptr && newval->is_InlineType()) {
3178 // Re-execute the unsafe access if allocation triggers deoptimization.
3179 PreserveReexecuteState preexecs(this);
3180 jvms()->set_should_reexecute(true);
3181 newval = newval->as_InlineType()->buffer(this)->get_oop();
3182 }
3183
3184 // Transformation of a value which could be null pointer (CastPP #null)
3185 // could be delayed during Parse (for example, in adjust_map_after_if()).
3186 // Execute transformation here to avoid barrier generation in such case.
3187 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3188 newval = _gvn.makecon(TypePtr::NULL_PTR);
3189
3190 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3191 // Refine the value to a null constant, when it is known to be null
3192 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3193 }
3194 }
3195
3196 Node* result = nullptr;
3197 switch (kind) {
3198 case LS_cmp_exchange: {
3199 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3200 oldval, newval, value_type, type, decorators);
3201 break;
3202 }
3203 case LS_cmp_swap_weak:
3204 decorators |= C2_WEAK_CMPXCHG;
3205 case LS_cmp_swap: {
3206 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3207 oldval, newval, value_type, type, decorators);
3208 break;
3209 }
3210 case LS_get_set: {
3211 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3212 newval, value_type, type, decorators);
3213 break;
3214 }
3215 case LS_get_add: {
3216 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3217 newval, value_type, type, decorators);
3218 break;
3219 }
3220 default:
3221 ShouldNotReachHere();
3222 }
3223
3224 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3225 set_result(result);
3226 return true;
3227 }
3228
3229 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3230 // Regardless of form, don't allow previous ld/st to move down,
3231 // then issue acquire, release, or volatile mem_bar.
3232 insert_mem_bar(Op_MemBarCPUOrder);
3233 switch(id) {
3234 case vmIntrinsics::_loadFence:
3235 insert_mem_bar(Op_LoadFence);
3236 return true;
3237 case vmIntrinsics::_storeFence:
3238 insert_mem_bar(Op_StoreFence);
3239 return true;
3240 case vmIntrinsics::_storeStoreFence:
3241 insert_mem_bar(Op_StoreStoreFence);
3242 return true;
3243 case vmIntrinsics::_fullFence:
3244 insert_mem_bar(Op_MemBarVolatile);
3245 return true;
3246 default:
3247 fatal_unexpected_iid(id);
3248 return false;
3249 }
3250 }
3251
3252 bool LibraryCallKit::inline_onspinwait() {
3253 insert_mem_bar(Op_OnSpinWait);
3254 return true;
3255 }
3256
3257 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3258 if (!kls->is_Con()) {
3259 return true;
3260 }
3261 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3262 if (klsptr == nullptr) {
3263 return true;
3264 }
3265 ciInstanceKlass* ik = klsptr->instance_klass();
3266 // don't need a guard for a klass that is already initialized
3267 return !ik->is_initialized();
3268 }
3269
3270 //----------------------------inline_unsafe_writeback0-------------------------
3271 // public native void Unsafe.writeback0(long address)
3272 bool LibraryCallKit::inline_unsafe_writeback0() {
3273 if (!Matcher::has_match_rule(Op_CacheWB)) {
3274 return false;
3275 }
3276 #ifndef PRODUCT
3277 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3278 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3279 ciSignature* sig = callee()->signature();
3280 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3281 #endif
3282 null_check_receiver(); // null-check, then ignore
3283 Node *addr = argument(1);
3284 addr = new CastX2PNode(addr);
3285 addr = _gvn.transform(addr);
3286 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3287 flush = _gvn.transform(flush);
3288 set_memory(flush, TypeRawPtr::BOTTOM);
3289 return true;
3290 }
3291
3292 //----------------------------inline_unsafe_writeback0-------------------------
3293 // public native void Unsafe.writeback0(long address)
3294 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3295 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3296 return false;
3297 }
3298 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3299 return false;
3300 }
3301 #ifndef PRODUCT
3302 assert(Matcher::has_match_rule(Op_CacheWB),
3303 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3304 : "found match rule for CacheWBPostSync but not CacheWB"));
3305
3306 #endif
3307 null_check_receiver(); // null-check, then ignore
3308 Node *sync;
3309 if (is_pre) {
3310 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3311 } else {
3312 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3313 }
3314 sync = _gvn.transform(sync);
3315 set_memory(sync, TypeRawPtr::BOTTOM);
3316 return true;
3317 }
3318
3319 //----------------------------inline_unsafe_allocate---------------------------
3320 // public native Object Unsafe.allocateInstance(Class<?> cls);
3321 bool LibraryCallKit::inline_unsafe_allocate() {
3322
3323 #if INCLUDE_JVMTI
3324 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3325 return false;
3326 }
3327 #endif //INCLUDE_JVMTI
3328
3329 if (callee()->is_static()) return false; // caller must have the capability!
3330
3331 null_check_receiver(); // null-check, then ignore
3332 Node* cls = null_check(argument(1));
3333 if (stopped()) return true;
3334
3335 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3336 kls = null_check(kls);
3337 if (stopped()) return true; // argument was like int.class
3338
3339 #if INCLUDE_JVMTI
3340 // Don't try to access new allocated obj in the intrinsic.
3341 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3342 // Deoptimize and allocate in interpreter instead.
3343 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3344 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3345 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3346 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3347 {
3348 BuildCutout unless(this, tst, PROB_MAX);
3349 uncommon_trap(Deoptimization::Reason_intrinsic,
3350 Deoptimization::Action_make_not_entrant);
3351 }
3352 if (stopped()) {
3353 return true;
3354 }
3355 #endif //INCLUDE_JVMTI
3356
3357 Node* test = nullptr;
3358 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3359 // Note: The argument might still be an illegal value like
3360 // Serializable.class or Object[].class. The runtime will handle it.
3361 // But we must make an explicit check for initialization.
3362 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3363 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3364 // can generate code to load it as unsigned byte.
3365 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3366 Node* bits = intcon(InstanceKlass::fully_initialized);
3367 test = _gvn.transform(new SubINode(inst, bits));
3368 // The 'test' is non-zero if we need to take a slow path.
3369 }
3370 Node* obj = nullptr;
3371 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3372 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3373 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3374 } else {
3375 obj = new_instance(kls, test);
3376 }
3377 set_result(obj);
3378 return true;
3379 }
3380
3381 //------------------------inline_native_time_funcs--------------
3382 // inline code for System.currentTimeMillis() and System.nanoTime()
3383 // these have the same type and signature
3384 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3385 const TypeFunc* tf = OptoRuntime::void_long_Type();
3386 const TypePtr* no_memory_effects = nullptr;
3387 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3388 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3389 #ifdef ASSERT
3390 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3391 assert(value_top == top(), "second value must be top");
3392 #endif
3393 set_result(value);
3394 return true;
3395 }
3396
3397
3398 #if INCLUDE_JVMTI
3399
3400 // When notifications are disabled then just update the VTMS transition bit and return.
3401 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3402 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3403 if (!DoJVMTIVirtualThreadTransitions) {
3404 return true;
3405 }
3406 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3407 IdealKit ideal(this);
3408
3409 Node* ONE = ideal.ConI(1);
3410 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3411 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3412 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3413
3414 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3415 sync_kit(ideal);
3416 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3417 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3418 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3419 ideal.sync_kit(this);
3420 } ideal.else_(); {
3421 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3422 Node* thread = ideal.thread();
3423 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3424 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3425
3426 sync_kit(ideal);
3427 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3428 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3429
3430 ideal.sync_kit(this);
3431 } ideal.end_if();
3432 final_sync(ideal);
3433
3434 return true;
3435 }
3436
3437 // Always update the is_disable_suspend bit.
3438 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3439 if (!DoJVMTIVirtualThreadTransitions) {
3440 return true;
3441 }
3442 IdealKit ideal(this);
3443
3444 {
3445 // unconditionally update the is_disable_suspend bit in current JavaThread
3446 Node* thread = ideal.thread();
3447 Node* arg = _gvn.transform(argument(0)); // argument for notification
3448 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3449 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3450
3451 sync_kit(ideal);
3452 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3453 ideal.sync_kit(this);
3454 }
3455 final_sync(ideal);
3456
3457 return true;
3458 }
3459
3460 #endif // INCLUDE_JVMTI
3461
3462 #ifdef JFR_HAVE_INTRINSICS
3463
3464 /**
3465 * if oop->klass != null
3466 * // normal class
3467 * epoch = _epoch_state ? 2 : 1
3468 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3469 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3470 * }
3471 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3472 * else
3473 * // primitive class
3474 * if oop->array_klass != null
3475 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3476 * else
3477 * id = LAST_TYPE_ID + 1 // void class path
3478 * if (!signaled)
3479 * signaled = true
3480 */
3481 bool LibraryCallKit::inline_native_classID() {
3482 Node* cls = argument(0);
3483
3484 IdealKit ideal(this);
3485 #define __ ideal.
3486 IdealVariable result(ideal); __ declarations_done();
3487 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3488 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3489 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3490
3491
3492 __ if_then(kls, BoolTest::ne, null()); {
3493 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3494 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3495
3496 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3497 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3498 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3499 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3500 mask = _gvn.transform(new OrLNode(mask, epoch));
3501 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3502
3503 float unlikely = PROB_UNLIKELY(0.999);
3504 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3505 sync_kit(ideal);
3506 make_runtime_call(RC_LEAF,
3507 OptoRuntime::class_id_load_barrier_Type(),
3508 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3509 "class id load barrier",
3510 TypePtr::BOTTOM,
3511 kls);
3512 ideal.sync_kit(this);
3513 } __ end_if();
3514
3515 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3516 } __ else_(); {
3517 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3518 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3519 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3520 __ if_then(array_kls, BoolTest::ne, null()); {
3521 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3522 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3523 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3524 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3525 } __ else_(); {
3526 // void class case
3527 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3528 } __ end_if();
3529
3530 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3531 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3532 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3533 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3534 } __ end_if();
3535 } __ end_if();
3536
3537 final_sync(ideal);
3538 set_result(ideal.value(result));
3539 #undef __
3540 return true;
3541 }
3542
3543 //------------------------inline_native_jvm_commit------------------
3544 bool LibraryCallKit::inline_native_jvm_commit() {
3545 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3546
3547 // Save input memory and i_o state.
3548 Node* input_memory_state = reset_memory();
3549 set_all_memory(input_memory_state);
3550 Node* input_io_state = i_o();
3551
3552 // TLS.
3553 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3554 // Jfr java buffer.
3555 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3556 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3557 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3558
3559 // Load the current value of the notified field in the JfrThreadLocal.
3560 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3561 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3562
3563 // Test for notification.
3564 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3565 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3566 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3567
3568 // True branch, is notified.
3569 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3570 set_control(is_notified);
3571
3572 // Reset notified state.
3573 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3574 Node* notified_reset_memory = reset_memory();
3575
3576 // 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.
3577 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3578 // Convert the machine-word to a long.
3579 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3580
3581 // False branch, not notified.
3582 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3583 set_control(not_notified);
3584 set_all_memory(input_memory_state);
3585
3586 // Arg is the next position as a long.
3587 Node* arg = argument(0);
3588 // Convert long to machine-word.
3589 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3590
3591 // Store the next_position to the underlying jfr java buffer.
3592 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3593
3594 Node* commit_memory = reset_memory();
3595 set_all_memory(commit_memory);
3596
3597 // 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.
3598 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3599 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3600 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3601
3602 // And flags with lease constant.
3603 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3604
3605 // Branch on lease to conditionalize returning the leased java buffer.
3606 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3607 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3608 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3609
3610 // False branch, not a lease.
3611 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3612
3613 // True branch, is lease.
3614 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3615 set_control(is_lease);
3616
3617 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3618 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3619 OptoRuntime::void_void_Type(),
3620 SharedRuntime::jfr_return_lease(),
3621 "return_lease", TypePtr::BOTTOM);
3622 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3623
3624 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3625 record_for_igvn(lease_compare_rgn);
3626 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3627 record_for_igvn(lease_compare_mem);
3628 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3629 record_for_igvn(lease_compare_io);
3630 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3631 record_for_igvn(lease_result_value);
3632
3633 // Update control and phi nodes.
3634 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3635 lease_compare_rgn->init_req(_false_path, not_lease);
3636
3637 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3638 lease_compare_mem->init_req(_false_path, commit_memory);
3639
3640 lease_compare_io->init_req(_true_path, i_o());
3641 lease_compare_io->init_req(_false_path, input_io_state);
3642
3643 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3644 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3645
3646 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3647 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3648 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3649 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3650
3651 // Update control and phi nodes.
3652 result_rgn->init_req(_true_path, is_notified);
3653 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3654
3655 result_mem->init_req(_true_path, notified_reset_memory);
3656 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3657
3658 result_io->init_req(_true_path, input_io_state);
3659 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3660
3661 result_value->init_req(_true_path, current_pos);
3662 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3663
3664 // Set output state.
3665 set_control(_gvn.transform(result_rgn));
3666 set_all_memory(_gvn.transform(result_mem));
3667 set_i_o(_gvn.transform(result_io));
3668 set_result(result_rgn, result_value);
3669 return true;
3670 }
3671
3672 /*
3673 * The intrinsic is a model of this pseudo-code:
3674 *
3675 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3676 * jobject h_event_writer = tl->java_event_writer();
3677 * if (h_event_writer == nullptr) {
3678 * return nullptr;
3679 * }
3680 * oop threadObj = Thread::threadObj();
3681 * oop vthread = java_lang_Thread::vthread(threadObj);
3682 * traceid tid;
3683 * bool pinVirtualThread;
3684 * bool excluded;
3685 * if (vthread != threadObj) { // i.e. current thread is virtual
3686 * tid = java_lang_Thread::tid(vthread);
3687 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3688 * pinVirtualThread = VMContinuations;
3689 * excluded = vthread_epoch_raw & excluded_mask;
3690 * if (!excluded) {
3691 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3692 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3693 * if (vthread_epoch != current_epoch) {
3694 * write_checkpoint();
3695 * }
3696 * }
3697 * } else {
3698 * tid = java_lang_Thread::tid(threadObj);
3699 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3700 * pinVirtualThread = false;
3701 * excluded = thread_epoch_raw & excluded_mask;
3702 * }
3703 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3704 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3705 * if (tid_in_event_writer != tid) {
3706 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3707 * setField(event_writer, "excluded", excluded);
3708 * setField(event_writer, "threadID", tid);
3709 * }
3710 * return event_writer
3711 */
3712 bool LibraryCallKit::inline_native_getEventWriter() {
3713 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3714
3715 // Save input memory and i_o state.
3716 Node* input_memory_state = reset_memory();
3717 set_all_memory(input_memory_state);
3718 Node* input_io_state = i_o();
3719
3720 // The most significant bit of the u2 is used to denote thread exclusion
3721 Node* excluded_shift = _gvn.intcon(15);
3722 Node* excluded_mask = _gvn.intcon(1 << 15);
3723 // The epoch generation is the range [1-32767]
3724 Node* epoch_mask = _gvn.intcon(32767);
3725
3726 // TLS
3727 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3728
3729 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3730 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3731
3732 // Load the eventwriter jobject handle.
3733 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3734
3735 // Null check the jobject handle.
3736 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3737 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3738 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3739
3740 // False path, jobj is null.
3741 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3742
3743 // True path, jobj is not null.
3744 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3745
3746 set_control(jobj_is_not_null);
3747
3748 // Load the threadObj for the CarrierThread.
3749 Node* threadObj = generate_current_thread(tls_ptr);
3750
3751 // Load the vthread.
3752 Node* vthread = generate_virtual_thread(tls_ptr);
3753
3754 // If vthread != threadObj, this is a virtual thread.
3755 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3756 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3757 IfNode* iff_vthread_not_equal_threadObj =
3758 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3759
3760 // False branch, fallback to threadObj.
3761 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3762 set_control(vthread_equal_threadObj);
3763
3764 // Load the tid field from the vthread object.
3765 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3766
3767 // Load the raw epoch value from the threadObj.
3768 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3769 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3770 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3771 TypeInt::CHAR, T_CHAR,
3772 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3773
3774 // Mask off the excluded information from the epoch.
3775 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3776
3777 // True branch, this is a virtual thread.
3778 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3779 set_control(vthread_not_equal_threadObj);
3780
3781 // Load the tid field from the vthread object.
3782 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3783
3784 // Continuation support determines if a virtual thread should be pinned.
3785 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3786 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3787
3788 // Load the raw epoch value from the vthread.
3789 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3790 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3791 TypeInt::CHAR, T_CHAR,
3792 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3793
3794 // Mask off the excluded information from the epoch.
3795 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3796
3797 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3798 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3799 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3800 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3801
3802 // False branch, vthread is excluded, no need to write epoch info.
3803 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3804
3805 // True branch, vthread is included, update epoch info.
3806 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3807 set_control(included);
3808
3809 // Get epoch value.
3810 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3811
3812 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3813 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3814 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3815
3816 // Compare the epoch in the vthread to the current epoch generation.
3817 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3818 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3819 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3820
3821 // False path, epoch is equal, checkpoint information is valid.
3822 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3823
3824 // True path, epoch is not equal, write a checkpoint for the vthread.
3825 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3826
3827 set_control(epoch_is_not_equal);
3828
3829 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3830 // The call also updates the native thread local thread id and the vthread with the current epoch.
3831 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3832 OptoRuntime::jfr_write_checkpoint_Type(),
3833 SharedRuntime::jfr_write_checkpoint(),
3834 "write_checkpoint", TypePtr::BOTTOM);
3835 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3836
3837 // vthread epoch != current epoch
3838 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3839 record_for_igvn(epoch_compare_rgn);
3840 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3841 record_for_igvn(epoch_compare_mem);
3842 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3843 record_for_igvn(epoch_compare_io);
3844
3845 // Update control and phi nodes.
3846 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3847 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3848 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3849 epoch_compare_mem->init_req(_false_path, input_memory_state);
3850 epoch_compare_io->init_req(_true_path, i_o());
3851 epoch_compare_io->init_req(_false_path, input_io_state);
3852
3853 // excluded != true
3854 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3855 record_for_igvn(exclude_compare_rgn);
3856 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3857 record_for_igvn(exclude_compare_mem);
3858 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3859 record_for_igvn(exclude_compare_io);
3860
3861 // Update control and phi nodes.
3862 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3863 exclude_compare_rgn->init_req(_false_path, excluded);
3864 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3865 exclude_compare_mem->init_req(_false_path, input_memory_state);
3866 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3867 exclude_compare_io->init_req(_false_path, input_io_state);
3868
3869 // vthread != threadObj
3870 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3871 record_for_igvn(vthread_compare_rgn);
3872 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3873 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3874 record_for_igvn(vthread_compare_io);
3875 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3876 record_for_igvn(tid);
3877 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3878 record_for_igvn(exclusion);
3879 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3880 record_for_igvn(pinVirtualThread);
3881
3882 // Update control and phi nodes.
3883 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3884 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3885 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3886 vthread_compare_mem->init_req(_false_path, input_memory_state);
3887 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3888 vthread_compare_io->init_req(_false_path, input_io_state);
3889 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3890 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3891 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3892 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3893 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3894 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3895
3896 // Update branch state.
3897 set_control(_gvn.transform(vthread_compare_rgn));
3898 set_all_memory(_gvn.transform(vthread_compare_mem));
3899 set_i_o(_gvn.transform(vthread_compare_io));
3900
3901 // Load the event writer oop by dereferencing the jobject handle.
3902 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3903 assert(klass_EventWriter->is_loaded(), "invariant");
3904 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3905 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3906 const TypeOopPtr* const xtype = aklass->as_instance_type();
3907 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3908 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3909
3910 // Load the current thread id from the event writer object.
3911 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3912 // Get the field offset to, conditionally, store an updated tid value later.
3913 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3914 // Get the field offset to, conditionally, store an updated exclusion value later.
3915 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3916 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3917 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3918
3919 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3920 record_for_igvn(event_writer_tid_compare_rgn);
3921 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3922 record_for_igvn(event_writer_tid_compare_mem);
3923 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3924 record_for_igvn(event_writer_tid_compare_io);
3925
3926 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3927 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3928 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3929 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3930
3931 // False path, tids are the same.
3932 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3933
3934 // True path, tid is not equal, need to update the tid in the event writer.
3935 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3936 record_for_igvn(tid_is_not_equal);
3937
3938 // Store the pin state to the event writer.
3939 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3940
3941 // Store the exclusion state to the event writer.
3942 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3943 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3944
3945 // Store the tid to the event writer.
3946 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3947
3948 // Update control and phi nodes.
3949 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3950 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3951 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3952 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3953 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3954 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3955
3956 // Result of top level CFG, Memory, IO and Value.
3957 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3958 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3959 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3960 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3961
3962 // Result control.
3963 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3964 result_rgn->init_req(_false_path, jobj_is_null);
3965
3966 // Result memory.
3967 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3968 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3969
3970 // Result IO.
3971 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3972 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3973
3974 // Result value.
3975 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3976 result_value->init_req(_false_path, null()); // return null
3977
3978 // Set output state.
3979 set_control(_gvn.transform(result_rgn));
3980 set_all_memory(_gvn.transform(result_mem));
3981 set_i_o(_gvn.transform(result_io));
3982 set_result(result_rgn, result_value);
3983 return true;
3984 }
3985
3986 /*
3987 * The intrinsic is a model of this pseudo-code:
3988 *
3989 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3990 * if (carrierThread != thread) { // is virtual thread
3991 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3992 * bool excluded = vthread_epoch_raw & excluded_mask;
3993 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3994 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3995 * if (!excluded) {
3996 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3997 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3998 * }
3999 * Atomic::release_store(&tl->_vthread, true);
4000 * return;
4001 * }
4002 * Atomic::release_store(&tl->_vthread, false);
4003 */
4004 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4005 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4006
4007 Node* input_memory_state = reset_memory();
4008 set_all_memory(input_memory_state);
4009
4010 // The most significant bit of the u2 is used to denote thread exclusion
4011 Node* excluded_mask = _gvn.intcon(1 << 15);
4012 // The epoch generation is the range [1-32767]
4013 Node* epoch_mask = _gvn.intcon(32767);
4014
4015 Node* const carrierThread = generate_current_thread(jt);
4016 // If thread != carrierThread, this is a virtual thread.
4017 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4018 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4019 IfNode* iff_thread_not_equal_carrierThread =
4020 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4021
4022 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4023
4024 // False branch, is carrierThread.
4025 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4026 // Store release
4027 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4028
4029 set_all_memory(input_memory_state);
4030
4031 // True branch, is virtual thread.
4032 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4033 set_control(thread_not_equal_carrierThread);
4034
4035 // Load the raw epoch value from the vthread.
4036 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4037 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4038 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4039
4040 // Mask off the excluded information from the epoch.
4041 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4042
4043 // Load the tid field from the thread.
4044 Node* tid = load_field_from_object(thread, "tid", "J");
4045
4046 // Store the vthread tid to the jfr thread local.
4047 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4048 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4049
4050 // Branch is_excluded to conditionalize updating the epoch .
4051 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4052 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4053 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4054
4055 // True branch, vthread is excluded, no need to write epoch info.
4056 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4057 set_control(excluded);
4058 Node* vthread_is_excluded = _gvn.intcon(1);
4059
4060 // False branch, vthread is included, update epoch info.
4061 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4062 set_control(included);
4063 Node* vthread_is_included = _gvn.intcon(0);
4064
4065 // Get epoch value.
4066 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4067
4068 // Store the vthread epoch to the jfr thread local.
4069 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4070 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4071
4072 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4073 record_for_igvn(excluded_rgn);
4074 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4075 record_for_igvn(excluded_mem);
4076 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4077 record_for_igvn(exclusion);
4078
4079 // Merge the excluded control and memory.
4080 excluded_rgn->init_req(_true_path, excluded);
4081 excluded_rgn->init_req(_false_path, included);
4082 excluded_mem->init_req(_true_path, tid_memory);
4083 excluded_mem->init_req(_false_path, included_memory);
4084 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4085 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4086
4087 // Set intermediate state.
4088 set_control(_gvn.transform(excluded_rgn));
4089 set_all_memory(excluded_mem);
4090
4091 // Store the vthread exclusion state to the jfr thread local.
4092 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4093 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4094
4095 // Store release
4096 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4097
4098 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4099 record_for_igvn(thread_compare_rgn);
4100 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4101 record_for_igvn(thread_compare_mem);
4102 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4103 record_for_igvn(vthread);
4104
4105 // Merge the thread_compare control and memory.
4106 thread_compare_rgn->init_req(_true_path, control());
4107 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4108 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4109 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4110
4111 // Set output state.
4112 set_control(_gvn.transform(thread_compare_rgn));
4113 set_all_memory(_gvn.transform(thread_compare_mem));
4114 }
4115
4116 #endif // JFR_HAVE_INTRINSICS
4117
4118 //------------------------inline_native_currentCarrierThread------------------
4119 bool LibraryCallKit::inline_native_currentCarrierThread() {
4120 Node* junk = nullptr;
4121 set_result(generate_current_thread(junk));
4122 return true;
4123 }
4124
4125 //------------------------inline_native_currentThread------------------
4126 bool LibraryCallKit::inline_native_currentThread() {
4127 Node* junk = nullptr;
4128 set_result(generate_virtual_thread(junk));
4129 return true;
4130 }
4131
4132 //------------------------inline_native_setVthread------------------
4133 bool LibraryCallKit::inline_native_setCurrentThread() {
4134 assert(C->method()->changes_current_thread(),
4135 "method changes current Thread but is not annotated ChangesCurrentThread");
4136 Node* arr = argument(1);
4137 Node* thread = _gvn.transform(new ThreadLocalNode());
4138 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4139 Node* thread_obj_handle
4140 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4141 thread_obj_handle = _gvn.transform(thread_obj_handle);
4142 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4143 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4144
4145 // Change the _monitor_owner_id of the JavaThread
4146 Node* tid = load_field_from_object(arr, "tid", "J");
4147 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4148 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4149
4150 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4151 return true;
4152 }
4153
4154 const Type* LibraryCallKit::scopedValueCache_type() {
4155 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4156 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4157 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
4158
4159 // Because we create the scopedValue cache lazily we have to make the
4160 // type of the result BotPTR.
4161 bool xk = etype->klass_is_exact();
4162 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4163 return objects_type;
4164 }
4165
4166 Node* LibraryCallKit::scopedValueCache_helper() {
4167 Node* thread = _gvn.transform(new ThreadLocalNode());
4168 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4169 // We cannot use immutable_memory() because we might flip onto a
4170 // different carrier thread, at which point we'll need to use that
4171 // carrier thread's cache.
4172 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4173 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4174 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4175 }
4176
4177 //------------------------inline_native_scopedValueCache------------------
4178 bool LibraryCallKit::inline_native_scopedValueCache() {
4179 Node* cache_obj_handle = scopedValueCache_helper();
4180 const Type* objects_type = scopedValueCache_type();
4181 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4182
4183 return true;
4184 }
4185
4186 //------------------------inline_native_setScopedValueCache------------------
4187 bool LibraryCallKit::inline_native_setScopedValueCache() {
4188 Node* arr = argument(0);
4189 Node* cache_obj_handle = scopedValueCache_helper();
4190 const Type* objects_type = scopedValueCache_type();
4191
4192 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4193 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4194
4195 return true;
4196 }
4197
4198 //------------------------inline_native_Continuation_pin and unpin-----------
4199
4200 // Shared implementation routine for both pin and unpin.
4201 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4202 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4203
4204 // Save input memory.
4205 Node* input_memory_state = reset_memory();
4206 set_all_memory(input_memory_state);
4207
4208 // TLS
4209 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4210 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4211 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4212
4213 // Null check the last continuation object.
4214 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4215 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4216 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4217
4218 // False path, last continuation is null.
4219 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4220
4221 // True path, last continuation is not null.
4222 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4223
4224 set_control(continuation_is_not_null);
4225
4226 // Load the pin count from the last continuation.
4227 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4228 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4229
4230 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4231 Node* pin_count_rhs;
4232 if (unpin) {
4233 pin_count_rhs = _gvn.intcon(0);
4234 } else {
4235 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4236 }
4237 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4238 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4239 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4240
4241 // True branch, pin count over/underflow.
4242 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4243 {
4244 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4245 // which will throw IllegalStateException for pin count over/underflow.
4246 // No memory changed so far - we can use memory create by reset_memory()
4247 // at the beginning of this intrinsic. No need to call reset_memory() again.
4248 PreserveJVMState pjvms(this);
4249 set_control(pin_count_over_underflow);
4250 uncommon_trap(Deoptimization::Reason_intrinsic,
4251 Deoptimization::Action_none);
4252 assert(stopped(), "invariant");
4253 }
4254
4255 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4256 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4257 set_control(valid_pin_count);
4258
4259 Node* next_pin_count;
4260 if (unpin) {
4261 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4262 } else {
4263 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4264 }
4265
4266 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4267
4268 // Result of top level CFG and Memory.
4269 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4270 record_for_igvn(result_rgn);
4271 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4272 record_for_igvn(result_mem);
4273
4274 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4275 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4276 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4277 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4278
4279 // Set output state.
4280 set_control(_gvn.transform(result_rgn));
4281 set_all_memory(_gvn.transform(result_mem));
4282
4283 return true;
4284 }
4285
4286 //-----------------------load_klass_from_mirror_common-------------------------
4287 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4288 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4289 // and branch to the given path on the region.
4290 // If never_see_null, take an uncommon trap on null, so we can optimistically
4291 // compile for the non-null case.
4292 // If the region is null, force never_see_null = true.
4293 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4294 bool never_see_null,
4295 RegionNode* region,
4296 int null_path,
4297 int offset) {
4298 if (region == nullptr) never_see_null = true;
4299 Node* p = basic_plus_adr(mirror, offset);
4300 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4301 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4302 Node* null_ctl = top();
4303 kls = null_check_oop(kls, &null_ctl, never_see_null);
4304 if (region != nullptr) {
4305 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4306 region->init_req(null_path, null_ctl);
4307 } else {
4308 assert(null_ctl == top(), "no loose ends");
4309 }
4310 return kls;
4311 }
4312
4313 //--------------------(inline_native_Class_query helpers)---------------------
4314 // Use this for JVM_ACC_INTERFACE.
4315 // Fall through if (mods & mask) == bits, take the guard otherwise.
4316 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4317 ByteSize offset, const Type* type, BasicType bt) {
4318 // Branch around if the given klass has the given modifier bit set.
4319 // Like generate_guard, adds a new path onto the region.
4320 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4321 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4322 Node* mask = intcon(modifier_mask);
4323 Node* bits = intcon(modifier_bits);
4324 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4325 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4326 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4327 return generate_fair_guard(bol, region);
4328 }
4329
4330 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4331 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4332 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4333 }
4334
4335 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4336 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4337 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4338 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4339 }
4340
4341 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4342 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4343 }
4344
4345 //-------------------------inline_native_Class_query-------------------
4346 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4347 const Type* return_type = TypeInt::BOOL;
4348 Node* prim_return_value = top(); // what happens if it's a primitive class?
4349 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4350 bool expect_prim = false; // most of these guys expect to work on refs
4351
4352 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4353
4354 Node* mirror = argument(0);
4355 Node* obj = top();
4356
4357 switch (id) {
4358 case vmIntrinsics::_isInstance:
4359 // nothing is an instance of a primitive type
4360 prim_return_value = intcon(0);
4361 obj = argument(1);
4362 break;
4363 case vmIntrinsics::_isHidden:
4364 prim_return_value = intcon(0);
4365 break;
4366 case vmIntrinsics::_getSuperclass:
4367 prim_return_value = null();
4368 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4369 break;
4370 case vmIntrinsics::_getClassAccessFlags:
4371 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
4372 return_type = TypeInt::CHAR;
4373 break;
4374 default:
4375 fatal_unexpected_iid(id);
4376 break;
4377 }
4378
4379 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4380 if (mirror_con == nullptr) return false; // cannot happen?
4381
4382 #ifndef PRODUCT
4383 if (C->print_intrinsics() || C->print_inlining()) {
4384 ciType* k = mirror_con->java_mirror_type();
4385 if (k) {
4386 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4387 k->print_name();
4388 tty->cr();
4389 }
4390 }
4391 #endif
4392
4393 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4394 RegionNode* region = new RegionNode(PATH_LIMIT);
4395 record_for_igvn(region);
4396 PhiNode* phi = new PhiNode(region, return_type);
4397
4398 // The mirror will never be null of Reflection.getClassAccessFlags, however
4399 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4400 // if it is. See bug 4774291.
4401
4402 // For Reflection.getClassAccessFlags(), the null check occurs in
4403 // the wrong place; see inline_unsafe_access(), above, for a similar
4404 // situation.
4405 mirror = null_check(mirror);
4406 // If mirror or obj is dead, only null-path is taken.
4407 if (stopped()) return true;
4408
4409 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4410
4411 // Now load the mirror's klass metaobject, and null-check it.
4412 // Side-effects region with the control path if the klass is null.
4413 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4414 // If kls is null, we have a primitive mirror.
4415 phi->init_req(_prim_path, prim_return_value);
4416 if (stopped()) { set_result(region, phi); return true; }
4417 bool safe_for_replace = (region->in(_prim_path) == top());
4418
4419 Node* p; // handy temp
4420 Node* null_ctl;
4421
4422 // Now that we have the non-null klass, we can perform the real query.
4423 // For constant classes, the query will constant-fold in LoadNode::Value.
4424 Node* query_value = top();
4425 switch (id) {
4426 case vmIntrinsics::_isInstance:
4427 // nothing is an instance of a primitive type
4428 query_value = gen_instanceof(obj, kls, safe_for_replace);
4429 break;
4430
4431 case vmIntrinsics::_isHidden:
4432 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4433 if (generate_hidden_class_guard(kls, region) != nullptr)
4434 // A guard was added. If the guard is taken, it was an hidden class.
4435 phi->add_req(intcon(1));
4436 // If we fall through, it's a plain class.
4437 query_value = intcon(0);
4438 break;
4439
4440
4441 case vmIntrinsics::_getSuperclass:
4442 // The rules here are somewhat unfortunate, but we can still do better
4443 // with random logic than with a JNI call.
4444 // Interfaces store null or Object as _super, but must report null.
4445 // Arrays store an intermediate super as _super, but must report Object.
4446 // Other types can report the actual _super.
4447 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4448 if (generate_interface_guard(kls, region) != nullptr)
4449 // A guard was added. If the guard is taken, it was an interface.
4450 phi->add_req(null());
4451 if (generate_array_guard(kls, region) != nullptr)
4452 // A guard was added. If the guard is taken, it was an array.
4453 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4454 // If we fall through, it's a plain class. Get its _super.
4455 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4456 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4457 null_ctl = top();
4458 kls = null_check_oop(kls, &null_ctl);
4459 if (null_ctl != top()) {
4460 // If the guard is taken, Object.superClass is null (both klass and mirror).
4461 region->add_req(null_ctl);
4462 phi ->add_req(null());
4463 }
4464 if (!stopped()) {
4465 query_value = load_mirror_from_klass(kls);
4466 }
4467 break;
4468
4469 case vmIntrinsics::_getClassAccessFlags:
4470 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4471 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4472 break;
4473
4474 default:
4475 fatal_unexpected_iid(id);
4476 break;
4477 }
4478
4479 // Fall-through is the normal case of a query to a real class.
4480 phi->init_req(1, query_value);
4481 region->init_req(1, control());
4482
4483 C->set_has_split_ifs(true); // Has chance for split-if optimization
4484 set_result(region, phi);
4485 return true;
4486 }
4487
4488
4489 //-------------------------inline_Class_cast-------------------
4490 bool LibraryCallKit::inline_Class_cast() {
4491 Node* mirror = argument(0); // Class
4492 Node* obj = argument(1);
4493 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4494 if (mirror_con == nullptr) {
4495 return false; // dead path (mirror->is_top()).
4496 }
4497 if (obj == nullptr || obj->is_top()) {
4498 return false; // dead path
4499 }
4500 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4501
4502 // First, see if Class.cast() can be folded statically.
4503 // java_mirror_type() returns non-null for compile-time Class constants.
4504 ciType* tm = mirror_con->java_mirror_type();
4505 if (tm != nullptr && tm->is_klass() &&
4506 tp != nullptr) {
4507 if (!tp->is_loaded()) {
4508 // Don't use intrinsic when class is not loaded.
4509 return false;
4510 } else {
4511 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4512 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4513 if (static_res == Compile::SSC_always_true) {
4514 // isInstance() is true - fold the code.
4515 set_result(obj);
4516 return true;
4517 } else if (static_res == Compile::SSC_always_false) {
4518 // Don't use intrinsic, have to throw ClassCastException.
4519 // If the reference is null, the non-intrinsic bytecode will
4520 // be optimized appropriately.
4521 return false;
4522 }
4523 }
4524 }
4525
4526 // Bailout intrinsic and do normal inlining if exception path is frequent.
4527 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4528 return false;
4529 }
4530
4531 // Generate dynamic checks.
4532 // Class.cast() is java implementation of _checkcast bytecode.
4533 // Do checkcast (Parse::do_checkcast()) optimizations here.
4534
4535 mirror = null_check(mirror);
4536 // If mirror is dead, only null-path is taken.
4537 if (stopped()) {
4538 return true;
4539 }
4540
4541 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4542 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4543 RegionNode* region = new RegionNode(PATH_LIMIT);
4544 record_for_igvn(region);
4545
4546 // Now load the mirror's klass metaobject, and null-check it.
4547 // If kls is null, we have a primitive mirror and
4548 // nothing is an instance of a primitive type.
4549 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4550
4551 Node* res = top();
4552 Node* io = i_o();
4553 Node* mem = merged_memory();
4554 if (!stopped()) {
4555
4556 Node* bad_type_ctrl = top();
4557 // Do checkcast optimizations.
4558 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4559 region->init_req(_bad_type_path, bad_type_ctrl);
4560 }
4561 if (region->in(_prim_path) != top() ||
4562 region->in(_bad_type_path) != top() ||
4563 region->in(_npe_path) != top()) {
4564 // Let Interpreter throw ClassCastException.
4565 PreserveJVMState pjvms(this);
4566 set_control(_gvn.transform(region));
4567 // Set IO and memory because gen_checkcast may override them when buffering inline types
4568 set_i_o(io);
4569 set_all_memory(mem);
4570 uncommon_trap(Deoptimization::Reason_intrinsic,
4571 Deoptimization::Action_maybe_recompile);
4572 }
4573 if (!stopped()) {
4574 set_result(res);
4575 }
4576 return true;
4577 }
4578
4579
4580 //--------------------------inline_native_subtype_check------------------------
4581 // This intrinsic takes the JNI calls out of the heart of
4582 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4583 bool LibraryCallKit::inline_native_subtype_check() {
4584 // Pull both arguments off the stack.
4585 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4586 args[0] = argument(0);
4587 args[1] = argument(1);
4588 Node* klasses[2]; // corresponding Klasses: superk, subk
4589 klasses[0] = klasses[1] = top();
4590
4591 enum {
4592 // A full decision tree on {superc is prim, subc is prim}:
4593 _prim_0_path = 1, // {P,N} => false
4594 // {P,P} & superc!=subc => false
4595 _prim_same_path, // {P,P} & superc==subc => true
4596 _prim_1_path, // {N,P} => false
4597 _ref_subtype_path, // {N,N} & subtype check wins => true
4598 _both_ref_path, // {N,N} & subtype check loses => false
4599 PATH_LIMIT
4600 };
4601
4602 RegionNode* region = new RegionNode(PATH_LIMIT);
4603 RegionNode* prim_region = new RegionNode(2);
4604 Node* phi = new PhiNode(region, TypeInt::BOOL);
4605 record_for_igvn(region);
4606 record_for_igvn(prim_region);
4607
4608 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4609 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4610 int class_klass_offset = java_lang_Class::klass_offset();
4611
4612 // First null-check both mirrors and load each mirror's klass metaobject.
4613 int which_arg;
4614 for (which_arg = 0; which_arg <= 1; which_arg++) {
4615 Node* arg = args[which_arg];
4616 arg = null_check(arg);
4617 if (stopped()) break;
4618 args[which_arg] = arg;
4619
4620 Node* p = basic_plus_adr(arg, class_klass_offset);
4621 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4622 klasses[which_arg] = _gvn.transform(kls);
4623 }
4624
4625 // Having loaded both klasses, test each for null.
4626 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4627 for (which_arg = 0; which_arg <= 1; which_arg++) {
4628 Node* kls = klasses[which_arg];
4629 Node* null_ctl = top();
4630 kls = null_check_oop(kls, &null_ctl, never_see_null);
4631 if (which_arg == 0) {
4632 prim_region->init_req(1, null_ctl);
4633 } else {
4634 region->init_req(_prim_1_path, null_ctl);
4635 }
4636 if (stopped()) break;
4637 klasses[which_arg] = kls;
4638 }
4639
4640 if (!stopped()) {
4641 // now we have two reference types, in klasses[0..1]
4642 Node* subk = klasses[1]; // the argument to isAssignableFrom
4643 Node* superk = klasses[0]; // the receiver
4644 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4645 region->set_req(_ref_subtype_path, control());
4646 }
4647
4648 // If both operands are primitive (both klasses null), then
4649 // we must return true when they are identical primitives.
4650 // It is convenient to test this after the first null klass check.
4651 // This path is also used if superc is a value mirror.
4652 set_control(_gvn.transform(prim_region));
4653 if (!stopped()) {
4654 // Since superc is primitive, make a guard for the superc==subc case.
4655 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4656 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4657 generate_fair_guard(bol_eq, region);
4658 if (region->req() == PATH_LIMIT+1) {
4659 // A guard was added. If the added guard is taken, superc==subc.
4660 region->swap_edges(PATH_LIMIT, _prim_same_path);
4661 region->del_req(PATH_LIMIT);
4662 }
4663 region->set_req(_prim_0_path, control()); // Not equal after all.
4664 }
4665
4666 // these are the only paths that produce 'true':
4667 phi->set_req(_prim_same_path, intcon(1));
4668 phi->set_req(_ref_subtype_path, intcon(1));
4669
4670 // pull together the cases:
4671 assert(region->req() == PATH_LIMIT, "sane region");
4672 for (uint i = 1; i < region->req(); i++) {
4673 Node* ctl = region->in(i);
4674 if (ctl == nullptr || ctl == top()) {
4675 region->set_req(i, top());
4676 phi ->set_req(i, top());
4677 } else if (phi->in(i) == nullptr) {
4678 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4679 }
4680 }
4681
4682 set_control(_gvn.transform(region));
4683 set_result(_gvn.transform(phi));
4684 return true;
4685 }
4686
4687 //---------------------generate_array_guard_common------------------------
4688 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4689
4690 if (stopped()) {
4691 return nullptr;
4692 }
4693
4694 // Like generate_guard, adds a new path onto the region.
4695 jint layout_con = 0;
4696 Node* layout_val = get_layout_helper(kls, layout_con);
4697 if (layout_val == nullptr) {
4698 bool query = 0;
4699 switch(kind) {
4700 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4701 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4702 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4703 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4704 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4705 default:
4706 ShouldNotReachHere();
4707 }
4708 if (!query) {
4709 return nullptr; // never a branch
4710 } else { // always a branch
4711 Node* always_branch = control();
4712 if (region != nullptr)
4713 region->add_req(always_branch);
4714 set_control(top());
4715 return always_branch;
4716 }
4717 }
4718 unsigned int value = 0;
4719 BoolTest::mask btest = BoolTest::illegal;
4720 switch(kind) {
4721 case RefArray:
4722 case NonRefArray: {
4723 value = Klass::_lh_array_tag_ref_value;
4724 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4725 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4726 break;
4727 }
4728 case TypeArray: {
4729 value = Klass::_lh_array_tag_type_value;
4730 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4731 btest = BoolTest::eq;
4732 break;
4733 }
4734 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4735 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4736 default:
4737 ShouldNotReachHere();
4738 }
4739 // Now test the correct condition.
4740 jint nval = (jint)value;
4741 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4742 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4743 Node* ctrl = generate_fair_guard(bol, region);
4744 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4745 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4746 // Keep track of the fact that 'obj' is an array to prevent
4747 // array specific accesses from floating above the guard.
4748 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4749 }
4750 return ctrl;
4751 }
4752
4753 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4754 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4755 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4756 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4757 assert(null_free || atomic, "nullable implies atomic");
4758 Node* componentType = argument(0);
4759 Node* length = argument(1);
4760 Node* init_val = null_free ? argument(2) : nullptr;
4761
4762 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4763 if (tp != nullptr) {
4764 ciInstanceKlass* ik = tp->instance_klass();
4765 if (ik == C->env()->Class_klass()) {
4766 ciType* t = tp->java_mirror_type();
4767 if (t != nullptr && t->is_inlinetype()) {
4768
4769 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4770 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4771 assert(array_klass->is_elem_atomic() == atomic, "inconsistency");
4772
4773 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4774 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4775 return false;
4776 }
4777
4778 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4779 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces, true);
4780 if (null_free) {
4781 if (init_val->is_InlineType()) {
4782 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4783 // Zeroing is enough because the init value is the all-zero value
4784 init_val = nullptr;
4785 } else {
4786 init_val = init_val->as_InlineType()->buffer(this);
4787 }
4788 }
4789 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4790 }
4791 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4792 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4793 assert(arytype->is_null_free() == null_free, "inconsistency");
4794 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4795 assert(arytype->is_atomic() == atomic, "inconsistency");
4796 set_result(obj);
4797 return true;
4798 }
4799 }
4800 }
4801 }
4802 return false;
4803 }
4804
4805 Node* LibraryCallKit::load_default_array_klass(Node* klass_node) {
4806 // TODO 8366668
4807 // - Fred suggested that we could just have the first entry in the refined list point to the array with ArrayKlass::ArrayProperties::DEFAULT property
4808 // For now, we just load from ObjArrayKlass::_next_refined_array_klass, which would always be the refKlass for non-values, and deopt if it's not
4809 // - Convert this to an IGVN optimization, so it's also folded after parsing
4810 // - The generate_typeArray_guard is not needed by all callers, double-check that it's folded
4811
4812 const Type* klass_t = _gvn.type(klass_node);
4813 const TypeAryKlassPtr* ary_klass_t = klass_t->isa_aryklassptr();
4814 if (ary_klass_t && ary_klass_t->klass_is_exact()) {
4815 if (ary_klass_t->exact_klass()->is_obj_array_klass()) {
4816 ary_klass_t = ary_klass_t->get_vm_type(false);
4817 return makecon(ary_klass_t);
4818 } else {
4819 return klass_node;
4820 }
4821 }
4822
4823 // Load next refined array klass if klass is an ObjArrayKlass
4824 RegionNode* refined_region = new RegionNode(2);
4825 Node* refined_phi = new PhiNode(refined_region, klass_t);
4826
4827 generate_typeArray_guard(klass_node, refined_region);
4828 if (refined_region->req() == 3) {
4829 refined_phi->add_req(klass_node);
4830 }
4831
4832 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4833 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4834
4835 RegionNode* refined_region2 = new RegionNode(3);
4836 Node* refined_phi2 = new PhiNode(refined_region2, klass_t);
4837
4838 Node* null_ctl = top();
4839 Node* null_free_klass = null_check_common(refined_klass, T_OBJECT, false, &null_ctl);
4840 refined_region2->init_req(1, null_ctl);
4841 refined_phi2->init_req(1, klass_node);
4842
4843 refined_region2->init_req(2, control());
4844 refined_phi2->init_req(2, null_free_klass);
4845
4846 set_control(_gvn.transform(refined_region2));
4847 refined_klass = _gvn.transform(refined_phi2);
4848
4849 Node* adr_properties = basic_plus_adr(refined_klass, in_bytes(ObjArrayKlass::properties_offset()));
4850
4851 Node* properties = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_properties, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
4852 Node* default_val = makecon(TypeInt::make(ArrayKlass::ArrayProperties::DEFAULT));
4853 Node* chk = _gvn.transform(new CmpINode(properties, default_val));
4854 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
4855
4856 { // Deoptimize if not the default property
4857 BuildCutout unless(this, tst, PROB_MAX);
4858 uncommon_trap_exact(Deoptimization::Reason_class_check, Deoptimization::Action_none);
4859 }
4860
4861 refined_region->init_req(1, control());
4862 refined_phi->init_req(1, refined_klass);
4863
4864 set_control(_gvn.transform(refined_region));
4865 klass_node = _gvn.transform(refined_phi);
4866
4867 return klass_node;
4868 }
4869
4870 //-----------------------inline_native_newArray--------------------------
4871 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4872 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4873 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4874 Node* mirror;
4875 Node* count_val;
4876 if (uninitialized) {
4877 null_check_receiver();
4878 mirror = argument(1);
4879 count_val = argument(2);
4880 } else {
4881 mirror = argument(0);
4882 count_val = argument(1);
4883 }
4884
4885 mirror = null_check(mirror);
4886 // If mirror or obj is dead, only null-path is taken.
4887 if (stopped()) return true;
4888
4889 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4890 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4891 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4892 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4893 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4894
4895 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4896 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4897 result_reg, _slow_path);
4898 Node* normal_ctl = control();
4899 Node* no_array_ctl = result_reg->in(_slow_path);
4900
4901 // Generate code for the slow case. We make a call to newArray().
4902 set_control(no_array_ctl);
4903 if (!stopped()) {
4904 // Either the input type is void.class, or else the
4905 // array klass has not yet been cached. Either the
4906 // ensuing call will throw an exception, or else it
4907 // will cache the array klass for next time.
4908 PreserveJVMState pjvms(this);
4909 CallJavaNode* slow_call = nullptr;
4910 if (uninitialized) {
4911 // Generate optimized virtual call (holder class 'Unsafe' is final)
4912 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4913 } else {
4914 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4915 }
4916 Node* slow_result = set_results_for_java_call(slow_call);
4917 // this->control() comes from set_results_for_java_call
4918 result_reg->set_req(_slow_path, control());
4919 result_val->set_req(_slow_path, slow_result);
4920 result_io ->set_req(_slow_path, i_o());
4921 result_mem->set_req(_slow_path, reset_memory());
4922 }
4923
4924 set_control(normal_ctl);
4925 if (!stopped()) {
4926 // Normal case: The array type has been cached in the java.lang.Class.
4927 // The following call works fine even if the array type is polymorphic.
4928 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4929
4930 klass_node = load_default_array_klass(klass_node);
4931
4932 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4933 result_reg->init_req(_normal_path, control());
4934 result_val->init_req(_normal_path, obj);
4935 result_io ->init_req(_normal_path, i_o());
4936 result_mem->init_req(_normal_path, reset_memory());
4937
4938 if (uninitialized) {
4939 // Mark the allocation so that zeroing is skipped
4940 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4941 alloc->maybe_set_complete(&_gvn);
4942 }
4943 }
4944
4945 // Return the combined state.
4946 set_i_o( _gvn.transform(result_io) );
4947 set_all_memory( _gvn.transform(result_mem));
4948
4949 C->set_has_split_ifs(true); // Has chance for split-if optimization
4950 set_result(result_reg, result_val);
4951 return true;
4952 }
4953
4954 //----------------------inline_native_getLength--------------------------
4955 // public static native int java.lang.reflect.Array.getLength(Object array);
4956 bool LibraryCallKit::inline_native_getLength() {
4957 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4958
4959 Node* array = null_check(argument(0));
4960 // If array is dead, only null-path is taken.
4961 if (stopped()) return true;
4962
4963 // Deoptimize if it is a non-array.
4964 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4965
4966 if (non_array != nullptr) {
4967 PreserveJVMState pjvms(this);
4968 set_control(non_array);
4969 uncommon_trap(Deoptimization::Reason_intrinsic,
4970 Deoptimization::Action_maybe_recompile);
4971 }
4972
4973 // If control is dead, only non-array-path is taken.
4974 if (stopped()) return true;
4975
4976 // The works fine even if the array type is polymorphic.
4977 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4978 Node* result = load_array_length(array);
4979
4980 C->set_has_split_ifs(true); // Has chance for split-if optimization
4981 set_result(result);
4982 return true;
4983 }
4984
4985 //------------------------inline_array_copyOf----------------------------
4986 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4987 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4988 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4989 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4990
4991 // Get the arguments.
4992 Node* original = argument(0);
4993 Node* start = is_copyOfRange? argument(1): intcon(0);
4994 Node* end = is_copyOfRange? argument(2): argument(1);
4995 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4996
4997 Node* newcopy = nullptr;
4998
4999 // Set the original stack and the reexecute bit for the interpreter to reexecute
5000 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5001 { PreserveReexecuteState preexecs(this);
5002 jvms()->set_should_reexecute(true);
5003
5004 array_type_mirror = null_check(array_type_mirror);
5005 original = null_check(original);
5006
5007 // Check if a null path was taken unconditionally.
5008 if (stopped()) return true;
5009
5010 Node* orig_length = load_array_length(original);
5011
5012 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5013 klass_node = null_check(klass_node);
5014
5015 RegionNode* bailout = new RegionNode(1);
5016 record_for_igvn(bailout);
5017
5018 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5019 // Bail out if that is so.
5020 // Inline type array may have object field that would require a
5021 // write barrier. Conservatively, go to slow path.
5022 // TODO 8251971: Optimize for the case when flat src/dst are later found
5023 // to not contain oops (i.e., move this check to the macro expansion phase).
5024 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5025 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5026 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5027 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5028 // Can src array be flat and contain oops?
5029 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5030 // Can dest array be flat and contain oops?
5031 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5032 // TODO 8366668 generate_non_refArray_guard also passed for ref arrays??
5033 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5034
5035 klass_node = load_default_array_klass(klass_node);
5036
5037 if (not_objArray != nullptr) {
5038 // Improve the klass node's type from the new optimistic assumption:
5039 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5040 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
5041 Node* cast = new CastPPNode(control(), klass_node, akls);
5042 klass_node = _gvn.transform(cast);
5043 }
5044
5045 // Bail out if either start or end is negative.
5046 generate_negative_guard(start, bailout, &start);
5047 generate_negative_guard(end, bailout, &end);
5048
5049 Node* length = end;
5050 if (_gvn.type(start) != TypeInt::ZERO) {
5051 length = _gvn.transform(new SubINode(end, start));
5052 }
5053
5054 // Bail out if length is negative (i.e., if start > end).
5055 // Without this the new_array would throw
5056 // NegativeArraySizeException but IllegalArgumentException is what
5057 // should be thrown
5058 generate_negative_guard(length, bailout, &length);
5059
5060 // Handle inline type arrays
5061 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5062 if (!stopped()) {
5063 // TODO JDK-8329224
5064 if (!orig_t->is_null_free()) {
5065 // Not statically known to be null free, add a check
5066 generate_fair_guard(null_free_array_test(original), bailout);
5067 }
5068 orig_t = _gvn.type(original)->isa_aryptr();
5069 if (orig_t != nullptr && orig_t->is_flat()) {
5070 // Src is flat, check that dest is flat as well
5071 if (exclude_flat) {
5072 // Dest can't be flat, bail out
5073 bailout->add_req(control());
5074 set_control(top());
5075 } else {
5076 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
5077 }
5078 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5079 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5080 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5081 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5082 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5083 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5084 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5085 if (orig_t != nullptr) {
5086 orig_t = orig_t->cast_to_not_flat();
5087 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5088 }
5089 }
5090 if (!can_validate) {
5091 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5092 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5093 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5094 generate_fair_guard(flat_array_test(klass_node), bailout);
5095 generate_fair_guard(null_free_array_test(original), bailout);
5096 }
5097 }
5098
5099 // Bail out if start is larger than the original length
5100 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5101 generate_negative_guard(orig_tail, bailout, &orig_tail);
5102
5103 if (bailout->req() > 1) {
5104 PreserveJVMState pjvms(this);
5105 set_control(_gvn.transform(bailout));
5106 uncommon_trap(Deoptimization::Reason_intrinsic,
5107 Deoptimization::Action_maybe_recompile);
5108 }
5109
5110 if (!stopped()) {
5111 // How many elements will we copy from the original?
5112 // The answer is MinI(orig_tail, length).
5113 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5114
5115 // Generate a direct call to the right arraycopy function(s).
5116 // We know the copy is disjoint but we might not know if the
5117 // oop stores need checking.
5118 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5119 // This will fail a store-check if x contains any non-nulls.
5120
5121 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5122 // loads/stores but it is legal only if we're sure the
5123 // Arrays.copyOf would succeed. So we need all input arguments
5124 // to the copyOf to be validated, including that the copy to the
5125 // new array won't trigger an ArrayStoreException. That subtype
5126 // check can be optimized if we know something on the type of
5127 // the input array from type speculation.
5128 if (_gvn.type(klass_node)->singleton()) {
5129 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5130 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5131
5132 int test = C->static_subtype_check(superk, subk);
5133 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5134 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5135 if (t_original->speculative_type() != nullptr) {
5136 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5137 }
5138 }
5139 }
5140
5141 bool validated = false;
5142 // Reason_class_check rather than Reason_intrinsic because we
5143 // want to intrinsify even if this traps.
5144 if (can_validate) {
5145 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5146
5147 if (not_subtype_ctrl != top()) {
5148 PreserveJVMState pjvms(this);
5149 set_control(not_subtype_ctrl);
5150 uncommon_trap(Deoptimization::Reason_class_check,
5151 Deoptimization::Action_make_not_entrant);
5152 assert(stopped(), "Should be stopped");
5153 }
5154 validated = true;
5155 }
5156
5157 if (!stopped()) {
5158 newcopy = new_array(klass_node, length, 0); // no arguments to push
5159
5160 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5161 load_object_klass(original), klass_node);
5162 if (!is_copyOfRange) {
5163 ac->set_copyof(validated);
5164 } else {
5165 ac->set_copyofrange(validated);
5166 }
5167 Node* n = _gvn.transform(ac);
5168 if (n == ac) {
5169 ac->connect_outputs(this);
5170 } else {
5171 assert(validated, "shouldn't transform if all arguments not validated");
5172 set_all_memory(n);
5173 }
5174 }
5175 }
5176 } // original reexecute is set back here
5177
5178 C->set_has_split_ifs(true); // Has chance for split-if optimization
5179 if (!stopped()) {
5180 set_result(newcopy);
5181 }
5182 return true;
5183 }
5184
5185
5186 //----------------------generate_virtual_guard---------------------------
5187 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5188 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5189 RegionNode* slow_region) {
5190 ciMethod* method = callee();
5191 int vtable_index = method->vtable_index();
5192 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5193 "bad index %d", vtable_index);
5194 // Get the Method* out of the appropriate vtable entry.
5195 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5196 vtable_index*vtableEntry::size_in_bytes() +
5197 in_bytes(vtableEntry::method_offset());
5198 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5199 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5200
5201 // Compare the target method with the expected method (e.g., Object.hashCode).
5202 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5203
5204 Node* native_call = makecon(native_call_addr);
5205 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5206 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5207
5208 return generate_slow_guard(test_native, slow_region);
5209 }
5210
5211 //-----------------------generate_method_call----------------------------
5212 // Use generate_method_call to make a slow-call to the real
5213 // method if the fast path fails. An alternative would be to
5214 // use a stub like OptoRuntime::slow_arraycopy_Java.
5215 // This only works for expanding the current library call,
5216 // not another intrinsic. (E.g., don't use this for making an
5217 // arraycopy call inside of the copyOf intrinsic.)
5218 CallJavaNode*
5219 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5220 // When compiling the intrinsic method itself, do not use this technique.
5221 guarantee(callee() != C->method(), "cannot make slow-call to self");
5222
5223 ciMethod* method = callee();
5224 // ensure the JVMS we have will be correct for this call
5225 guarantee(method_id == method->intrinsic_id(), "must match");
5226
5227 const TypeFunc* tf = TypeFunc::make(method);
5228 if (res_not_null) {
5229 assert(tf->return_type() == T_OBJECT, "");
5230 const TypeTuple* range = tf->range_cc();
5231 const Type** fields = TypeTuple::fields(range->cnt());
5232 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5233 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5234 tf = TypeFunc::make(tf->domain_cc(), new_range);
5235 }
5236 CallJavaNode* slow_call;
5237 if (is_static) {
5238 assert(!is_virtual, "");
5239 slow_call = new CallStaticJavaNode(C, tf,
5240 SharedRuntime::get_resolve_static_call_stub(), method);
5241 } else if (is_virtual) {
5242 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5243 int vtable_index = Method::invalid_vtable_index;
5244 if (UseInlineCaches) {
5245 // Suppress the vtable call
5246 } else {
5247 // hashCode and clone are not a miranda methods,
5248 // so the vtable index is fixed.
5249 // No need to use the linkResolver to get it.
5250 vtable_index = method->vtable_index();
5251 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5252 "bad index %d", vtable_index);
5253 }
5254 slow_call = new CallDynamicJavaNode(tf,
5255 SharedRuntime::get_resolve_virtual_call_stub(),
5256 method, vtable_index);
5257 } else { // neither virtual nor static: opt_virtual
5258 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5259 slow_call = new CallStaticJavaNode(C, tf,
5260 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5261 slow_call->set_optimized_virtual(true);
5262 }
5263 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5264 // To be able to issue a direct call (optimized virtual or virtual)
5265 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5266 // about the method being invoked should be attached to the call site to
5267 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5268 slow_call->set_override_symbolic_info(true);
5269 }
5270 set_arguments_for_java_call(slow_call);
5271 set_edges_for_java_call(slow_call);
5272 return slow_call;
5273 }
5274
5275
5276 /**
5277 * Build special case code for calls to hashCode on an object. This call may
5278 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5279 * slightly different code.
5280 */
5281 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5282 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5283 assert(!(is_virtual && is_static), "either virtual, special, or static");
5284
5285 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5286
5287 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5288 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5289 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5290 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5291 Node* obj = argument(0);
5292
5293 // Don't intrinsify hashcode on inline types for now.
5294 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5295 if (gvn().type(obj)->is_inlinetypeptr()) {
5296 return false;
5297 }
5298
5299 if (!is_static) {
5300 // Check for hashing null object
5301 obj = null_check_receiver();
5302 if (stopped()) return true; // unconditionally null
5303 result_reg->init_req(_null_path, top());
5304 result_val->init_req(_null_path, top());
5305 } else {
5306 // Do a null check, and return zero if null.
5307 // System.identityHashCode(null) == 0
5308 Node* null_ctl = top();
5309 obj = null_check_oop(obj, &null_ctl);
5310 result_reg->init_req(_null_path, null_ctl);
5311 result_val->init_req(_null_path, _gvn.intcon(0));
5312 }
5313
5314 // Unconditionally null? Then return right away.
5315 if (stopped()) {
5316 set_control( result_reg->in(_null_path));
5317 if (!stopped())
5318 set_result(result_val->in(_null_path));
5319 return true;
5320 }
5321
5322 // We only go to the fast case code if we pass a number of guards. The
5323 // paths which do not pass are accumulated in the slow_region.
5324 RegionNode* slow_region = new RegionNode(1);
5325 record_for_igvn(slow_region);
5326
5327 // If this is a virtual call, we generate a funny guard. We pull out
5328 // the vtable entry corresponding to hashCode() from the target object.
5329 // If the target method which we are calling happens to be the native
5330 // Object hashCode() method, we pass the guard. We do not need this
5331 // guard for non-virtual calls -- the caller is known to be the native
5332 // Object hashCode().
5333 if (is_virtual) {
5334 // After null check, get the object's klass.
5335 Node* obj_klass = load_object_klass(obj);
5336 generate_virtual_guard(obj_klass, slow_region);
5337 }
5338
5339 // Get the header out of the object, use LoadMarkNode when available
5340 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5341 // The control of the load must be null. Otherwise, the load can move before
5342 // the null check after castPP removal.
5343 Node* no_ctrl = nullptr;
5344 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5345
5346 if (!UseObjectMonitorTable) {
5347 // Test the header to see if it is safe to read w.r.t. locking.
5348 // This also serves as guard against inline types
5349 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5350 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5351 if (LockingMode == LM_LIGHTWEIGHT) {
5352 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5353 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5354 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5355
5356 generate_slow_guard(test_monitor, slow_region);
5357 } else {
5358 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5359 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5360 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5361
5362 generate_slow_guard(test_not_unlocked, slow_region);
5363 }
5364 }
5365
5366 // Get the hash value and check to see that it has been properly assigned.
5367 // We depend on hash_mask being at most 32 bits and avoid the use of
5368 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5369 // vm: see markWord.hpp.
5370 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5371 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5372 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5373 // This hack lets the hash bits live anywhere in the mark object now, as long
5374 // as the shift drops the relevant bits into the low 32 bits. Note that
5375 // Java spec says that HashCode is an int so there's no point in capturing
5376 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5377 hshifted_header = ConvX2I(hshifted_header);
5378 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5379
5380 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5381 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5382 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5383
5384 generate_slow_guard(test_assigned, slow_region);
5385
5386 Node* init_mem = reset_memory();
5387 // fill in the rest of the null path:
5388 result_io ->init_req(_null_path, i_o());
5389 result_mem->init_req(_null_path, init_mem);
5390
5391 result_val->init_req(_fast_path, hash_val);
5392 result_reg->init_req(_fast_path, control());
5393 result_io ->init_req(_fast_path, i_o());
5394 result_mem->init_req(_fast_path, init_mem);
5395
5396 // Generate code for the slow case. We make a call to hashCode().
5397 set_control(_gvn.transform(slow_region));
5398 if (!stopped()) {
5399 // No need for PreserveJVMState, because we're using up the present state.
5400 set_all_memory(init_mem);
5401 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5402 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5403 Node* slow_result = set_results_for_java_call(slow_call);
5404 // this->control() comes from set_results_for_java_call
5405 result_reg->init_req(_slow_path, control());
5406 result_val->init_req(_slow_path, slow_result);
5407 result_io ->set_req(_slow_path, i_o());
5408 result_mem ->set_req(_slow_path, reset_memory());
5409 }
5410
5411 // Return the combined state.
5412 set_i_o( _gvn.transform(result_io) );
5413 set_all_memory( _gvn.transform(result_mem));
5414
5415 set_result(result_reg, result_val);
5416 return true;
5417 }
5418
5419 //---------------------------inline_native_getClass----------------------------
5420 // public final native Class<?> java.lang.Object.getClass();
5421 //
5422 // Build special case code for calls to getClass on an object.
5423 bool LibraryCallKit::inline_native_getClass() {
5424 Node* obj = argument(0);
5425 if (obj->is_InlineType()) {
5426 const Type* t = _gvn.type(obj);
5427 if (t->maybe_null()) {
5428 null_check(obj);
5429 }
5430 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5431 return true;
5432 }
5433 obj = null_check_receiver();
5434 if (stopped()) return true;
5435 set_result(load_mirror_from_klass(load_object_klass(obj)));
5436 return true;
5437 }
5438
5439 //-----------------inline_native_Reflection_getCallerClass---------------------
5440 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5441 //
5442 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5443 //
5444 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5445 // in that it must skip particular security frames and checks for
5446 // caller sensitive methods.
5447 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5448 #ifndef PRODUCT
5449 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5450 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5451 }
5452 #endif
5453
5454 if (!jvms()->has_method()) {
5455 #ifndef PRODUCT
5456 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5457 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5458 }
5459 #endif
5460 return false;
5461 }
5462
5463 // Walk back up the JVM state to find the caller at the required
5464 // depth.
5465 JVMState* caller_jvms = jvms();
5466
5467 // Cf. JVM_GetCallerClass
5468 // NOTE: Start the loop at depth 1 because the current JVM state does
5469 // not include the Reflection.getCallerClass() frame.
5470 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5471 ciMethod* m = caller_jvms->method();
5472 switch (n) {
5473 case 0:
5474 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5475 break;
5476 case 1:
5477 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5478 if (!m->caller_sensitive()) {
5479 #ifndef PRODUCT
5480 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5481 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5482 }
5483 #endif
5484 return false; // bail-out; let JVM_GetCallerClass do the work
5485 }
5486 break;
5487 default:
5488 if (!m->is_ignored_by_security_stack_walk()) {
5489 // We have reached the desired frame; return the holder class.
5490 // Acquire method holder as java.lang.Class and push as constant.
5491 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5492 ciInstance* caller_mirror = caller_klass->java_mirror();
5493 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5494
5495 #ifndef PRODUCT
5496 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5497 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());
5498 tty->print_cr(" JVM state at this point:");
5499 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5500 ciMethod* m = jvms()->of_depth(i)->method();
5501 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5502 }
5503 }
5504 #endif
5505 return true;
5506 }
5507 break;
5508 }
5509 }
5510
5511 #ifndef PRODUCT
5512 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5513 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5514 tty->print_cr(" JVM state at this point:");
5515 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5516 ciMethod* m = jvms()->of_depth(i)->method();
5517 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5518 }
5519 }
5520 #endif
5521
5522 return false; // bail-out; let JVM_GetCallerClass do the work
5523 }
5524
5525 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5526 Node* arg = argument(0);
5527 Node* result = nullptr;
5528
5529 switch (id) {
5530 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5531 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5532 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5533 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5534 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5535 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5536
5537 case vmIntrinsics::_doubleToLongBits: {
5538 // two paths (plus control) merge in a wood
5539 RegionNode *r = new RegionNode(3);
5540 Node *phi = new PhiNode(r, TypeLong::LONG);
5541
5542 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5543 // Build the boolean node
5544 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5545
5546 // Branch either way.
5547 // NaN case is less traveled, which makes all the difference.
5548 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5549 Node *opt_isnan = _gvn.transform(ifisnan);
5550 assert( opt_isnan->is_If(), "Expect an IfNode");
5551 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5552 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5553
5554 set_control(iftrue);
5555
5556 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5557 Node *slow_result = longcon(nan_bits); // return NaN
5558 phi->init_req(1, _gvn.transform( slow_result ));
5559 r->init_req(1, iftrue);
5560
5561 // Else fall through
5562 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5563 set_control(iffalse);
5564
5565 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5566 r->init_req(2, iffalse);
5567
5568 // Post merge
5569 set_control(_gvn.transform(r));
5570 record_for_igvn(r);
5571
5572 C->set_has_split_ifs(true); // Has chance for split-if optimization
5573 result = phi;
5574 assert(result->bottom_type()->isa_long(), "must be");
5575 break;
5576 }
5577
5578 case vmIntrinsics::_floatToIntBits: {
5579 // two paths (plus control) merge in a wood
5580 RegionNode *r = new RegionNode(3);
5581 Node *phi = new PhiNode(r, TypeInt::INT);
5582
5583 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5584 // Build the boolean node
5585 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5586
5587 // Branch either way.
5588 // NaN case is less traveled, which makes all the difference.
5589 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5590 Node *opt_isnan = _gvn.transform(ifisnan);
5591 assert( opt_isnan->is_If(), "Expect an IfNode");
5592 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5593 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5594
5595 set_control(iftrue);
5596
5597 static const jint nan_bits = 0x7fc00000;
5598 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5599 phi->init_req(1, _gvn.transform( slow_result ));
5600 r->init_req(1, iftrue);
5601
5602 // Else fall through
5603 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5604 set_control(iffalse);
5605
5606 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5607 r->init_req(2, iffalse);
5608
5609 // Post merge
5610 set_control(_gvn.transform(r));
5611 record_for_igvn(r);
5612
5613 C->set_has_split_ifs(true); // Has chance for split-if optimization
5614 result = phi;
5615 assert(result->bottom_type()->isa_int(), "must be");
5616 break;
5617 }
5618
5619 default:
5620 fatal_unexpected_iid(id);
5621 break;
5622 }
5623 set_result(_gvn.transform(result));
5624 return true;
5625 }
5626
5627 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5628 Node* arg = argument(0);
5629 Node* result = nullptr;
5630
5631 switch (id) {
5632 case vmIntrinsics::_floatIsInfinite:
5633 result = new IsInfiniteFNode(arg);
5634 break;
5635 case vmIntrinsics::_floatIsFinite:
5636 result = new IsFiniteFNode(arg);
5637 break;
5638 case vmIntrinsics::_doubleIsInfinite:
5639 result = new IsInfiniteDNode(arg);
5640 break;
5641 case vmIntrinsics::_doubleIsFinite:
5642 result = new IsFiniteDNode(arg);
5643 break;
5644 default:
5645 fatal_unexpected_iid(id);
5646 break;
5647 }
5648 set_result(_gvn.transform(result));
5649 return true;
5650 }
5651
5652 //----------------------inline_unsafe_copyMemory-------------------------
5653 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5654
5655 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5656 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5657 const Type* base_t = gvn.type(base);
5658
5659 bool in_native = (base_t == TypePtr::NULL_PTR);
5660 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5661 bool is_mixed = !in_heap && !in_native;
5662
5663 if (is_mixed) {
5664 return true; // mixed accesses can touch both on-heap and off-heap memory
5665 }
5666 if (in_heap) {
5667 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5668 if (!is_prim_array) {
5669 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5670 // there's not enough type information available to determine proper memory slice for it.
5671 return true;
5672 }
5673 }
5674 return false;
5675 }
5676
5677 bool LibraryCallKit::inline_unsafe_copyMemory() {
5678 if (callee()->is_static()) return false; // caller must have the capability!
5679 null_check_receiver(); // null-check receiver
5680 if (stopped()) return true;
5681
5682 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5683
5684 Node* src_base = argument(1); // type: oop
5685 Node* src_off = ConvL2X(argument(2)); // type: long
5686 Node* dst_base = argument(4); // type: oop
5687 Node* dst_off = ConvL2X(argument(5)); // type: long
5688 Node* size = ConvL2X(argument(7)); // type: long
5689
5690 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5691 "fieldOffset must be byte-scaled");
5692
5693 Node* src_addr = make_unsafe_address(src_base, src_off);
5694 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5695
5696 Node* thread = _gvn.transform(new ThreadLocalNode());
5697 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5698 BasicType doing_unsafe_access_bt = T_BYTE;
5699 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5700
5701 // update volatile field
5702 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5703
5704 int flags = RC_LEAF | RC_NO_FP;
5705
5706 const TypePtr* dst_type = TypePtr::BOTTOM;
5707
5708 // Adjust memory effects of the runtime call based on input values.
5709 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5710 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5711 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5712
5713 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5714 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5715 flags |= RC_NARROW_MEM; // narrow in memory
5716 }
5717 }
5718
5719 // Call it. Note that the length argument is not scaled.
5720 make_runtime_call(flags,
5721 OptoRuntime::fast_arraycopy_Type(),
5722 StubRoutines::unsafe_arraycopy(),
5723 "unsafe_arraycopy",
5724 dst_type,
5725 src_addr, dst_addr, size XTOP);
5726
5727 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5728
5729 return true;
5730 }
5731
5732 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5733 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5734 bool LibraryCallKit::inline_unsafe_setMemory() {
5735 if (callee()->is_static()) return false; // caller must have the capability!
5736 null_check_receiver(); // null-check receiver
5737 if (stopped()) return true;
5738
5739 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5740
5741 Node* dst_base = argument(1); // type: oop
5742 Node* dst_off = ConvL2X(argument(2)); // type: long
5743 Node* size = ConvL2X(argument(4)); // type: long
5744 Node* byte = argument(6); // type: byte
5745
5746 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5747 "fieldOffset must be byte-scaled");
5748
5749 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5750
5751 Node* thread = _gvn.transform(new ThreadLocalNode());
5752 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5753 BasicType doing_unsafe_access_bt = T_BYTE;
5754 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5755
5756 // update volatile field
5757 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5758
5759 int flags = RC_LEAF | RC_NO_FP;
5760
5761 const TypePtr* dst_type = TypePtr::BOTTOM;
5762
5763 // Adjust memory effects of the runtime call based on input values.
5764 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5765 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5766
5767 flags |= RC_NARROW_MEM; // narrow in memory
5768 }
5769
5770 // Call it. Note that the length argument is not scaled.
5771 make_runtime_call(flags,
5772 OptoRuntime::unsafe_setmemory_Type(),
5773 StubRoutines::unsafe_setmemory(),
5774 "unsafe_setmemory",
5775 dst_type,
5776 dst_addr, size XTOP, byte);
5777
5778 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5779
5780 return true;
5781 }
5782
5783 #undef XTOP
5784
5785 //------------------------clone_coping-----------------------------------
5786 // Helper function for inline_native_clone.
5787 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5788 assert(obj_size != nullptr, "");
5789 Node* raw_obj = alloc_obj->in(1);
5790 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5791
5792 AllocateNode* alloc = nullptr;
5793 if (ReduceBulkZeroing &&
5794 // If we are implementing an array clone without knowing its source type
5795 // (can happen when compiling the array-guarded branch of a reflective
5796 // Object.clone() invocation), initialize the array within the allocation.
5797 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5798 // to a runtime clone call that assumes fully initialized source arrays.
5799 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5800 // We will be completely responsible for initializing this object -
5801 // mark Initialize node as complete.
5802 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5803 // The object was just allocated - there should be no any stores!
5804 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5805 // Mark as complete_with_arraycopy so that on AllocateNode
5806 // expansion, we know this AllocateNode is initialized by an array
5807 // copy and a StoreStore barrier exists after the array copy.
5808 alloc->initialization()->set_complete_with_arraycopy();
5809 }
5810
5811 Node* size = _gvn.transform(obj_size);
5812 access_clone(obj, alloc_obj, size, is_array);
5813
5814 // Do not let reads from the cloned object float above the arraycopy.
5815 if (alloc != nullptr) {
5816 // Do not let stores that initialize this object be reordered with
5817 // a subsequent store that would make this object accessible by
5818 // other threads.
5819 // Record what AllocateNode this StoreStore protects so that
5820 // escape analysis can go from the MemBarStoreStoreNode to the
5821 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5822 // based on the escape status of the AllocateNode.
5823 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5824 } else {
5825 insert_mem_bar(Op_MemBarCPUOrder);
5826 }
5827 }
5828
5829 //------------------------inline_native_clone----------------------------
5830 // protected native Object java.lang.Object.clone();
5831 //
5832 // Here are the simple edge cases:
5833 // null receiver => normal trap
5834 // virtual and clone was overridden => slow path to out-of-line clone
5835 // not cloneable or finalizer => slow path to out-of-line Object.clone
5836 //
5837 // The general case has two steps, allocation and copying.
5838 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5839 //
5840 // Copying also has two cases, oop arrays and everything else.
5841 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5842 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5843 //
5844 // These steps fold up nicely if and when the cloned object's klass
5845 // can be sharply typed as an object array, a type array, or an instance.
5846 //
5847 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5848 PhiNode* result_val;
5849
5850 // Set the reexecute bit for the interpreter to reexecute
5851 // the bytecode that invokes Object.clone if deoptimization happens.
5852 { PreserveReexecuteState preexecs(this);
5853 jvms()->set_should_reexecute(true);
5854
5855 Node* obj = argument(0);
5856 obj = null_check_receiver();
5857 if (stopped()) return true;
5858
5859 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5860 if (obj_type->is_inlinetypeptr()) {
5861 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5862 // no identity.
5863 set_result(obj);
5864 return true;
5865 }
5866
5867 // If we are going to clone an instance, we need its exact type to
5868 // know the number and types of fields to convert the clone to
5869 // loads/stores. Maybe a speculative type can help us.
5870 if (!obj_type->klass_is_exact() &&
5871 obj_type->speculative_type() != nullptr &&
5872 obj_type->speculative_type()->is_instance_klass() &&
5873 !obj_type->speculative_type()->is_inlinetype()) {
5874 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5875 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5876 !spec_ik->has_injected_fields()) {
5877 if (!obj_type->isa_instptr() ||
5878 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5879 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5880 }
5881 }
5882 }
5883
5884 // Conservatively insert a memory barrier on all memory slices.
5885 // Do not let writes into the original float below the clone.
5886 insert_mem_bar(Op_MemBarCPUOrder);
5887
5888 // paths into result_reg:
5889 enum {
5890 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5891 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5892 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5893 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5894 PATH_LIMIT
5895 };
5896 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5897 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5898 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5899 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5900 record_for_igvn(result_reg);
5901
5902 Node* obj_klass = load_object_klass(obj);
5903 // We only go to the fast case code if we pass a number of guards.
5904 // The paths which do not pass are accumulated in the slow_region.
5905 RegionNode* slow_region = new RegionNode(1);
5906 record_for_igvn(slow_region);
5907
5908 Node* array_obj = obj;
5909 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5910 if (array_ctl != nullptr) {
5911 // It's an array.
5912 PreserveJVMState pjvms(this);
5913 set_control(array_ctl);
5914
5915 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5916 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5917 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5918 obj_type->can_be_inline_array() &&
5919 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5920 // Flat inline type array may have object field that would require a
5921 // write barrier. Conservatively, go to slow path.
5922 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5923 }
5924
5925 if (!stopped()) {
5926 Node* obj_length = load_array_length(array_obj);
5927 Node* array_size = nullptr; // Size of the array without object alignment padding.
5928 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5929
5930 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5931 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5932 // If it is an oop array, it requires very special treatment,
5933 // because gc barriers are required when accessing the array.
5934 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
5935 if (is_obja != nullptr) {
5936 PreserveJVMState pjvms2(this);
5937 set_control(is_obja);
5938 // Generate a direct call to the right arraycopy function(s).
5939 // Clones are always tightly coupled.
5940 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5941 ac->set_clone_oop_array();
5942 Node* n = _gvn.transform(ac);
5943 assert(n == ac, "cannot disappear");
5944 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5945
5946 result_reg->init_req(_objArray_path, control());
5947 result_val->init_req(_objArray_path, alloc_obj);
5948 result_i_o ->set_req(_objArray_path, i_o());
5949 result_mem ->set_req(_objArray_path, reset_memory());
5950 }
5951 }
5952 // Otherwise, there are no barriers to worry about.
5953 // (We can dispense with card marks if we know the allocation
5954 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5955 // causes the non-eden paths to take compensating steps to
5956 // simulate a fresh allocation, so that no further
5957 // card marks are required in compiled code to initialize
5958 // the object.)
5959
5960 if (!stopped()) {
5961 copy_to_clone(obj, alloc_obj, array_size, true);
5962
5963 // Present the results of the copy.
5964 result_reg->init_req(_array_path, control());
5965 result_val->init_req(_array_path, alloc_obj);
5966 result_i_o ->set_req(_array_path, i_o());
5967 result_mem ->set_req(_array_path, reset_memory());
5968 }
5969 }
5970 }
5971
5972 if (!stopped()) {
5973 // It's an instance (we did array above). Make the slow-path tests.
5974 // If this is a virtual call, we generate a funny guard. We grab
5975 // the vtable entry corresponding to clone() from the target object.
5976 // If the target method which we are calling happens to be the
5977 // Object clone() method, we pass the guard. We do not need this
5978 // guard for non-virtual calls; the caller is known to be the native
5979 // Object clone().
5980 if (is_virtual) {
5981 generate_virtual_guard(obj_klass, slow_region);
5982 }
5983
5984 // The object must be easily cloneable and must not have a finalizer.
5985 // Both of these conditions may be checked in a single test.
5986 // We could optimize the test further, but we don't care.
5987 generate_misc_flags_guard(obj_klass,
5988 // Test both conditions:
5989 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5990 // Must be cloneable but not finalizer:
5991 KlassFlags::_misc_is_cloneable_fast,
5992 slow_region);
5993 }
5994
5995 if (!stopped()) {
5996 // It's an instance, and it passed the slow-path tests.
5997 PreserveJVMState pjvms(this);
5998 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5999 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6000 // is reexecuted if deoptimization occurs and there could be problems when merging
6001 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6002 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6003
6004 copy_to_clone(obj, alloc_obj, obj_size, false);
6005
6006 // Present the results of the slow call.
6007 result_reg->init_req(_instance_path, control());
6008 result_val->init_req(_instance_path, alloc_obj);
6009 result_i_o ->set_req(_instance_path, i_o());
6010 result_mem ->set_req(_instance_path, reset_memory());
6011 }
6012
6013 // Generate code for the slow case. We make a call to clone().
6014 set_control(_gvn.transform(slow_region));
6015 if (!stopped()) {
6016 PreserveJVMState pjvms(this);
6017 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6018 // We need to deoptimize on exception (see comment above)
6019 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6020 // this->control() comes from set_results_for_java_call
6021 result_reg->init_req(_slow_path, control());
6022 result_val->init_req(_slow_path, slow_result);
6023 result_i_o ->set_req(_slow_path, i_o());
6024 result_mem ->set_req(_slow_path, reset_memory());
6025 }
6026
6027 // Return the combined state.
6028 set_control( _gvn.transform(result_reg));
6029 set_i_o( _gvn.transform(result_i_o));
6030 set_all_memory( _gvn.transform(result_mem));
6031 } // original reexecute is set back here
6032
6033 set_result(_gvn.transform(result_val));
6034 return true;
6035 }
6036
6037 // If we have a tightly coupled allocation, the arraycopy may take care
6038 // of the array initialization. If one of the guards we insert between
6039 // the allocation and the arraycopy causes a deoptimization, an
6040 // uninitialized array will escape the compiled method. To prevent that
6041 // we set the JVM state for uncommon traps between the allocation and
6042 // the arraycopy to the state before the allocation so, in case of
6043 // deoptimization, we'll reexecute the allocation and the
6044 // initialization.
6045 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6046 if (alloc != nullptr) {
6047 ciMethod* trap_method = alloc->jvms()->method();
6048 int trap_bci = alloc->jvms()->bci();
6049
6050 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6051 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6052 // Make sure there's no store between the allocation and the
6053 // arraycopy otherwise visible side effects could be rexecuted
6054 // in case of deoptimization and cause incorrect execution.
6055 bool no_interfering_store = true;
6056 Node* mem = alloc->in(TypeFunc::Memory);
6057 if (mem->is_MergeMem()) {
6058 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6059 Node* n = mms.memory();
6060 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6061 assert(n->is_Store(), "what else?");
6062 no_interfering_store = false;
6063 break;
6064 }
6065 }
6066 } else {
6067 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6068 Node* n = mms.memory();
6069 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6070 assert(n->is_Store(), "what else?");
6071 no_interfering_store = false;
6072 break;
6073 }
6074 }
6075 }
6076
6077 if (no_interfering_store) {
6078 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6079
6080 JVMState* saved_jvms = jvms();
6081 saved_reexecute_sp = _reexecute_sp;
6082
6083 set_jvms(sfpt->jvms());
6084 _reexecute_sp = jvms()->sp();
6085
6086 return saved_jvms;
6087 }
6088 }
6089 }
6090 return nullptr;
6091 }
6092
6093 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6094 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6095 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6096 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6097 uint size = alloc->req();
6098 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6099 old_jvms->set_map(sfpt);
6100 for (uint i = 0; i < size; i++) {
6101 sfpt->init_req(i, alloc->in(i));
6102 }
6103 int adjustment = 1;
6104 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6105 if (ary_klass_ptr->is_null_free()) {
6106 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6107 // also requires the componentType and initVal on stack for re-execution.
6108 // Re-create and push the componentType.
6109 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6110 ciInstance* instance = klass->component_mirror_instance();
6111 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6112 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6113 adjustment++;
6114 }
6115 // re-push array length for deoptimization
6116 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6117 if (ary_klass_ptr->is_null_free()) {
6118 // Re-create and push the initVal.
6119 Node* init_val = alloc->in(AllocateNode::InitValue);
6120 if (init_val == nullptr) {
6121 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6122 } else if (UseCompressedOops) {
6123 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6124 }
6125 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6126 adjustment++;
6127 }
6128 old_jvms->set_sp(old_jvms->sp() + adjustment);
6129 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6130 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6131 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6132 old_jvms->set_should_reexecute(true);
6133
6134 sfpt->set_i_o(map()->i_o());
6135 sfpt->set_memory(map()->memory());
6136 sfpt->set_control(map()->control());
6137 return sfpt;
6138 }
6139
6140 // In case of a deoptimization, we restart execution at the
6141 // allocation, allocating a new array. We would leave an uninitialized
6142 // array in the heap that GCs wouldn't expect. Move the allocation
6143 // after the traps so we don't allocate the array if we
6144 // deoptimize. This is possible because tightly_coupled_allocation()
6145 // guarantees there's no observer of the allocated array at this point
6146 // and the control flow is simple enough.
6147 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6148 int saved_reexecute_sp, uint new_idx) {
6149 if (saved_jvms_before_guards != nullptr && !stopped()) {
6150 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6151
6152 assert(alloc != nullptr, "only with a tightly coupled allocation");
6153 // restore JVM state to the state at the arraycopy
6154 saved_jvms_before_guards->map()->set_control(map()->control());
6155 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6156 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6157 // If we've improved the types of some nodes (null check) while
6158 // emitting the guards, propagate them to the current state
6159 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6160 set_jvms(saved_jvms_before_guards);
6161 _reexecute_sp = saved_reexecute_sp;
6162
6163 // Remove the allocation from above the guards
6164 CallProjections* callprojs = alloc->extract_projections(true);
6165 InitializeNode* init = alloc->initialization();
6166 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6167 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6168 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
6169
6170 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6171 // the allocation (i.e. is only valid if the allocation succeeds):
6172 // 1) replace CastIINode with AllocateArrayNode's length here
6173 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6174 //
6175 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6176 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6177 Node* init_control = init->proj_out(TypeFunc::Control);
6178 Node* alloc_length = alloc->Ideal_length();
6179 #ifdef ASSERT
6180 Node* prev_cast = nullptr;
6181 #endif
6182 for (uint i = 0; i < init_control->outcnt(); i++) {
6183 Node* init_out = init_control->raw_out(i);
6184 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6185 #ifdef ASSERT
6186 if (prev_cast == nullptr) {
6187 prev_cast = init_out;
6188 } else {
6189 if (prev_cast->cmp(*init_out) == false) {
6190 prev_cast->dump();
6191 init_out->dump();
6192 assert(false, "not equal CastIINode");
6193 }
6194 }
6195 #endif
6196 C->gvn_replace_by(init_out, alloc_length);
6197 }
6198 }
6199 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6200
6201 // move the allocation here (after the guards)
6202 _gvn.hash_delete(alloc);
6203 alloc->set_req(TypeFunc::Control, control());
6204 alloc->set_req(TypeFunc::I_O, i_o());
6205 Node *mem = reset_memory();
6206 set_all_memory(mem);
6207 alloc->set_req(TypeFunc::Memory, mem);
6208 set_control(init->proj_out_or_null(TypeFunc::Control));
6209 set_i_o(callprojs->fallthrough_ioproj);
6210
6211 // Update memory as done in GraphKit::set_output_for_allocation()
6212 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6213 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6214 if (ary_type->isa_aryptr() && length_type != nullptr) {
6215 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6216 }
6217 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6218 int elemidx = C->get_alias_index(telemref);
6219 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
6220 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
6221
6222 Node* allocx = _gvn.transform(alloc);
6223 assert(allocx == alloc, "where has the allocation gone?");
6224 assert(dest->is_CheckCastPP(), "not an allocation result?");
6225
6226 _gvn.hash_delete(dest);
6227 dest->set_req(0, control());
6228 Node* destx = _gvn.transform(dest);
6229 assert(destx == dest, "where has the allocation result gone?");
6230
6231 array_ideal_length(alloc, ary_type, true);
6232 }
6233 }
6234
6235 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6236 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6237 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6238 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6239 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6240 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6241 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6242 JVMState* saved_jvms_before_guards) {
6243 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6244 // There is at least one unrelated uncommon trap which needs to be replaced.
6245 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6246
6247 JVMState* saved_jvms = jvms();
6248 const int saved_reexecute_sp = _reexecute_sp;
6249 set_jvms(sfpt->jvms());
6250 _reexecute_sp = jvms()->sp();
6251
6252 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6253
6254 // Restore state
6255 set_jvms(saved_jvms);
6256 _reexecute_sp = saved_reexecute_sp;
6257 }
6258 }
6259
6260 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6261 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6262 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6263 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6264 while (if_proj->is_IfProj()) {
6265 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6266 if (uncommon_trap != nullptr) {
6267 create_new_uncommon_trap(uncommon_trap);
6268 }
6269 assert(if_proj->in(0)->is_If(), "must be If");
6270 if_proj = if_proj->in(0)->in(0);
6271 }
6272 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6273 "must have reached control projection of init node");
6274 }
6275
6276 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6277 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6278 assert(trap_request != 0, "no valid UCT trap request");
6279 PreserveJVMState pjvms(this);
6280 set_control(uncommon_trap_call->in(0));
6281 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6282 Deoptimization::trap_request_action(trap_request));
6283 assert(stopped(), "Should be stopped");
6284 _gvn.hash_delete(uncommon_trap_call);
6285 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6286 }
6287
6288 // Common checks for array sorting intrinsics arguments.
6289 // Returns `true` if checks passed.
6290 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6291 // check address of the class
6292 if (elementType == nullptr || elementType->is_top()) {
6293 return false; // dead path
6294 }
6295 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6296 if (elem_klass == nullptr) {
6297 return false; // dead path
6298 }
6299 // java_mirror_type() returns non-null for compile-time Class constants only
6300 ciType* elem_type = elem_klass->java_mirror_type();
6301 if (elem_type == nullptr) {
6302 return false;
6303 }
6304 bt = elem_type->basic_type();
6305 // Disable the intrinsic if the CPU does not support SIMD sort
6306 if (!Matcher::supports_simd_sort(bt)) {
6307 return false;
6308 }
6309 // check address of the array
6310 if (obj == nullptr || obj->is_top()) {
6311 return false; // dead path
6312 }
6313 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6314 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6315 return false; // failed input validation
6316 }
6317 return true;
6318 }
6319
6320 //------------------------------inline_array_partition-----------------------
6321 bool LibraryCallKit::inline_array_partition() {
6322 address stubAddr = StubRoutines::select_array_partition_function();
6323 if (stubAddr == nullptr) {
6324 return false; // Intrinsic's stub is not implemented on this platform
6325 }
6326 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6327
6328 // no receiver because it is a static method
6329 Node* elementType = argument(0);
6330 Node* obj = argument(1);
6331 Node* offset = argument(2); // long
6332 Node* fromIndex = argument(4);
6333 Node* toIndex = argument(5);
6334 Node* indexPivot1 = argument(6);
6335 Node* indexPivot2 = argument(7);
6336 // PartitionOperation: argument(8) is ignored
6337
6338 Node* pivotIndices = nullptr;
6339 BasicType bt = T_ILLEGAL;
6340
6341 if (!check_array_sort_arguments(elementType, obj, bt)) {
6342 return false;
6343 }
6344 null_check(obj);
6345 // If obj is dead, only null-path is taken.
6346 if (stopped()) {
6347 return true;
6348 }
6349 // Set the original stack and the reexecute bit for the interpreter to reexecute
6350 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6351 { PreserveReexecuteState preexecs(this);
6352 jvms()->set_should_reexecute(true);
6353
6354 Node* obj_adr = make_unsafe_address(obj, offset);
6355
6356 // create the pivotIndices array of type int and size = 2
6357 Node* size = intcon(2);
6358 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6359 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6360 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6361 guarantee(alloc != nullptr, "created above");
6362 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6363
6364 // pass the basic type enum to the stub
6365 Node* elemType = intcon(bt);
6366
6367 // Call the stub
6368 const char *stubName = "array_partition_stub";
6369 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6370 stubAddr, stubName, TypePtr::BOTTOM,
6371 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6372 indexPivot1, indexPivot2);
6373
6374 } // original reexecute is set back here
6375
6376 if (!stopped()) {
6377 set_result(pivotIndices);
6378 }
6379
6380 return true;
6381 }
6382
6383
6384 //------------------------------inline_array_sort-----------------------
6385 bool LibraryCallKit::inline_array_sort() {
6386 address stubAddr = StubRoutines::select_arraysort_function();
6387 if (stubAddr == nullptr) {
6388 return false; // Intrinsic's stub is not implemented on this platform
6389 }
6390 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6391
6392 // no receiver because it is a static method
6393 Node* elementType = argument(0);
6394 Node* obj = argument(1);
6395 Node* offset = argument(2); // long
6396 Node* fromIndex = argument(4);
6397 Node* toIndex = argument(5);
6398 // SortOperation: argument(6) is ignored
6399
6400 BasicType bt = T_ILLEGAL;
6401
6402 if (!check_array_sort_arguments(elementType, obj, bt)) {
6403 return false;
6404 }
6405 null_check(obj);
6406 // If obj is dead, only null-path is taken.
6407 if (stopped()) {
6408 return true;
6409 }
6410 Node* obj_adr = make_unsafe_address(obj, offset);
6411
6412 // pass the basic type enum to the stub
6413 Node* elemType = intcon(bt);
6414
6415 // Call the stub.
6416 const char *stubName = "arraysort_stub";
6417 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6418 stubAddr, stubName, TypePtr::BOTTOM,
6419 obj_adr, elemType, fromIndex, toIndex);
6420
6421 return true;
6422 }
6423
6424
6425 //------------------------------inline_arraycopy-----------------------
6426 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6427 // Object dest, int destPos,
6428 // int length);
6429 bool LibraryCallKit::inline_arraycopy() {
6430 // Get the arguments.
6431 Node* src = argument(0); // type: oop
6432 Node* src_offset = argument(1); // type: int
6433 Node* dest = argument(2); // type: oop
6434 Node* dest_offset = argument(3); // type: int
6435 Node* length = argument(4); // type: int
6436
6437 uint new_idx = C->unique();
6438
6439 // Check for allocation before we add nodes that would confuse
6440 // tightly_coupled_allocation()
6441 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6442
6443 int saved_reexecute_sp = -1;
6444 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6445 // See arraycopy_restore_alloc_state() comment
6446 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6447 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6448 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6449 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6450
6451 // The following tests must be performed
6452 // (1) src and dest are arrays.
6453 // (2) src and dest arrays must have elements of the same BasicType
6454 // (3) src and dest must not be null.
6455 // (4) src_offset must not be negative.
6456 // (5) dest_offset must not be negative.
6457 // (6) length must not be negative.
6458 // (7) src_offset + length must not exceed length of src.
6459 // (8) dest_offset + length must not exceed length of dest.
6460 // (9) each element of an oop array must be assignable
6461
6462 // (3) src and dest must not be null.
6463 // always do this here because we need the JVM state for uncommon traps
6464 Node* null_ctl = top();
6465 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6466 assert(null_ctl->is_top(), "no null control here");
6467 dest = null_check(dest, T_ARRAY);
6468
6469 if (!can_emit_guards) {
6470 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6471 // guards but the arraycopy node could still take advantage of a
6472 // tightly allocated allocation. tightly_coupled_allocation() is
6473 // called again to make sure it takes the null check above into
6474 // account: the null check is mandatory and if it caused an
6475 // uncommon trap to be emitted then the allocation can't be
6476 // considered tightly coupled in this context.
6477 alloc = tightly_coupled_allocation(dest);
6478 }
6479
6480 bool validated = false;
6481
6482 const Type* src_type = _gvn.type(src);
6483 const Type* dest_type = _gvn.type(dest);
6484 const TypeAryPtr* top_src = src_type->isa_aryptr();
6485 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6486
6487 // Do we have the type of src?
6488 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6489 // Do we have the type of dest?
6490 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6491 // Is the type for src from speculation?
6492 bool src_spec = false;
6493 // Is the type for dest from speculation?
6494 bool dest_spec = false;
6495
6496 if ((!has_src || !has_dest) && can_emit_guards) {
6497 // We don't have sufficient type information, let's see if
6498 // speculative types can help. We need to have types for both src
6499 // and dest so that it pays off.
6500
6501 // Do we already have or could we have type information for src
6502 bool could_have_src = has_src;
6503 // Do we already have or could we have type information for dest
6504 bool could_have_dest = has_dest;
6505
6506 ciKlass* src_k = nullptr;
6507 if (!has_src) {
6508 src_k = src_type->speculative_type_not_null();
6509 if (src_k != nullptr && src_k->is_array_klass()) {
6510 could_have_src = true;
6511 }
6512 }
6513
6514 ciKlass* dest_k = nullptr;
6515 if (!has_dest) {
6516 dest_k = dest_type->speculative_type_not_null();
6517 if (dest_k != nullptr && dest_k->is_array_klass()) {
6518 could_have_dest = true;
6519 }
6520 }
6521
6522 if (could_have_src && could_have_dest) {
6523 // This is going to pay off so emit the required guards
6524 if (!has_src) {
6525 src = maybe_cast_profiled_obj(src, src_k, true);
6526 src_type = _gvn.type(src);
6527 top_src = src_type->isa_aryptr();
6528 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6529 src_spec = true;
6530 }
6531 if (!has_dest) {
6532 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6533 dest_type = _gvn.type(dest);
6534 top_dest = dest_type->isa_aryptr();
6535 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6536 dest_spec = true;
6537 }
6538 }
6539 }
6540
6541 if (has_src && has_dest && can_emit_guards) {
6542 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6543 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6544 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6545 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6546
6547 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6548 // If both arrays are object arrays then having the exact types
6549 // for both will remove the need for a subtype check at runtime
6550 // before the call and may make it possible to pick a faster copy
6551 // routine (without a subtype check on every element)
6552 // Do we have the exact type of src?
6553 bool could_have_src = src_spec;
6554 // Do we have the exact type of dest?
6555 bool could_have_dest = dest_spec;
6556 ciKlass* src_k = nullptr;
6557 ciKlass* dest_k = nullptr;
6558 if (!src_spec) {
6559 src_k = src_type->speculative_type_not_null();
6560 if (src_k != nullptr && src_k->is_array_klass()) {
6561 could_have_src = true;
6562 }
6563 }
6564 if (!dest_spec) {
6565 dest_k = dest_type->speculative_type_not_null();
6566 if (dest_k != nullptr && dest_k->is_array_klass()) {
6567 could_have_dest = true;
6568 }
6569 }
6570 if (could_have_src && could_have_dest) {
6571 // If we can have both exact types, emit the missing guards
6572 if (could_have_src && !src_spec) {
6573 src = maybe_cast_profiled_obj(src, src_k, true);
6574 src_type = _gvn.type(src);
6575 top_src = src_type->isa_aryptr();
6576 }
6577 if (could_have_dest && !dest_spec) {
6578 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6579 dest_type = _gvn.type(dest);
6580 top_dest = dest_type->isa_aryptr();
6581 }
6582 }
6583 }
6584 }
6585
6586 ciMethod* trap_method = method();
6587 int trap_bci = bci();
6588 if (saved_jvms_before_guards != nullptr) {
6589 trap_method = alloc->jvms()->method();
6590 trap_bci = alloc->jvms()->bci();
6591 }
6592
6593 bool negative_length_guard_generated = false;
6594
6595 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6596 can_emit_guards && !src->is_top() && !dest->is_top()) {
6597 // validate arguments: enables transformation the ArrayCopyNode
6598 validated = true;
6599
6600 RegionNode* slow_region = new RegionNode(1);
6601 record_for_igvn(slow_region);
6602
6603 // (1) src and dest are arrays.
6604 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6605 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6606
6607 // (2) src and dest arrays must have elements of the same BasicType
6608 // done at macro expansion or at Ideal transformation time
6609
6610 // (4) src_offset must not be negative.
6611 generate_negative_guard(src_offset, slow_region);
6612
6613 // (5) dest_offset must not be negative.
6614 generate_negative_guard(dest_offset, slow_region);
6615
6616 // (7) src_offset + length must not exceed length of src.
6617 generate_limit_guard(src_offset, length,
6618 load_array_length(src),
6619 slow_region);
6620
6621 // (8) dest_offset + length must not exceed length of dest.
6622 generate_limit_guard(dest_offset, length,
6623 load_array_length(dest),
6624 slow_region);
6625
6626 // (6) length must not be negative.
6627 // This is also checked in generate_arraycopy() during macro expansion, but
6628 // we also have to check it here for the case where the ArrayCopyNode will
6629 // be eliminated by Escape Analysis.
6630 if (EliminateAllocations) {
6631 generate_negative_guard(length, slow_region);
6632 negative_length_guard_generated = true;
6633 }
6634
6635 // (9) each element of an oop array must be assignable
6636 Node* dest_klass = load_object_klass(dest);
6637 if (src != dest) {
6638 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6639 slow_region->add_req(not_subtype_ctrl);
6640 }
6641
6642 // TODO 8350865 Fix below logic. Also handle atomicity.
6643 generate_fair_guard(flat_array_test(src), slow_region);
6644 generate_fair_guard(flat_array_test(dest), slow_region);
6645
6646 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6647 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6648 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6649 src_type = _gvn.type(src);
6650 top_src = src_type->isa_aryptr();
6651
6652 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6653 if (!stopped() && UseArrayFlattening) {
6654 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6655 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6656 if (top_src != nullptr && top_src->is_flat()) {
6657 // Src is flat, check that dest is flat as well
6658 if (top_dest != nullptr && !top_dest->is_flat()) {
6659 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6660 // Since dest is flat and src <: dest, dest must have the same type as src.
6661 top_dest = top_src->cast_to_exactness(false);
6662 assert(top_dest->is_flat(), "dest must be flat");
6663 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6664 }
6665 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6666 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6667 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6668 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6669 generate_fair_guard(flat_array_test(src), slow_region);
6670 if (top_src != nullptr) {
6671 top_src = top_src->cast_to_not_flat();
6672 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6673 }
6674 }
6675 }
6676
6677 {
6678 PreserveJVMState pjvms(this);
6679 set_control(_gvn.transform(slow_region));
6680 uncommon_trap(Deoptimization::Reason_intrinsic,
6681 Deoptimization::Action_make_not_entrant);
6682 assert(stopped(), "Should be stopped");
6683 }
6684 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6685 }
6686
6687 if (stopped()) {
6688 return true;
6689 }
6690
6691 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6692 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6693 // so the compiler has a chance to eliminate them: during macro expansion,
6694 // we have to set their control (CastPP nodes are eliminated).
6695 load_object_klass(src), load_object_klass(dest),
6696 load_array_length(src), load_array_length(dest));
6697
6698 ac->set_arraycopy(validated);
6699
6700 Node* n = _gvn.transform(ac);
6701 if (n == ac) {
6702 ac->connect_outputs(this);
6703 } else {
6704 assert(validated, "shouldn't transform if all arguments not validated");
6705 set_all_memory(n);
6706 }
6707 clear_upper_avx();
6708
6709
6710 return true;
6711 }
6712
6713
6714 // Helper function which determines if an arraycopy immediately follows
6715 // an allocation, with no intervening tests or other escapes for the object.
6716 AllocateArrayNode*
6717 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6718 if (stopped()) return nullptr; // no fast path
6719 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6720
6721 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6722 if (alloc == nullptr) return nullptr;
6723
6724 Node* rawmem = memory(Compile::AliasIdxRaw);
6725 // Is the allocation's memory state untouched?
6726 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6727 // Bail out if there have been raw-memory effects since the allocation.
6728 // (Example: There might have been a call or safepoint.)
6729 return nullptr;
6730 }
6731 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6732 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6733 return nullptr;
6734 }
6735
6736 // There must be no unexpected observers of this allocation.
6737 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6738 Node* obs = ptr->fast_out(i);
6739 if (obs != this->map()) {
6740 return nullptr;
6741 }
6742 }
6743
6744 // This arraycopy must unconditionally follow the allocation of the ptr.
6745 Node* alloc_ctl = ptr->in(0);
6746 Node* ctl = control();
6747 while (ctl != alloc_ctl) {
6748 // There may be guards which feed into the slow_region.
6749 // Any other control flow means that we might not get a chance
6750 // to finish initializing the allocated object.
6751 // Various low-level checks bottom out in uncommon traps. These
6752 // are considered safe since we've already checked above that
6753 // there is no unexpected observer of this allocation.
6754 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6755 assert(ctl->in(0)->is_If(), "must be If");
6756 ctl = ctl->in(0)->in(0);
6757 } else {
6758 return nullptr;
6759 }
6760 }
6761
6762 // If we get this far, we have an allocation which immediately
6763 // precedes the arraycopy, and we can take over zeroing the new object.
6764 // The arraycopy will finish the initialization, and provide
6765 // a new control state to which we will anchor the destination pointer.
6766
6767 return alloc;
6768 }
6769
6770 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6771 if (node->is_IfProj()) {
6772 Node* other_proj = node->as_IfProj()->other_if_proj();
6773 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6774 Node* obs = other_proj->fast_out(j);
6775 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6776 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6777 return obs->as_CallStaticJava();
6778 }
6779 }
6780 }
6781 return nullptr;
6782 }
6783
6784 //-------------inline_encodeISOArray-----------------------------------
6785 // encode char[] to byte[] in ISO_8859_1 or ASCII
6786 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6787 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6788 // no receiver since it is static method
6789 Node *src = argument(0);
6790 Node *src_offset = argument(1);
6791 Node *dst = argument(2);
6792 Node *dst_offset = argument(3);
6793 Node *length = argument(4);
6794
6795 src = must_be_not_null(src, true);
6796 dst = must_be_not_null(dst, true);
6797
6798 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6799 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6800 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6801 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6802 // failed array check
6803 return false;
6804 }
6805
6806 // Figure out the size and type of the elements we will be copying.
6807 BasicType src_elem = src_type->elem()->array_element_basic_type();
6808 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6809 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6810 return false;
6811 }
6812
6813 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6814 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6815 // 'src_start' points to src array + scaled offset
6816 // 'dst_start' points to dst array + scaled offset
6817
6818 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6819 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6820 enc = _gvn.transform(enc);
6821 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6822 set_memory(res_mem, mtype);
6823 set_result(enc);
6824 clear_upper_avx();
6825
6826 return true;
6827 }
6828
6829 //-------------inline_multiplyToLen-----------------------------------
6830 bool LibraryCallKit::inline_multiplyToLen() {
6831 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6832
6833 address stubAddr = StubRoutines::multiplyToLen();
6834 if (stubAddr == nullptr) {
6835 return false; // Intrinsic's stub is not implemented on this platform
6836 }
6837 const char* stubName = "multiplyToLen";
6838
6839 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6840
6841 // no receiver because it is a static method
6842 Node* x = argument(0);
6843 Node* xlen = argument(1);
6844 Node* y = argument(2);
6845 Node* ylen = argument(3);
6846 Node* z = argument(4);
6847
6848 x = must_be_not_null(x, true);
6849 y = must_be_not_null(y, true);
6850
6851 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6852 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6853 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6854 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6855 // failed array check
6856 return false;
6857 }
6858
6859 BasicType x_elem = x_type->elem()->array_element_basic_type();
6860 BasicType y_elem = y_type->elem()->array_element_basic_type();
6861 if (x_elem != T_INT || y_elem != T_INT) {
6862 return false;
6863 }
6864
6865 Node* x_start = array_element_address(x, intcon(0), x_elem);
6866 Node* y_start = array_element_address(y, intcon(0), y_elem);
6867 // 'x_start' points to x array + scaled xlen
6868 // 'y_start' points to y array + scaled ylen
6869
6870 Node* z_start = array_element_address(z, intcon(0), T_INT);
6871
6872 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6873 OptoRuntime::multiplyToLen_Type(),
6874 stubAddr, stubName, TypePtr::BOTTOM,
6875 x_start, xlen, y_start, ylen, z_start);
6876
6877 C->set_has_split_ifs(true); // Has chance for split-if optimization
6878 set_result(z);
6879 return true;
6880 }
6881
6882 //-------------inline_squareToLen------------------------------------
6883 bool LibraryCallKit::inline_squareToLen() {
6884 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6885
6886 address stubAddr = StubRoutines::squareToLen();
6887 if (stubAddr == nullptr) {
6888 return false; // Intrinsic's stub is not implemented on this platform
6889 }
6890 const char* stubName = "squareToLen";
6891
6892 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6893
6894 Node* x = argument(0);
6895 Node* len = argument(1);
6896 Node* z = argument(2);
6897 Node* zlen = argument(3);
6898
6899 x = must_be_not_null(x, true);
6900 z = must_be_not_null(z, true);
6901
6902 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6903 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6904 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6905 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6906 // failed array check
6907 return false;
6908 }
6909
6910 BasicType x_elem = x_type->elem()->array_element_basic_type();
6911 BasicType z_elem = z_type->elem()->array_element_basic_type();
6912 if (x_elem != T_INT || z_elem != T_INT) {
6913 return false;
6914 }
6915
6916
6917 Node* x_start = array_element_address(x, intcon(0), x_elem);
6918 Node* z_start = array_element_address(z, intcon(0), z_elem);
6919
6920 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6921 OptoRuntime::squareToLen_Type(),
6922 stubAddr, stubName, TypePtr::BOTTOM,
6923 x_start, len, z_start, zlen);
6924
6925 set_result(z);
6926 return true;
6927 }
6928
6929 //-------------inline_mulAdd------------------------------------------
6930 bool LibraryCallKit::inline_mulAdd() {
6931 assert(UseMulAddIntrinsic, "not implemented on this platform");
6932
6933 address stubAddr = StubRoutines::mulAdd();
6934 if (stubAddr == nullptr) {
6935 return false; // Intrinsic's stub is not implemented on this platform
6936 }
6937 const char* stubName = "mulAdd";
6938
6939 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6940
6941 Node* out = argument(0);
6942 Node* in = argument(1);
6943 Node* offset = argument(2);
6944 Node* len = argument(3);
6945 Node* k = argument(4);
6946
6947 in = must_be_not_null(in, true);
6948 out = must_be_not_null(out, true);
6949
6950 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6951 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6952 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6953 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6954 // failed array check
6955 return false;
6956 }
6957
6958 BasicType out_elem = out_type->elem()->array_element_basic_type();
6959 BasicType in_elem = in_type->elem()->array_element_basic_type();
6960 if (out_elem != T_INT || in_elem != T_INT) {
6961 return false;
6962 }
6963
6964 Node* outlen = load_array_length(out);
6965 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6966 Node* out_start = array_element_address(out, intcon(0), out_elem);
6967 Node* in_start = array_element_address(in, intcon(0), in_elem);
6968
6969 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6970 OptoRuntime::mulAdd_Type(),
6971 stubAddr, stubName, TypePtr::BOTTOM,
6972 out_start,in_start, new_offset, len, k);
6973 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6974 set_result(result);
6975 return true;
6976 }
6977
6978 //-------------inline_montgomeryMultiply-----------------------------------
6979 bool LibraryCallKit::inline_montgomeryMultiply() {
6980 address stubAddr = StubRoutines::montgomeryMultiply();
6981 if (stubAddr == nullptr) {
6982 return false; // Intrinsic's stub is not implemented on this platform
6983 }
6984
6985 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6986 const char* stubName = "montgomery_multiply";
6987
6988 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6989
6990 Node* a = argument(0);
6991 Node* b = argument(1);
6992 Node* n = argument(2);
6993 Node* len = argument(3);
6994 Node* inv = argument(4);
6995 Node* m = argument(6);
6996
6997 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6998 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6999 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7000 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7001 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7002 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7003 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7004 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7005 // failed array check
7006 return false;
7007 }
7008
7009 BasicType a_elem = a_type->elem()->array_element_basic_type();
7010 BasicType b_elem = b_type->elem()->array_element_basic_type();
7011 BasicType n_elem = n_type->elem()->array_element_basic_type();
7012 BasicType m_elem = m_type->elem()->array_element_basic_type();
7013 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7014 return false;
7015 }
7016
7017 // Make the call
7018 {
7019 Node* a_start = array_element_address(a, intcon(0), a_elem);
7020 Node* b_start = array_element_address(b, intcon(0), b_elem);
7021 Node* n_start = array_element_address(n, intcon(0), n_elem);
7022 Node* m_start = array_element_address(m, intcon(0), m_elem);
7023
7024 Node* call = make_runtime_call(RC_LEAF,
7025 OptoRuntime::montgomeryMultiply_Type(),
7026 stubAddr, stubName, TypePtr::BOTTOM,
7027 a_start, b_start, n_start, len, inv, top(),
7028 m_start);
7029 set_result(m);
7030 }
7031
7032 return true;
7033 }
7034
7035 bool LibraryCallKit::inline_montgomerySquare() {
7036 address stubAddr = StubRoutines::montgomerySquare();
7037 if (stubAddr == nullptr) {
7038 return false; // Intrinsic's stub is not implemented on this platform
7039 }
7040
7041 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7042 const char* stubName = "montgomery_square";
7043
7044 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7045
7046 Node* a = argument(0);
7047 Node* n = argument(1);
7048 Node* len = argument(2);
7049 Node* inv = argument(3);
7050 Node* m = argument(5);
7051
7052 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7053 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7054 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7055 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7056 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7057 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7058 // failed array check
7059 return false;
7060 }
7061
7062 BasicType a_elem = a_type->elem()->array_element_basic_type();
7063 BasicType n_elem = n_type->elem()->array_element_basic_type();
7064 BasicType m_elem = m_type->elem()->array_element_basic_type();
7065 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7066 return false;
7067 }
7068
7069 // Make the call
7070 {
7071 Node* a_start = array_element_address(a, intcon(0), a_elem);
7072 Node* n_start = array_element_address(n, intcon(0), n_elem);
7073 Node* m_start = array_element_address(m, intcon(0), m_elem);
7074
7075 Node* call = make_runtime_call(RC_LEAF,
7076 OptoRuntime::montgomerySquare_Type(),
7077 stubAddr, stubName, TypePtr::BOTTOM,
7078 a_start, n_start, len, inv, top(),
7079 m_start);
7080 set_result(m);
7081 }
7082
7083 return true;
7084 }
7085
7086 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7087 address stubAddr = nullptr;
7088 const char* stubName = nullptr;
7089
7090 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7091 if (stubAddr == nullptr) {
7092 return false; // Intrinsic's stub is not implemented on this platform
7093 }
7094
7095 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7096
7097 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7098
7099 Node* newArr = argument(0);
7100 Node* oldArr = argument(1);
7101 Node* newIdx = argument(2);
7102 Node* shiftCount = argument(3);
7103 Node* numIter = argument(4);
7104
7105 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7106 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7107 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7108 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7109 return false;
7110 }
7111
7112 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7113 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7114 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7115 return false;
7116 }
7117
7118 // Make the call
7119 {
7120 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7121 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7122
7123 Node* call = make_runtime_call(RC_LEAF,
7124 OptoRuntime::bigIntegerShift_Type(),
7125 stubAddr,
7126 stubName,
7127 TypePtr::BOTTOM,
7128 newArr_start,
7129 oldArr_start,
7130 newIdx,
7131 shiftCount,
7132 numIter);
7133 }
7134
7135 return true;
7136 }
7137
7138 //-------------inline_vectorizedMismatch------------------------------
7139 bool LibraryCallKit::inline_vectorizedMismatch() {
7140 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7141
7142 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7143 Node* obja = argument(0); // Object
7144 Node* aoffset = argument(1); // long
7145 Node* objb = argument(3); // Object
7146 Node* boffset = argument(4); // long
7147 Node* length = argument(6); // int
7148 Node* scale = argument(7); // int
7149
7150 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7151 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7152 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7153 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7154 scale == top()) {
7155 return false; // failed input validation
7156 }
7157
7158 Node* obja_adr = make_unsafe_address(obja, aoffset);
7159 Node* objb_adr = make_unsafe_address(objb, boffset);
7160
7161 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7162 //
7163 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7164 // if (length <= inline_limit) {
7165 // inline_path:
7166 // vmask = VectorMaskGen length
7167 // vload1 = LoadVectorMasked obja, vmask
7168 // vload2 = LoadVectorMasked objb, vmask
7169 // result1 = VectorCmpMasked vload1, vload2, vmask
7170 // } else {
7171 // call_stub_path:
7172 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7173 // }
7174 // exit_block:
7175 // return Phi(result1, result2);
7176 //
7177 enum { inline_path = 1, // input is small enough to process it all at once
7178 stub_path = 2, // input is too large; call into the VM
7179 PATH_LIMIT = 3
7180 };
7181
7182 Node* exit_block = new RegionNode(PATH_LIMIT);
7183 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7184 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7185
7186 Node* call_stub_path = control();
7187
7188 BasicType elem_bt = T_ILLEGAL;
7189
7190 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7191 if (scale_t->is_con()) {
7192 switch (scale_t->get_con()) {
7193 case 0: elem_bt = T_BYTE; break;
7194 case 1: elem_bt = T_SHORT; break;
7195 case 2: elem_bt = T_INT; break;
7196 case 3: elem_bt = T_LONG; break;
7197
7198 default: elem_bt = T_ILLEGAL; break; // not supported
7199 }
7200 }
7201
7202 int inline_limit = 0;
7203 bool do_partial_inline = false;
7204
7205 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7206 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7207 do_partial_inline = inline_limit >= 16;
7208 }
7209
7210 if (do_partial_inline) {
7211 assert(elem_bt != T_ILLEGAL, "sanity");
7212
7213 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7214 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7215 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7216
7217 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7218 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7219 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7220
7221 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7222
7223 if (!stopped()) {
7224 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7225
7226 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7227 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7228 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7229 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7230
7231 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7232 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7233 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7234 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7235
7236 exit_block->init_req(inline_path, control());
7237 memory_phi->init_req(inline_path, map()->memory());
7238 result_phi->init_req(inline_path, result);
7239
7240 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7241 clear_upper_avx();
7242 }
7243 }
7244 }
7245
7246 if (call_stub_path != nullptr) {
7247 set_control(call_stub_path);
7248
7249 Node* call = make_runtime_call(RC_LEAF,
7250 OptoRuntime::vectorizedMismatch_Type(),
7251 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7252 obja_adr, objb_adr, length, scale);
7253
7254 exit_block->init_req(stub_path, control());
7255 memory_phi->init_req(stub_path, map()->memory());
7256 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7257 }
7258
7259 exit_block = _gvn.transform(exit_block);
7260 memory_phi = _gvn.transform(memory_phi);
7261 result_phi = _gvn.transform(result_phi);
7262
7263 record_for_igvn(exit_block);
7264 record_for_igvn(memory_phi);
7265 record_for_igvn(result_phi);
7266
7267 set_control(exit_block);
7268 set_all_memory(memory_phi);
7269 set_result(result_phi);
7270
7271 return true;
7272 }
7273
7274 //------------------------------inline_vectorizedHashcode----------------------------
7275 bool LibraryCallKit::inline_vectorizedHashCode() {
7276 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7277
7278 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7279 Node* array = argument(0);
7280 Node* offset = argument(1);
7281 Node* length = argument(2);
7282 Node* initialValue = argument(3);
7283 Node* basic_type = argument(4);
7284
7285 if (basic_type == top()) {
7286 return false; // failed input validation
7287 }
7288
7289 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7290 if (!basic_type_t->is_con()) {
7291 return false; // Only intrinsify if mode argument is constant
7292 }
7293
7294 array = must_be_not_null(array, true);
7295
7296 BasicType bt = (BasicType)basic_type_t->get_con();
7297
7298 // Resolve address of first element
7299 Node* array_start = array_element_address(array, offset, bt);
7300
7301 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7302 array_start, length, initialValue, basic_type)));
7303 clear_upper_avx();
7304
7305 return true;
7306 }
7307
7308 /**
7309 * Calculate CRC32 for byte.
7310 * int java.util.zip.CRC32.update(int crc, int b)
7311 */
7312 bool LibraryCallKit::inline_updateCRC32() {
7313 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7314 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7315 // no receiver since it is static method
7316 Node* crc = argument(0); // type: int
7317 Node* b = argument(1); // type: int
7318
7319 /*
7320 * int c = ~ crc;
7321 * b = timesXtoThe32[(b ^ c) & 0xFF];
7322 * b = b ^ (c >>> 8);
7323 * crc = ~b;
7324 */
7325
7326 Node* M1 = intcon(-1);
7327 crc = _gvn.transform(new XorINode(crc, M1));
7328 Node* result = _gvn.transform(new XorINode(crc, b));
7329 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7330
7331 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7332 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7333 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7334 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7335
7336 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7337 result = _gvn.transform(new XorINode(crc, result));
7338 result = _gvn.transform(new XorINode(result, M1));
7339 set_result(result);
7340 return true;
7341 }
7342
7343 /**
7344 * Calculate CRC32 for byte[] array.
7345 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7346 */
7347 bool LibraryCallKit::inline_updateBytesCRC32() {
7348 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7349 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7350 // no receiver since it is static method
7351 Node* crc = argument(0); // type: int
7352 Node* src = argument(1); // type: oop
7353 Node* offset = argument(2); // type: int
7354 Node* length = argument(3); // type: int
7355
7356 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7357 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7358 // failed array check
7359 return false;
7360 }
7361
7362 // Figure out the size and type of the elements we will be copying.
7363 BasicType src_elem = src_type->elem()->array_element_basic_type();
7364 if (src_elem != T_BYTE) {
7365 return false;
7366 }
7367
7368 // 'src_start' points to src array + scaled offset
7369 src = must_be_not_null(src, true);
7370 Node* src_start = array_element_address(src, offset, src_elem);
7371
7372 // We assume that range check is done by caller.
7373 // TODO: generate range check (offset+length < src.length) in debug VM.
7374
7375 // Call the stub.
7376 address stubAddr = StubRoutines::updateBytesCRC32();
7377 const char *stubName = "updateBytesCRC32";
7378
7379 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7380 stubAddr, stubName, TypePtr::BOTTOM,
7381 crc, src_start, length);
7382 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7383 set_result(result);
7384 return true;
7385 }
7386
7387 /**
7388 * Calculate CRC32 for ByteBuffer.
7389 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7390 */
7391 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7392 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7393 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7394 // no receiver since it is static method
7395 Node* crc = argument(0); // type: int
7396 Node* src = argument(1); // type: long
7397 Node* offset = argument(3); // type: int
7398 Node* length = argument(4); // type: int
7399
7400 src = ConvL2X(src); // adjust Java long to machine word
7401 Node* base = _gvn.transform(new CastX2PNode(src));
7402 offset = ConvI2X(offset);
7403
7404 // 'src_start' points to src array + scaled offset
7405 Node* src_start = basic_plus_adr(top(), base, offset);
7406
7407 // Call the stub.
7408 address stubAddr = StubRoutines::updateBytesCRC32();
7409 const char *stubName = "updateBytesCRC32";
7410
7411 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7412 stubAddr, stubName, TypePtr::BOTTOM,
7413 crc, src_start, length);
7414 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7415 set_result(result);
7416 return true;
7417 }
7418
7419 //------------------------------get_table_from_crc32c_class-----------------------
7420 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7421 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7422 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7423
7424 return table;
7425 }
7426
7427 //------------------------------inline_updateBytesCRC32C-----------------------
7428 //
7429 // Calculate CRC32C for byte[] array.
7430 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7431 //
7432 bool LibraryCallKit::inline_updateBytesCRC32C() {
7433 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7434 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7435 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7436 // no receiver since it is a static method
7437 Node* crc = argument(0); // type: int
7438 Node* src = argument(1); // type: oop
7439 Node* offset = argument(2); // type: int
7440 Node* end = argument(3); // type: int
7441
7442 Node* length = _gvn.transform(new SubINode(end, offset));
7443
7444 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7445 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7446 // failed array check
7447 return false;
7448 }
7449
7450 // Figure out the size and type of the elements we will be copying.
7451 BasicType src_elem = src_type->elem()->array_element_basic_type();
7452 if (src_elem != T_BYTE) {
7453 return false;
7454 }
7455
7456 // 'src_start' points to src array + scaled offset
7457 src = must_be_not_null(src, true);
7458 Node* src_start = array_element_address(src, offset, src_elem);
7459
7460 // static final int[] byteTable in class CRC32C
7461 Node* table = get_table_from_crc32c_class(callee()->holder());
7462 table = must_be_not_null(table, true);
7463 Node* table_start = array_element_address(table, intcon(0), T_INT);
7464
7465 // We assume that range check is done by caller.
7466 // TODO: generate range check (offset+length < src.length) in debug VM.
7467
7468 // Call the stub.
7469 address stubAddr = StubRoutines::updateBytesCRC32C();
7470 const char *stubName = "updateBytesCRC32C";
7471
7472 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7473 stubAddr, stubName, TypePtr::BOTTOM,
7474 crc, src_start, length, table_start);
7475 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7476 set_result(result);
7477 return true;
7478 }
7479
7480 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7481 //
7482 // Calculate CRC32C for DirectByteBuffer.
7483 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7484 //
7485 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7486 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7487 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7488 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7489 // no receiver since it is a static method
7490 Node* crc = argument(0); // type: int
7491 Node* src = argument(1); // type: long
7492 Node* offset = argument(3); // type: int
7493 Node* end = argument(4); // type: int
7494
7495 Node* length = _gvn.transform(new SubINode(end, offset));
7496
7497 src = ConvL2X(src); // adjust Java long to machine word
7498 Node* base = _gvn.transform(new CastX2PNode(src));
7499 offset = ConvI2X(offset);
7500
7501 // 'src_start' points to src array + scaled offset
7502 Node* src_start = basic_plus_adr(top(), base, offset);
7503
7504 // static final int[] byteTable in class CRC32C
7505 Node* table = get_table_from_crc32c_class(callee()->holder());
7506 table = must_be_not_null(table, true);
7507 Node* table_start = array_element_address(table, intcon(0), T_INT);
7508
7509 // Call the stub.
7510 address stubAddr = StubRoutines::updateBytesCRC32C();
7511 const char *stubName = "updateBytesCRC32C";
7512
7513 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7514 stubAddr, stubName, TypePtr::BOTTOM,
7515 crc, src_start, length, table_start);
7516 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7517 set_result(result);
7518 return true;
7519 }
7520
7521 //------------------------------inline_updateBytesAdler32----------------------
7522 //
7523 // Calculate Adler32 checksum for byte[] array.
7524 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7525 //
7526 bool LibraryCallKit::inline_updateBytesAdler32() {
7527 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7528 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7529 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7530 // no receiver since it is static method
7531 Node* crc = argument(0); // type: int
7532 Node* src = argument(1); // type: oop
7533 Node* offset = argument(2); // type: int
7534 Node* length = argument(3); // type: int
7535
7536 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7537 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7538 // failed array check
7539 return false;
7540 }
7541
7542 // Figure out the size and type of the elements we will be copying.
7543 BasicType src_elem = src_type->elem()->array_element_basic_type();
7544 if (src_elem != T_BYTE) {
7545 return false;
7546 }
7547
7548 // 'src_start' points to src array + scaled offset
7549 Node* src_start = array_element_address(src, offset, src_elem);
7550
7551 // We assume that range check is done by caller.
7552 // TODO: generate range check (offset+length < src.length) in debug VM.
7553
7554 // Call the stub.
7555 address stubAddr = StubRoutines::updateBytesAdler32();
7556 const char *stubName = "updateBytesAdler32";
7557
7558 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7559 stubAddr, stubName, TypePtr::BOTTOM,
7560 crc, src_start, length);
7561 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7562 set_result(result);
7563 return true;
7564 }
7565
7566 //------------------------------inline_updateByteBufferAdler32---------------
7567 //
7568 // Calculate Adler32 checksum for DirectByteBuffer.
7569 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7570 //
7571 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7572 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7573 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7574 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7575 // no receiver since it is static method
7576 Node* crc = argument(0); // type: int
7577 Node* src = argument(1); // type: long
7578 Node* offset = argument(3); // type: int
7579 Node* length = argument(4); // type: int
7580
7581 src = ConvL2X(src); // adjust Java long to machine word
7582 Node* base = _gvn.transform(new CastX2PNode(src));
7583 offset = ConvI2X(offset);
7584
7585 // 'src_start' points to src array + scaled offset
7586 Node* src_start = basic_plus_adr(top(), base, offset);
7587
7588 // Call the stub.
7589 address stubAddr = StubRoutines::updateBytesAdler32();
7590 const char *stubName = "updateBytesAdler32";
7591
7592 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7593 stubAddr, stubName, TypePtr::BOTTOM,
7594 crc, src_start, length);
7595
7596 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7597 set_result(result);
7598 return true;
7599 }
7600
7601 //----------------------------inline_reference_get0----------------------------
7602 // public T java.lang.ref.Reference.get();
7603 bool LibraryCallKit::inline_reference_get0() {
7604 const int referent_offset = java_lang_ref_Reference::referent_offset();
7605
7606 // Get the argument:
7607 Node* reference_obj = null_check_receiver();
7608 if (stopped()) return true;
7609
7610 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7611 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7612 decorators, /*is_static*/ false, nullptr);
7613 if (result == nullptr) return false;
7614
7615 // Add memory barrier to prevent commoning reads from this field
7616 // across safepoint since GC can change its value.
7617 insert_mem_bar(Op_MemBarCPUOrder);
7618
7619 set_result(result);
7620 return true;
7621 }
7622
7623 //----------------------------inline_reference_refersTo0----------------------------
7624 // bool java.lang.ref.Reference.refersTo0();
7625 // bool java.lang.ref.PhantomReference.refersTo0();
7626 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7627 // Get arguments:
7628 Node* reference_obj = null_check_receiver();
7629 Node* other_obj = argument(1);
7630 if (stopped()) return true;
7631
7632 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7633 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7634 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7635 decorators, /*is_static*/ false, nullptr);
7636 if (referent == nullptr) return false;
7637
7638 // Add memory barrier to prevent commoning reads from this field
7639 // across safepoint since GC can change its value.
7640 insert_mem_bar(Op_MemBarCPUOrder);
7641
7642 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7643 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7644 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7645
7646 RegionNode* region = new RegionNode(3);
7647 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7648
7649 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7650 region->init_req(1, if_true);
7651 phi->init_req(1, intcon(1));
7652
7653 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7654 region->init_req(2, if_false);
7655 phi->init_req(2, intcon(0));
7656
7657 set_control(_gvn.transform(region));
7658 record_for_igvn(region);
7659 set_result(_gvn.transform(phi));
7660 return true;
7661 }
7662
7663 //----------------------------inline_reference_clear0----------------------------
7664 // void java.lang.ref.Reference.clear0();
7665 // void java.lang.ref.PhantomReference.clear0();
7666 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7667 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7668
7669 // Get arguments
7670 Node* reference_obj = null_check_receiver();
7671 if (stopped()) return true;
7672
7673 // Common access parameters
7674 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7675 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7676 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7677 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7678 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7679
7680 Node* referent = access_load_at(reference_obj,
7681 referent_field_addr,
7682 referent_field_addr_type,
7683 val_type,
7684 T_OBJECT,
7685 decorators);
7686
7687 IdealKit ideal(this);
7688 #define __ ideal.
7689 __ if_then(referent, BoolTest::ne, null());
7690 sync_kit(ideal);
7691 access_store_at(reference_obj,
7692 referent_field_addr,
7693 referent_field_addr_type,
7694 null(),
7695 val_type,
7696 T_OBJECT,
7697 decorators);
7698 __ sync_kit(this);
7699 __ end_if();
7700 final_sync(ideal);
7701 #undef __
7702
7703 return true;
7704 }
7705
7706 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7707 DecoratorSet decorators, bool is_static,
7708 ciInstanceKlass* fromKls) {
7709 if (fromKls == nullptr) {
7710 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7711 assert(tinst != nullptr, "obj is null");
7712 assert(tinst->is_loaded(), "obj is not loaded");
7713 fromKls = tinst->instance_klass();
7714 } else {
7715 assert(is_static, "only for static field access");
7716 }
7717 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7718 ciSymbol::make(fieldTypeString),
7719 is_static);
7720
7721 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7722 if (field == nullptr) return (Node *) nullptr;
7723
7724 if (is_static) {
7725 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7726 fromObj = makecon(tip);
7727 }
7728
7729 // Next code copied from Parse::do_get_xxx():
7730
7731 // Compute address and memory type.
7732 int offset = field->offset_in_bytes();
7733 bool is_vol = field->is_volatile();
7734 ciType* field_klass = field->type();
7735 assert(field_klass->is_loaded(), "should be loaded");
7736 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7737 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7738 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7739 "slice of address and input slice don't match");
7740 BasicType bt = field->layout_type();
7741
7742 // Build the resultant type of the load
7743 const Type *type;
7744 if (bt == T_OBJECT) {
7745 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7746 } else {
7747 type = Type::get_const_basic_type(bt);
7748 }
7749
7750 if (is_vol) {
7751 decorators |= MO_SEQ_CST;
7752 }
7753
7754 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7755 }
7756
7757 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7758 bool is_exact /* true */, bool is_static /* false */,
7759 ciInstanceKlass * fromKls /* nullptr */) {
7760 if (fromKls == nullptr) {
7761 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7762 assert(tinst != nullptr, "obj is null");
7763 assert(tinst->is_loaded(), "obj is not loaded");
7764 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7765 fromKls = tinst->instance_klass();
7766 }
7767 else {
7768 assert(is_static, "only for static field access");
7769 }
7770 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7771 ciSymbol::make(fieldTypeString),
7772 is_static);
7773
7774 assert(field != nullptr, "undefined field");
7775 assert(!field->is_volatile(), "not defined for volatile fields");
7776
7777 if (is_static) {
7778 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7779 fromObj = makecon(tip);
7780 }
7781
7782 // Next code copied from Parse::do_get_xxx():
7783
7784 // Compute address and memory type.
7785 int offset = field->offset_in_bytes();
7786 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7787
7788 return adr;
7789 }
7790
7791 //------------------------------inline_aescrypt_Block-----------------------
7792 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7793 address stubAddr = nullptr;
7794 const char *stubName;
7795 assert(UseAES, "need AES instruction support");
7796
7797 switch(id) {
7798 case vmIntrinsics::_aescrypt_encryptBlock:
7799 stubAddr = StubRoutines::aescrypt_encryptBlock();
7800 stubName = "aescrypt_encryptBlock";
7801 break;
7802 case vmIntrinsics::_aescrypt_decryptBlock:
7803 stubAddr = StubRoutines::aescrypt_decryptBlock();
7804 stubName = "aescrypt_decryptBlock";
7805 break;
7806 default:
7807 break;
7808 }
7809 if (stubAddr == nullptr) return false;
7810
7811 Node* aescrypt_object = argument(0);
7812 Node* src = argument(1);
7813 Node* src_offset = argument(2);
7814 Node* dest = argument(3);
7815 Node* dest_offset = argument(4);
7816
7817 src = must_be_not_null(src, true);
7818 dest = must_be_not_null(dest, true);
7819
7820 // (1) src and dest are arrays.
7821 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7822 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7823 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7824 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7825
7826 // for the quick and dirty code we will skip all the checks.
7827 // we are just trying to get the call to be generated.
7828 Node* src_start = src;
7829 Node* dest_start = dest;
7830 if (src_offset != nullptr || dest_offset != nullptr) {
7831 assert(src_offset != nullptr && dest_offset != nullptr, "");
7832 src_start = array_element_address(src, src_offset, T_BYTE);
7833 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7834 }
7835
7836 // now need to get the start of its expanded key array
7837 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7838 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7839 if (k_start == nullptr) return false;
7840
7841 // Call the stub.
7842 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7843 stubAddr, stubName, TypePtr::BOTTOM,
7844 src_start, dest_start, k_start);
7845
7846 return true;
7847 }
7848
7849 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7850 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7851 address stubAddr = nullptr;
7852 const char *stubName = nullptr;
7853
7854 assert(UseAES, "need AES instruction support");
7855
7856 switch(id) {
7857 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7858 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7859 stubName = "cipherBlockChaining_encryptAESCrypt";
7860 break;
7861 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7862 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7863 stubName = "cipherBlockChaining_decryptAESCrypt";
7864 break;
7865 default:
7866 break;
7867 }
7868 if (stubAddr == nullptr) return false;
7869
7870 Node* cipherBlockChaining_object = argument(0);
7871 Node* src = argument(1);
7872 Node* src_offset = argument(2);
7873 Node* len = argument(3);
7874 Node* dest = argument(4);
7875 Node* dest_offset = argument(5);
7876
7877 src = must_be_not_null(src, false);
7878 dest = must_be_not_null(dest, false);
7879
7880 // (1) src and dest are arrays.
7881 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7882 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7883 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7884 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7885
7886 // checks are the responsibility of the caller
7887 Node* src_start = src;
7888 Node* dest_start = dest;
7889 if (src_offset != nullptr || dest_offset != nullptr) {
7890 assert(src_offset != nullptr && dest_offset != nullptr, "");
7891 src_start = array_element_address(src, src_offset, T_BYTE);
7892 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7893 }
7894
7895 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7896 // (because of the predicated logic executed earlier).
7897 // so we cast it here safely.
7898 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7899
7900 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7901 if (embeddedCipherObj == nullptr) return false;
7902
7903 // cast it to what we know it will be at runtime
7904 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7905 assert(tinst != nullptr, "CBC obj is null");
7906 assert(tinst->is_loaded(), "CBC obj is not loaded");
7907 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7908 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7909
7910 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7911 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7912 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7913 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7914 aescrypt_object = _gvn.transform(aescrypt_object);
7915
7916 // we need to get the start of the aescrypt_object's expanded key array
7917 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7918 if (k_start == nullptr) return false;
7919
7920 // similarly, get the start address of the r vector
7921 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7922 if (objRvec == nullptr) return false;
7923 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7924
7925 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7926 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7927 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7928 stubAddr, stubName, TypePtr::BOTTOM,
7929 src_start, dest_start, k_start, r_start, len);
7930
7931 // return cipher length (int)
7932 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7933 set_result(retvalue);
7934 return true;
7935 }
7936
7937 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7938 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7939 address stubAddr = nullptr;
7940 const char *stubName = nullptr;
7941
7942 assert(UseAES, "need AES instruction support");
7943
7944 switch (id) {
7945 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7946 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7947 stubName = "electronicCodeBook_encryptAESCrypt";
7948 break;
7949 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7950 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7951 stubName = "electronicCodeBook_decryptAESCrypt";
7952 break;
7953 default:
7954 break;
7955 }
7956
7957 if (stubAddr == nullptr) return false;
7958
7959 Node* electronicCodeBook_object = argument(0);
7960 Node* src = argument(1);
7961 Node* src_offset = argument(2);
7962 Node* len = argument(3);
7963 Node* dest = argument(4);
7964 Node* dest_offset = argument(5);
7965
7966 // (1) src and dest are arrays.
7967 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7968 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7969 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7970 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7971
7972 // checks are the responsibility of the caller
7973 Node* src_start = src;
7974 Node* dest_start = dest;
7975 if (src_offset != nullptr || dest_offset != nullptr) {
7976 assert(src_offset != nullptr && dest_offset != nullptr, "");
7977 src_start = array_element_address(src, src_offset, T_BYTE);
7978 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7979 }
7980
7981 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7982 // (because of the predicated logic executed earlier).
7983 // so we cast it here safely.
7984 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7985
7986 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7987 if (embeddedCipherObj == nullptr) return false;
7988
7989 // cast it to what we know it will be at runtime
7990 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7991 assert(tinst != nullptr, "ECB obj is null");
7992 assert(tinst->is_loaded(), "ECB obj is not loaded");
7993 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7994 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7995
7996 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7997 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7998 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7999 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8000 aescrypt_object = _gvn.transform(aescrypt_object);
8001
8002 // we need to get the start of the aescrypt_object's expanded key array
8003 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8004 if (k_start == nullptr) return false;
8005
8006 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8007 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8008 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8009 stubAddr, stubName, TypePtr::BOTTOM,
8010 src_start, dest_start, k_start, len);
8011
8012 // return cipher length (int)
8013 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8014 set_result(retvalue);
8015 return true;
8016 }
8017
8018 //------------------------------inline_counterMode_AESCrypt-----------------------
8019 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8020 assert(UseAES, "need AES instruction support");
8021 if (!UseAESCTRIntrinsics) return false;
8022
8023 address stubAddr = nullptr;
8024 const char *stubName = nullptr;
8025 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8026 stubAddr = StubRoutines::counterMode_AESCrypt();
8027 stubName = "counterMode_AESCrypt";
8028 }
8029 if (stubAddr == nullptr) return false;
8030
8031 Node* counterMode_object = argument(0);
8032 Node* src = argument(1);
8033 Node* src_offset = argument(2);
8034 Node* len = argument(3);
8035 Node* dest = argument(4);
8036 Node* dest_offset = argument(5);
8037
8038 // (1) src and dest are arrays.
8039 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8040 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8041 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8042 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8043
8044 // checks are the responsibility of the caller
8045 Node* src_start = src;
8046 Node* dest_start = dest;
8047 if (src_offset != nullptr || dest_offset != nullptr) {
8048 assert(src_offset != nullptr && dest_offset != nullptr, "");
8049 src_start = array_element_address(src, src_offset, T_BYTE);
8050 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8051 }
8052
8053 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8054 // (because of the predicated logic executed earlier).
8055 // so we cast it here safely.
8056 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8057 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8058 if (embeddedCipherObj == nullptr) return false;
8059 // cast it to what we know it will be at runtime
8060 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8061 assert(tinst != nullptr, "CTR obj is null");
8062 assert(tinst->is_loaded(), "CTR obj is not loaded");
8063 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8064 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
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 // we need to get the start of the aescrypt_object's expanded key array
8071 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8072 if (k_start == nullptr) return false;
8073 // similarly, get the start address of the r vector
8074 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8075 if (obj_counter == nullptr) return false;
8076 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8077
8078 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8079 if (saved_encCounter == nullptr) return false;
8080 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8081 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8082
8083 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8084 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8085 OptoRuntime::counterMode_aescrypt_Type(),
8086 stubAddr, stubName, TypePtr::BOTTOM,
8087 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8088
8089 // return cipher length (int)
8090 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8091 set_result(retvalue);
8092 return true;
8093 }
8094
8095 //------------------------------get_key_start_from_aescrypt_object-----------------------
8096 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
8097 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8098 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8099 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8100 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8101 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
8102 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
8103 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8104 if (objSessionK == nullptr) {
8105 return (Node *) nullptr;
8106 }
8107 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
8108 #else
8109 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
8110 #endif // PPC64
8111 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8112 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8113
8114 // now have the array, need to get the start address of the K array
8115 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8116 return k_start;
8117 }
8118
8119 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8120 // Return node representing slow path of predicate check.
8121 // the pseudo code we want to emulate with this predicate is:
8122 // for encryption:
8123 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8124 // for decryption:
8125 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8126 // note cipher==plain is more conservative than the original java code but that's OK
8127 //
8128 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8129 // The receiver was checked for null already.
8130 Node* objCBC = argument(0);
8131
8132 Node* src = argument(1);
8133 Node* dest = argument(4);
8134
8135 // Load embeddedCipher field of CipherBlockChaining object.
8136 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8137
8138 // get AESCrypt klass for instanceOf check
8139 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8140 // will have same classloader as CipherBlockChaining object
8141 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8142 assert(tinst != nullptr, "CBCobj is null");
8143 assert(tinst->is_loaded(), "CBCobj is not loaded");
8144
8145 // we want to do an instanceof comparison against the AESCrypt class
8146 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8147 if (!klass_AESCrypt->is_loaded()) {
8148 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8149 Node* ctrl = control();
8150 set_control(top()); // no regular fast path
8151 return ctrl;
8152 }
8153
8154 src = must_be_not_null(src, true);
8155 dest = must_be_not_null(dest, true);
8156
8157 // Resolve oops to stable for CmpP below.
8158 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8159
8160 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8161 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8162 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8163
8164 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8165
8166 // for encryption, we are done
8167 if (!decrypting)
8168 return instof_false; // even if it is null
8169
8170 // for decryption, we need to add a further check to avoid
8171 // taking the intrinsic path when cipher and plain are the same
8172 // see the original java code for why.
8173 RegionNode* region = new RegionNode(3);
8174 region->init_req(1, instof_false);
8175
8176 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8177 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8178 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8179 region->init_req(2, src_dest_conjoint);
8180
8181 record_for_igvn(region);
8182 return _gvn.transform(region);
8183 }
8184
8185 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8186 // Return node representing slow path of predicate check.
8187 // the pseudo code we want to emulate with this predicate is:
8188 // for encryption:
8189 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8190 // for decryption:
8191 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8192 // note cipher==plain is more conservative than the original java code but that's OK
8193 //
8194 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8195 // The receiver was checked for null already.
8196 Node* objECB = argument(0);
8197
8198 // Load embeddedCipher field of ElectronicCodeBook object.
8199 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8200
8201 // get AESCrypt klass for instanceOf check
8202 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8203 // will have same classloader as ElectronicCodeBook object
8204 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8205 assert(tinst != nullptr, "ECBobj is null");
8206 assert(tinst->is_loaded(), "ECBobj is not loaded");
8207
8208 // we want to do an instanceof comparison against the AESCrypt class
8209 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8210 if (!klass_AESCrypt->is_loaded()) {
8211 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8212 Node* ctrl = control();
8213 set_control(top()); // no regular fast path
8214 return ctrl;
8215 }
8216 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8217
8218 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8219 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8220 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8221
8222 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8223
8224 // for encryption, we are done
8225 if (!decrypting)
8226 return instof_false; // even if it is null
8227
8228 // for decryption, we need to add a further check to avoid
8229 // taking the intrinsic path when cipher and plain are the same
8230 // see the original java code for why.
8231 RegionNode* region = new RegionNode(3);
8232 region->init_req(1, instof_false);
8233 Node* src = argument(1);
8234 Node* dest = argument(4);
8235 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8236 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8237 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8238 region->init_req(2, src_dest_conjoint);
8239
8240 record_for_igvn(region);
8241 return _gvn.transform(region);
8242 }
8243
8244 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8245 // Return node representing slow path of predicate check.
8246 // the pseudo code we want to emulate with this predicate is:
8247 // for encryption:
8248 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8249 // for decryption:
8250 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8251 // note cipher==plain is more conservative than the original java code but that's OK
8252 //
8253
8254 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8255 // The receiver was checked for null already.
8256 Node* objCTR = argument(0);
8257
8258 // Load embeddedCipher field of CipherBlockChaining object.
8259 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8260
8261 // get AESCrypt klass for instanceOf check
8262 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8263 // will have same classloader as CipherBlockChaining object
8264 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8265 assert(tinst != nullptr, "CTRobj is null");
8266 assert(tinst->is_loaded(), "CTRobj is not loaded");
8267
8268 // we want to do an instanceof comparison against the AESCrypt class
8269 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8270 if (!klass_AESCrypt->is_loaded()) {
8271 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8272 Node* ctrl = control();
8273 set_control(top()); // no regular fast path
8274 return ctrl;
8275 }
8276
8277 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8278 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8279 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8280 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8281 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8282
8283 return instof_false; // even if it is null
8284 }
8285
8286 //------------------------------inline_ghash_processBlocks
8287 bool LibraryCallKit::inline_ghash_processBlocks() {
8288 address stubAddr;
8289 const char *stubName;
8290 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8291
8292 stubAddr = StubRoutines::ghash_processBlocks();
8293 stubName = "ghash_processBlocks";
8294
8295 Node* data = argument(0);
8296 Node* offset = argument(1);
8297 Node* len = argument(2);
8298 Node* state = argument(3);
8299 Node* subkeyH = argument(4);
8300
8301 state = must_be_not_null(state, true);
8302 subkeyH = must_be_not_null(subkeyH, true);
8303 data = must_be_not_null(data, true);
8304
8305 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8306 assert(state_start, "state is null");
8307 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8308 assert(subkeyH_start, "subkeyH is null");
8309 Node* data_start = array_element_address(data, offset, T_BYTE);
8310 assert(data_start, "data is null");
8311
8312 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8313 OptoRuntime::ghash_processBlocks_Type(),
8314 stubAddr, stubName, TypePtr::BOTTOM,
8315 state_start, subkeyH_start, data_start, len);
8316 return true;
8317 }
8318
8319 //------------------------------inline_chacha20Block
8320 bool LibraryCallKit::inline_chacha20Block() {
8321 address stubAddr;
8322 const char *stubName;
8323 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8324
8325 stubAddr = StubRoutines::chacha20Block();
8326 stubName = "chacha20Block";
8327
8328 Node* state = argument(0);
8329 Node* result = argument(1);
8330
8331 state = must_be_not_null(state, true);
8332 result = must_be_not_null(result, true);
8333
8334 Node* state_start = array_element_address(state, intcon(0), T_INT);
8335 assert(state_start, "state is null");
8336 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8337 assert(result_start, "result is null");
8338
8339 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8340 OptoRuntime::chacha20Block_Type(),
8341 stubAddr, stubName, TypePtr::BOTTOM,
8342 state_start, result_start);
8343 // return key stream length (int)
8344 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8345 set_result(retvalue);
8346 return true;
8347 }
8348
8349 //------------------------------inline_kyberNtt
8350 bool LibraryCallKit::inline_kyberNtt() {
8351 address stubAddr;
8352 const char *stubName;
8353 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8354 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8355
8356 stubAddr = StubRoutines::kyberNtt();
8357 stubName = "kyberNtt";
8358 if (!stubAddr) return false;
8359
8360 Node* coeffs = argument(0);
8361 Node* ntt_zetas = argument(1);
8362
8363 coeffs = must_be_not_null(coeffs, true);
8364 ntt_zetas = must_be_not_null(ntt_zetas, true);
8365
8366 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8367 assert(coeffs_start, "coeffs is null");
8368 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8369 assert(ntt_zetas_start, "ntt_zetas is null");
8370 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8371 OptoRuntime::kyberNtt_Type(),
8372 stubAddr, stubName, TypePtr::BOTTOM,
8373 coeffs_start, ntt_zetas_start);
8374 // return an int
8375 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8376 set_result(retvalue);
8377 return true;
8378 }
8379
8380 //------------------------------inline_kyberInverseNtt
8381 bool LibraryCallKit::inline_kyberInverseNtt() {
8382 address stubAddr;
8383 const char *stubName;
8384 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8385 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8386
8387 stubAddr = StubRoutines::kyberInverseNtt();
8388 stubName = "kyberInverseNtt";
8389 if (!stubAddr) return false;
8390
8391 Node* coeffs = argument(0);
8392 Node* zetas = argument(1);
8393
8394 coeffs = must_be_not_null(coeffs, true);
8395 zetas = must_be_not_null(zetas, true);
8396
8397 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8398 assert(coeffs_start, "coeffs is null");
8399 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8400 assert(zetas_start, "inverseNtt_zetas is null");
8401 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8402 OptoRuntime::kyberInverseNtt_Type(),
8403 stubAddr, stubName, TypePtr::BOTTOM,
8404 coeffs_start, zetas_start);
8405
8406 // return an int
8407 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8408 set_result(retvalue);
8409 return true;
8410 }
8411
8412 //------------------------------inline_kyberNttMult
8413 bool LibraryCallKit::inline_kyberNttMult() {
8414 address stubAddr;
8415 const char *stubName;
8416 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8417 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8418
8419 stubAddr = StubRoutines::kyberNttMult();
8420 stubName = "kyberNttMult";
8421 if (!stubAddr) return false;
8422
8423 Node* result = argument(0);
8424 Node* ntta = argument(1);
8425 Node* nttb = argument(2);
8426 Node* zetas = argument(3);
8427
8428 result = must_be_not_null(result, true);
8429 ntta = must_be_not_null(ntta, true);
8430 nttb = must_be_not_null(nttb, true);
8431 zetas = must_be_not_null(zetas, true);
8432
8433 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8434 assert(result_start, "result is null");
8435 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8436 assert(ntta_start, "ntta is null");
8437 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8438 assert(nttb_start, "nttb is null");
8439 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8440 assert(zetas_start, "nttMult_zetas is null");
8441 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8442 OptoRuntime::kyberNttMult_Type(),
8443 stubAddr, stubName, TypePtr::BOTTOM,
8444 result_start, ntta_start, nttb_start,
8445 zetas_start);
8446
8447 // return an int
8448 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8449 set_result(retvalue);
8450
8451 return true;
8452 }
8453
8454 //------------------------------inline_kyberAddPoly_2
8455 bool LibraryCallKit::inline_kyberAddPoly_2() {
8456 address stubAddr;
8457 const char *stubName;
8458 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8459 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8460
8461 stubAddr = StubRoutines::kyberAddPoly_2();
8462 stubName = "kyberAddPoly_2";
8463 if (!stubAddr) return false;
8464
8465 Node* result = argument(0);
8466 Node* a = argument(1);
8467 Node* b = argument(2);
8468
8469 result = must_be_not_null(result, true);
8470 a = must_be_not_null(a, true);
8471 b = must_be_not_null(b, true);
8472
8473 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8474 assert(result_start, "result is null");
8475 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8476 assert(a_start, "a is null");
8477 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8478 assert(b_start, "b is null");
8479 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8480 OptoRuntime::kyberAddPoly_2_Type(),
8481 stubAddr, stubName, TypePtr::BOTTOM,
8482 result_start, a_start, b_start);
8483 // return an int
8484 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8485 set_result(retvalue);
8486 return true;
8487 }
8488
8489 //------------------------------inline_kyberAddPoly_3
8490 bool LibraryCallKit::inline_kyberAddPoly_3() {
8491 address stubAddr;
8492 const char *stubName;
8493 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8494 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8495
8496 stubAddr = StubRoutines::kyberAddPoly_3();
8497 stubName = "kyberAddPoly_3";
8498 if (!stubAddr) return false;
8499
8500 Node* result = argument(0);
8501 Node* a = argument(1);
8502 Node* b = argument(2);
8503 Node* c = argument(3);
8504
8505 result = must_be_not_null(result, true);
8506 a = must_be_not_null(a, true);
8507 b = must_be_not_null(b, true);
8508 c = must_be_not_null(c, true);
8509
8510 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8511 assert(result_start, "result is null");
8512 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8513 assert(a_start, "a is null");
8514 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8515 assert(b_start, "b is null");
8516 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8517 assert(c_start, "c is null");
8518 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8519 OptoRuntime::kyberAddPoly_3_Type(),
8520 stubAddr, stubName, TypePtr::BOTTOM,
8521 result_start, a_start, b_start, c_start);
8522 // return an int
8523 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8524 set_result(retvalue);
8525 return true;
8526 }
8527
8528 //------------------------------inline_kyber12To16
8529 bool LibraryCallKit::inline_kyber12To16() {
8530 address stubAddr;
8531 const char *stubName;
8532 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8533 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8534
8535 stubAddr = StubRoutines::kyber12To16();
8536 stubName = "kyber12To16";
8537 if (!stubAddr) return false;
8538
8539 Node* condensed = argument(0);
8540 Node* condensedOffs = argument(1);
8541 Node* parsed = argument(2);
8542 Node* parsedLength = argument(3);
8543
8544 condensed = must_be_not_null(condensed, true);
8545 parsed = must_be_not_null(parsed, true);
8546
8547 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8548 assert(condensed_start, "condensed is null");
8549 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8550 assert(parsed_start, "parsed is null");
8551 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8552 OptoRuntime::kyber12To16_Type(),
8553 stubAddr, stubName, TypePtr::BOTTOM,
8554 condensed_start, condensedOffs, parsed_start, parsedLength);
8555 // return an int
8556 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8557 set_result(retvalue);
8558 return true;
8559
8560 }
8561
8562 //------------------------------inline_kyberBarrettReduce
8563 bool LibraryCallKit::inline_kyberBarrettReduce() {
8564 address stubAddr;
8565 const char *stubName;
8566 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8567 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8568
8569 stubAddr = StubRoutines::kyberBarrettReduce();
8570 stubName = "kyberBarrettReduce";
8571 if (!stubAddr) return false;
8572
8573 Node* coeffs = argument(0);
8574
8575 coeffs = must_be_not_null(coeffs, true);
8576
8577 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8578 assert(coeffs_start, "coeffs is null");
8579 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8580 OptoRuntime::kyberBarrettReduce_Type(),
8581 stubAddr, stubName, TypePtr::BOTTOM,
8582 coeffs_start);
8583 // return an int
8584 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8585 set_result(retvalue);
8586 return true;
8587 }
8588
8589 //------------------------------inline_dilithiumAlmostNtt
8590 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8591 address stubAddr;
8592 const char *stubName;
8593 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8594 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8595
8596 stubAddr = StubRoutines::dilithiumAlmostNtt();
8597 stubName = "dilithiumAlmostNtt";
8598 if (!stubAddr) return false;
8599
8600 Node* coeffs = argument(0);
8601 Node* ntt_zetas = argument(1);
8602
8603 coeffs = must_be_not_null(coeffs, true);
8604 ntt_zetas = must_be_not_null(ntt_zetas, true);
8605
8606 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8607 assert(coeffs_start, "coeffs is null");
8608 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8609 assert(ntt_zetas_start, "ntt_zetas is null");
8610 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8611 OptoRuntime::dilithiumAlmostNtt_Type(),
8612 stubAddr, stubName, TypePtr::BOTTOM,
8613 coeffs_start, ntt_zetas_start);
8614 // return an int
8615 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8616 set_result(retvalue);
8617 return true;
8618 }
8619
8620 //------------------------------inline_dilithiumAlmostInverseNtt
8621 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8622 address stubAddr;
8623 const char *stubName;
8624 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8625 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8626
8627 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8628 stubName = "dilithiumAlmostInverseNtt";
8629 if (!stubAddr) return false;
8630
8631 Node* coeffs = argument(0);
8632 Node* zetas = argument(1);
8633
8634 coeffs = must_be_not_null(coeffs, true);
8635 zetas = must_be_not_null(zetas, true);
8636
8637 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8638 assert(coeffs_start, "coeffs is null");
8639 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8640 assert(zetas_start, "inverseNtt_zetas is null");
8641 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8642 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8643 stubAddr, stubName, TypePtr::BOTTOM,
8644 coeffs_start, zetas_start);
8645 // return an int
8646 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8647 set_result(retvalue);
8648 return true;
8649 }
8650
8651 //------------------------------inline_dilithiumNttMult
8652 bool LibraryCallKit::inline_dilithiumNttMult() {
8653 address stubAddr;
8654 const char *stubName;
8655 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8656 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8657
8658 stubAddr = StubRoutines::dilithiumNttMult();
8659 stubName = "dilithiumNttMult";
8660 if (!stubAddr) return false;
8661
8662 Node* result = argument(0);
8663 Node* ntta = argument(1);
8664 Node* nttb = argument(2);
8665 Node* zetas = argument(3);
8666
8667 result = must_be_not_null(result, true);
8668 ntta = must_be_not_null(ntta, true);
8669 nttb = must_be_not_null(nttb, true);
8670 zetas = must_be_not_null(zetas, true);
8671
8672 Node* result_start = array_element_address(result, intcon(0), T_INT);
8673 assert(result_start, "result is null");
8674 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8675 assert(ntta_start, "ntta is null");
8676 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8677 assert(nttb_start, "nttb is null");
8678 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8679 OptoRuntime::dilithiumNttMult_Type(),
8680 stubAddr, stubName, TypePtr::BOTTOM,
8681 result_start, ntta_start, nttb_start);
8682
8683 // return an int
8684 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8685 set_result(retvalue);
8686
8687 return true;
8688 }
8689
8690 //------------------------------inline_dilithiumMontMulByConstant
8691 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8692 address stubAddr;
8693 const char *stubName;
8694 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8695 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8696
8697 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8698 stubName = "dilithiumMontMulByConstant";
8699 if (!stubAddr) return false;
8700
8701 Node* coeffs = argument(0);
8702 Node* constant = argument(1);
8703
8704 coeffs = must_be_not_null(coeffs, true);
8705
8706 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8707 assert(coeffs_start, "coeffs is null");
8708 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8709 OptoRuntime::dilithiumMontMulByConstant_Type(),
8710 stubAddr, stubName, TypePtr::BOTTOM,
8711 coeffs_start, constant);
8712
8713 // return an int
8714 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8715 set_result(retvalue);
8716 return true;
8717 }
8718
8719
8720 //------------------------------inline_dilithiumDecomposePoly
8721 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8722 address stubAddr;
8723 const char *stubName;
8724 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8725 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8726
8727 stubAddr = StubRoutines::dilithiumDecomposePoly();
8728 stubName = "dilithiumDecomposePoly";
8729 if (!stubAddr) return false;
8730
8731 Node* input = argument(0);
8732 Node* lowPart = argument(1);
8733 Node* highPart = argument(2);
8734 Node* twoGamma2 = argument(3);
8735 Node* multiplier = argument(4);
8736
8737 input = must_be_not_null(input, true);
8738 lowPart = must_be_not_null(lowPart, true);
8739 highPart = must_be_not_null(highPart, true);
8740
8741 Node* input_start = array_element_address(input, intcon(0), T_INT);
8742 assert(input_start, "input is null");
8743 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8744 assert(lowPart_start, "lowPart is null");
8745 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8746 assert(highPart_start, "highPart is null");
8747
8748 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8749 OptoRuntime::dilithiumDecomposePoly_Type(),
8750 stubAddr, stubName, TypePtr::BOTTOM,
8751 input_start, lowPart_start, highPart_start,
8752 twoGamma2, multiplier);
8753
8754 // return an int
8755 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8756 set_result(retvalue);
8757 return true;
8758 }
8759
8760 bool LibraryCallKit::inline_base64_encodeBlock() {
8761 address stubAddr;
8762 const char *stubName;
8763 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8764 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8765 stubAddr = StubRoutines::base64_encodeBlock();
8766 stubName = "encodeBlock";
8767
8768 if (!stubAddr) return false;
8769 Node* base64obj = argument(0);
8770 Node* src = argument(1);
8771 Node* offset = argument(2);
8772 Node* len = argument(3);
8773 Node* dest = argument(4);
8774 Node* dp = argument(5);
8775 Node* isURL = argument(6);
8776
8777 src = must_be_not_null(src, true);
8778 dest = must_be_not_null(dest, true);
8779
8780 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8781 assert(src_start, "source array is null");
8782 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8783 assert(dest_start, "destination array is null");
8784
8785 Node* base64 = make_runtime_call(RC_LEAF,
8786 OptoRuntime::base64_encodeBlock_Type(),
8787 stubAddr, stubName, TypePtr::BOTTOM,
8788 src_start, offset, len, dest_start, dp, isURL);
8789 return true;
8790 }
8791
8792 bool LibraryCallKit::inline_base64_decodeBlock() {
8793 address stubAddr;
8794 const char *stubName;
8795 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8796 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8797 stubAddr = StubRoutines::base64_decodeBlock();
8798 stubName = "decodeBlock";
8799
8800 if (!stubAddr) return false;
8801 Node* base64obj = argument(0);
8802 Node* src = argument(1);
8803 Node* src_offset = argument(2);
8804 Node* len = argument(3);
8805 Node* dest = argument(4);
8806 Node* dest_offset = argument(5);
8807 Node* isURL = argument(6);
8808 Node* isMIME = argument(7);
8809
8810 src = must_be_not_null(src, true);
8811 dest = must_be_not_null(dest, true);
8812
8813 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8814 assert(src_start, "source array is null");
8815 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8816 assert(dest_start, "destination array is null");
8817
8818 Node* call = make_runtime_call(RC_LEAF,
8819 OptoRuntime::base64_decodeBlock_Type(),
8820 stubAddr, stubName, TypePtr::BOTTOM,
8821 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8822 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8823 set_result(result);
8824 return true;
8825 }
8826
8827 bool LibraryCallKit::inline_poly1305_processBlocks() {
8828 address stubAddr;
8829 const char *stubName;
8830 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8831 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8832 stubAddr = StubRoutines::poly1305_processBlocks();
8833 stubName = "poly1305_processBlocks";
8834
8835 if (!stubAddr) return false;
8836 null_check_receiver(); // null-check receiver
8837 if (stopped()) return true;
8838
8839 Node* input = argument(1);
8840 Node* input_offset = argument(2);
8841 Node* len = argument(3);
8842 Node* alimbs = argument(4);
8843 Node* rlimbs = argument(5);
8844
8845 input = must_be_not_null(input, true);
8846 alimbs = must_be_not_null(alimbs, true);
8847 rlimbs = must_be_not_null(rlimbs, true);
8848
8849 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8850 assert(input_start, "input array is null");
8851 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8852 assert(acc_start, "acc array is null");
8853 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8854 assert(r_start, "r array is null");
8855
8856 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8857 OptoRuntime::poly1305_processBlocks_Type(),
8858 stubAddr, stubName, TypePtr::BOTTOM,
8859 input_start, len, acc_start, r_start);
8860 return true;
8861 }
8862
8863 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8864 address stubAddr;
8865 const char *stubName;
8866 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8867 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8868 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8869 stubName = "intpoly_montgomeryMult_P256";
8870
8871 if (!stubAddr) return false;
8872 null_check_receiver(); // null-check receiver
8873 if (stopped()) return true;
8874
8875 Node* a = argument(1);
8876 Node* b = argument(2);
8877 Node* r = argument(3);
8878
8879 a = must_be_not_null(a, true);
8880 b = must_be_not_null(b, true);
8881 r = must_be_not_null(r, true);
8882
8883 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8884 assert(a_start, "a array is null");
8885 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8886 assert(b_start, "b array is null");
8887 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8888 assert(r_start, "r array is null");
8889
8890 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8891 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8892 stubAddr, stubName, TypePtr::BOTTOM,
8893 a_start, b_start, r_start);
8894 return true;
8895 }
8896
8897 bool LibraryCallKit::inline_intpoly_assign() {
8898 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8899 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8900 const char *stubName = "intpoly_assign";
8901 address stubAddr = StubRoutines::intpoly_assign();
8902 if (!stubAddr) return false;
8903
8904 Node* set = argument(0);
8905 Node* a = argument(1);
8906 Node* b = argument(2);
8907 Node* arr_length = load_array_length(a);
8908
8909 a = must_be_not_null(a, true);
8910 b = must_be_not_null(b, true);
8911
8912 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8913 assert(a_start, "a array is null");
8914 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8915 assert(b_start, "b array is null");
8916
8917 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8918 OptoRuntime::intpoly_assign_Type(),
8919 stubAddr, stubName, TypePtr::BOTTOM,
8920 set, a_start, b_start, arr_length);
8921 return true;
8922 }
8923
8924 //------------------------------inline_digestBase_implCompress-----------------------
8925 //
8926 // Calculate MD5 for single-block byte[] array.
8927 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8928 //
8929 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8930 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8931 //
8932 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8933 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8934 //
8935 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8936 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8937 //
8938 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8939 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8940 //
8941 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8942 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8943
8944 Node* digestBase_obj = argument(0);
8945 Node* src = argument(1); // type oop
8946 Node* ofs = argument(2); // type int
8947
8948 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8949 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8950 // failed array check
8951 return false;
8952 }
8953 // Figure out the size and type of the elements we will be copying.
8954 BasicType src_elem = src_type->elem()->array_element_basic_type();
8955 if (src_elem != T_BYTE) {
8956 return false;
8957 }
8958 // 'src_start' points to src array + offset
8959 src = must_be_not_null(src, true);
8960 Node* src_start = array_element_address(src, ofs, src_elem);
8961 Node* state = nullptr;
8962 Node* block_size = nullptr;
8963 address stubAddr;
8964 const char *stubName;
8965
8966 switch(id) {
8967 case vmIntrinsics::_md5_implCompress:
8968 assert(UseMD5Intrinsics, "need MD5 instruction support");
8969 state = get_state_from_digest_object(digestBase_obj, T_INT);
8970 stubAddr = StubRoutines::md5_implCompress();
8971 stubName = "md5_implCompress";
8972 break;
8973 case vmIntrinsics::_sha_implCompress:
8974 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8975 state = get_state_from_digest_object(digestBase_obj, T_INT);
8976 stubAddr = StubRoutines::sha1_implCompress();
8977 stubName = "sha1_implCompress";
8978 break;
8979 case vmIntrinsics::_sha2_implCompress:
8980 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8981 state = get_state_from_digest_object(digestBase_obj, T_INT);
8982 stubAddr = StubRoutines::sha256_implCompress();
8983 stubName = "sha256_implCompress";
8984 break;
8985 case vmIntrinsics::_sha5_implCompress:
8986 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8987 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8988 stubAddr = StubRoutines::sha512_implCompress();
8989 stubName = "sha512_implCompress";
8990 break;
8991 case vmIntrinsics::_sha3_implCompress:
8992 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8993 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8994 stubAddr = StubRoutines::sha3_implCompress();
8995 stubName = "sha3_implCompress";
8996 block_size = get_block_size_from_digest_object(digestBase_obj);
8997 if (block_size == nullptr) return false;
8998 break;
8999 default:
9000 fatal_unexpected_iid(id);
9001 return false;
9002 }
9003 if (state == nullptr) return false;
9004
9005 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9006 if (stubAddr == nullptr) return false;
9007
9008 // Call the stub.
9009 Node* call;
9010 if (block_size == nullptr) {
9011 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9012 stubAddr, stubName, TypePtr::BOTTOM,
9013 src_start, state);
9014 } else {
9015 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9016 stubAddr, stubName, TypePtr::BOTTOM,
9017 src_start, state, block_size);
9018 }
9019
9020 return true;
9021 }
9022
9023 //------------------------------inline_double_keccak
9024 bool LibraryCallKit::inline_double_keccak() {
9025 address stubAddr;
9026 const char *stubName;
9027 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9028 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9029
9030 stubAddr = StubRoutines::double_keccak();
9031 stubName = "double_keccak";
9032 if (!stubAddr) return false;
9033
9034 Node* status0 = argument(0);
9035 Node* status1 = argument(1);
9036
9037 status0 = must_be_not_null(status0, true);
9038 status1 = must_be_not_null(status1, true);
9039
9040 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9041 assert(status0_start, "status0 is null");
9042 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9043 assert(status1_start, "status1 is null");
9044 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9045 OptoRuntime::double_keccak_Type(),
9046 stubAddr, stubName, TypePtr::BOTTOM,
9047 status0_start, status1_start);
9048 // return an int
9049 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9050 set_result(retvalue);
9051 return true;
9052 }
9053
9054
9055 //------------------------------inline_digestBase_implCompressMB-----------------------
9056 //
9057 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9058 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9059 //
9060 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9061 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9062 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9063 assert((uint)predicate < 5, "sanity");
9064 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9065
9066 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9067 Node* src = argument(1); // byte[] array
9068 Node* ofs = argument(2); // type int
9069 Node* limit = argument(3); // type int
9070
9071 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9072 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9073 // failed array check
9074 return false;
9075 }
9076 // Figure out the size and type of the elements we will be copying.
9077 BasicType src_elem = src_type->elem()->array_element_basic_type();
9078 if (src_elem != T_BYTE) {
9079 return false;
9080 }
9081 // 'src_start' points to src array + offset
9082 src = must_be_not_null(src, false);
9083 Node* src_start = array_element_address(src, ofs, src_elem);
9084
9085 const char* klass_digestBase_name = nullptr;
9086 const char* stub_name = nullptr;
9087 address stub_addr = nullptr;
9088 BasicType elem_type = T_INT;
9089
9090 switch (predicate) {
9091 case 0:
9092 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9093 klass_digestBase_name = "sun/security/provider/MD5";
9094 stub_name = "md5_implCompressMB";
9095 stub_addr = StubRoutines::md5_implCompressMB();
9096 }
9097 break;
9098 case 1:
9099 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9100 klass_digestBase_name = "sun/security/provider/SHA";
9101 stub_name = "sha1_implCompressMB";
9102 stub_addr = StubRoutines::sha1_implCompressMB();
9103 }
9104 break;
9105 case 2:
9106 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9107 klass_digestBase_name = "sun/security/provider/SHA2";
9108 stub_name = "sha256_implCompressMB";
9109 stub_addr = StubRoutines::sha256_implCompressMB();
9110 }
9111 break;
9112 case 3:
9113 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9114 klass_digestBase_name = "sun/security/provider/SHA5";
9115 stub_name = "sha512_implCompressMB";
9116 stub_addr = StubRoutines::sha512_implCompressMB();
9117 elem_type = T_LONG;
9118 }
9119 break;
9120 case 4:
9121 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9122 klass_digestBase_name = "sun/security/provider/SHA3";
9123 stub_name = "sha3_implCompressMB";
9124 stub_addr = StubRoutines::sha3_implCompressMB();
9125 elem_type = T_LONG;
9126 }
9127 break;
9128 default:
9129 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9130 }
9131 if (klass_digestBase_name != nullptr) {
9132 assert(stub_addr != nullptr, "Stub is generated");
9133 if (stub_addr == nullptr) return false;
9134
9135 // get DigestBase klass to lookup for SHA klass
9136 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9137 assert(tinst != nullptr, "digestBase_obj is not instance???");
9138 assert(tinst->is_loaded(), "DigestBase is not loaded");
9139
9140 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9141 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9142 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9143 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9144 }
9145 return false;
9146 }
9147
9148 //------------------------------inline_digestBase_implCompressMB-----------------------
9149 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9150 BasicType elem_type, address stubAddr, const char *stubName,
9151 Node* src_start, Node* ofs, Node* limit) {
9152 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9153 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9154 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9155 digest_obj = _gvn.transform(digest_obj);
9156
9157 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9158 if (state == nullptr) return false;
9159
9160 Node* block_size = nullptr;
9161 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9162 block_size = get_block_size_from_digest_object(digest_obj);
9163 if (block_size == nullptr) return false;
9164 }
9165
9166 // Call the stub.
9167 Node* call;
9168 if (block_size == nullptr) {
9169 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9170 OptoRuntime::digestBase_implCompressMB_Type(false),
9171 stubAddr, stubName, TypePtr::BOTTOM,
9172 src_start, state, ofs, limit);
9173 } else {
9174 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9175 OptoRuntime::digestBase_implCompressMB_Type(true),
9176 stubAddr, stubName, TypePtr::BOTTOM,
9177 src_start, state, block_size, ofs, limit);
9178 }
9179
9180 // return ofs (int)
9181 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9182 set_result(result);
9183
9184 return true;
9185 }
9186
9187 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9188 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9189 assert(UseAES, "need AES instruction support");
9190 address stubAddr = nullptr;
9191 const char *stubName = nullptr;
9192 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9193 stubName = "galoisCounterMode_AESCrypt";
9194
9195 if (stubAddr == nullptr) return false;
9196
9197 Node* in = argument(0);
9198 Node* inOfs = argument(1);
9199 Node* len = argument(2);
9200 Node* ct = argument(3);
9201 Node* ctOfs = argument(4);
9202 Node* out = argument(5);
9203 Node* outOfs = argument(6);
9204 Node* gctr_object = argument(7);
9205 Node* ghash_object = argument(8);
9206
9207 // (1) in, ct and out are arrays.
9208 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9209 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9210 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9211 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9212 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9213 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9214
9215 // checks are the responsibility of the caller
9216 Node* in_start = in;
9217 Node* ct_start = ct;
9218 Node* out_start = out;
9219 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9220 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9221 in_start = array_element_address(in, inOfs, T_BYTE);
9222 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9223 out_start = array_element_address(out, outOfs, T_BYTE);
9224 }
9225
9226 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9227 // (because of the predicated logic executed earlier).
9228 // so we cast it here safely.
9229 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9230 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9231 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9232 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9233 Node* state = load_field_from_object(ghash_object, "state", "[J");
9234
9235 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9236 return false;
9237 }
9238 // cast it to what we know it will be at runtime
9239 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9240 assert(tinst != nullptr, "GCTR obj is null");
9241 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9242 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9243 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9244 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9245 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9246 const TypeOopPtr* xtype = aklass->as_instance_type();
9247 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9248 aescrypt_object = _gvn.transform(aescrypt_object);
9249 // we need to get the start of the aescrypt_object's expanded key array
9250 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9251 if (k_start == nullptr) return false;
9252 // similarly, get the start address of the r vector
9253 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9254 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9255 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9256
9257
9258 // Call the stub, passing params
9259 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9260 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9261 stubAddr, stubName, TypePtr::BOTTOM,
9262 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9263
9264 // return cipher length (int)
9265 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9266 set_result(retvalue);
9267
9268 return true;
9269 }
9270
9271 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9272 // Return node representing slow path of predicate check.
9273 // the pseudo code we want to emulate with this predicate is:
9274 // for encryption:
9275 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9276 // for decryption:
9277 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9278 // note cipher==plain is more conservative than the original java code but that's OK
9279 //
9280
9281 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9282 // The receiver was checked for null already.
9283 Node* objGCTR = argument(7);
9284 // Load embeddedCipher field of GCTR object.
9285 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9286 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9287
9288 // get AESCrypt klass for instanceOf check
9289 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9290 // will have same classloader as CipherBlockChaining object
9291 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9292 assert(tinst != nullptr, "GCTR obj is null");
9293 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9294
9295 // we want to do an instanceof comparison against the AESCrypt class
9296 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9297 if (!klass_AESCrypt->is_loaded()) {
9298 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9299 Node* ctrl = control();
9300 set_control(top()); // no regular fast path
9301 return ctrl;
9302 }
9303
9304 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9305 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9306 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9307 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9308 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9309
9310 return instof_false; // even if it is null
9311 }
9312
9313 //------------------------------get_state_from_digest_object-----------------------
9314 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9315 const char* state_type;
9316 switch (elem_type) {
9317 case T_BYTE: state_type = "[B"; break;
9318 case T_INT: state_type = "[I"; break;
9319 case T_LONG: state_type = "[J"; break;
9320 default: ShouldNotReachHere();
9321 }
9322 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9323 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9324 if (digest_state == nullptr) return (Node *) nullptr;
9325
9326 // now have the array, need to get the start address of the state array
9327 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9328 return state;
9329 }
9330
9331 //------------------------------get_block_size_from_sha3_object----------------------------------
9332 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9333 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9334 assert (block_size != nullptr, "sanity");
9335 return block_size;
9336 }
9337
9338 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9339 // Return node representing slow path of predicate check.
9340 // the pseudo code we want to emulate with this predicate is:
9341 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9342 //
9343 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9344 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9345 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9346 assert((uint)predicate < 5, "sanity");
9347
9348 // The receiver was checked for null already.
9349 Node* digestBaseObj = argument(0);
9350
9351 // get DigestBase klass for instanceOf check
9352 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9353 assert(tinst != nullptr, "digestBaseObj is null");
9354 assert(tinst->is_loaded(), "DigestBase is not loaded");
9355
9356 const char* klass_name = nullptr;
9357 switch (predicate) {
9358 case 0:
9359 if (UseMD5Intrinsics) {
9360 // we want to do an instanceof comparison against the MD5 class
9361 klass_name = "sun/security/provider/MD5";
9362 }
9363 break;
9364 case 1:
9365 if (UseSHA1Intrinsics) {
9366 // we want to do an instanceof comparison against the SHA class
9367 klass_name = "sun/security/provider/SHA";
9368 }
9369 break;
9370 case 2:
9371 if (UseSHA256Intrinsics) {
9372 // we want to do an instanceof comparison against the SHA2 class
9373 klass_name = "sun/security/provider/SHA2";
9374 }
9375 break;
9376 case 3:
9377 if (UseSHA512Intrinsics) {
9378 // we want to do an instanceof comparison against the SHA5 class
9379 klass_name = "sun/security/provider/SHA5";
9380 }
9381 break;
9382 case 4:
9383 if (UseSHA3Intrinsics) {
9384 // we want to do an instanceof comparison against the SHA3 class
9385 klass_name = "sun/security/provider/SHA3";
9386 }
9387 break;
9388 default:
9389 fatal("unknown SHA intrinsic predicate: %d", predicate);
9390 }
9391
9392 ciKlass* klass = nullptr;
9393 if (klass_name != nullptr) {
9394 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9395 }
9396 if ((klass == nullptr) || !klass->is_loaded()) {
9397 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9398 Node* ctrl = control();
9399 set_control(top()); // no intrinsic path
9400 return ctrl;
9401 }
9402 ciInstanceKlass* instklass = klass->as_instance_klass();
9403
9404 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9405 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9406 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9407 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9408
9409 return instof_false; // even if it is null
9410 }
9411
9412 //-------------inline_fma-----------------------------------
9413 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9414 Node *a = nullptr;
9415 Node *b = nullptr;
9416 Node *c = nullptr;
9417 Node* result = nullptr;
9418 switch (id) {
9419 case vmIntrinsics::_fmaD:
9420 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9421 // no receiver since it is static method
9422 a = argument(0);
9423 b = argument(2);
9424 c = argument(4);
9425 result = _gvn.transform(new FmaDNode(a, b, c));
9426 break;
9427 case vmIntrinsics::_fmaF:
9428 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9429 a = argument(0);
9430 b = argument(1);
9431 c = argument(2);
9432 result = _gvn.transform(new FmaFNode(a, b, c));
9433 break;
9434 default:
9435 fatal_unexpected_iid(id); break;
9436 }
9437 set_result(result);
9438 return true;
9439 }
9440
9441 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9442 // argument(0) is receiver
9443 Node* codePoint = argument(1);
9444 Node* n = nullptr;
9445
9446 switch (id) {
9447 case vmIntrinsics::_isDigit :
9448 n = new DigitNode(control(), codePoint);
9449 break;
9450 case vmIntrinsics::_isLowerCase :
9451 n = new LowerCaseNode(control(), codePoint);
9452 break;
9453 case vmIntrinsics::_isUpperCase :
9454 n = new UpperCaseNode(control(), codePoint);
9455 break;
9456 case vmIntrinsics::_isWhitespace :
9457 n = new WhitespaceNode(control(), codePoint);
9458 break;
9459 default:
9460 fatal_unexpected_iid(id);
9461 }
9462
9463 set_result(_gvn.transform(n));
9464 return true;
9465 }
9466
9467 bool LibraryCallKit::inline_profileBoolean() {
9468 Node* counts = argument(1);
9469 const TypeAryPtr* ary = nullptr;
9470 ciArray* aobj = nullptr;
9471 if (counts->is_Con()
9472 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9473 && (aobj = ary->const_oop()->as_array()) != nullptr
9474 && (aobj->length() == 2)) {
9475 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9476 jint false_cnt = aobj->element_value(0).as_int();
9477 jint true_cnt = aobj->element_value(1).as_int();
9478
9479 if (C->log() != nullptr) {
9480 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9481 false_cnt, true_cnt);
9482 }
9483
9484 if (false_cnt + true_cnt == 0) {
9485 // According to profile, never executed.
9486 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9487 Deoptimization::Action_reinterpret);
9488 return true;
9489 }
9490
9491 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9492 // is a number of each value occurrences.
9493 Node* result = argument(0);
9494 if (false_cnt == 0 || true_cnt == 0) {
9495 // According to profile, one value has been never seen.
9496 int expected_val = (false_cnt == 0) ? 1 : 0;
9497
9498 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9499 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9500
9501 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9502 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9503 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9504
9505 { // Slow path: uncommon trap for never seen value and then reexecute
9506 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9507 // the value has been seen at least once.
9508 PreserveJVMState pjvms(this);
9509 PreserveReexecuteState preexecs(this);
9510 jvms()->set_should_reexecute(true);
9511
9512 set_control(slow_path);
9513 set_i_o(i_o());
9514
9515 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9516 Deoptimization::Action_reinterpret);
9517 }
9518 // The guard for never seen value enables sharpening of the result and
9519 // returning a constant. It allows to eliminate branches on the same value
9520 // later on.
9521 set_control(fast_path);
9522 result = intcon(expected_val);
9523 }
9524 // Stop profiling.
9525 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9526 // By replacing method body with profile data (represented as ProfileBooleanNode
9527 // on IR level) we effectively disable profiling.
9528 // It enables full speed execution once optimized code is generated.
9529 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9530 C->record_for_igvn(profile);
9531 set_result(profile);
9532 return true;
9533 } else {
9534 // Continue profiling.
9535 // Profile data isn't available at the moment. So, execute method's bytecode version.
9536 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9537 // is compiled and counters aren't available since corresponding MethodHandle
9538 // isn't a compile-time constant.
9539 return false;
9540 }
9541 }
9542
9543 bool LibraryCallKit::inline_isCompileConstant() {
9544 Node* n = argument(0);
9545 set_result(n->is_Con() ? intcon(1) : intcon(0));
9546 return true;
9547 }
9548
9549 //------------------------------- inline_getObjectSize --------------------------------------
9550 //
9551 // Calculate the runtime size of the object/array.
9552 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9553 //
9554 bool LibraryCallKit::inline_getObjectSize() {
9555 Node* obj = argument(3);
9556 Node* klass_node = load_object_klass(obj);
9557
9558 jint layout_con = Klass::_lh_neutral_value;
9559 Node* layout_val = get_layout_helper(klass_node, layout_con);
9560 int layout_is_con = (layout_val == nullptr);
9561
9562 if (layout_is_con) {
9563 // Layout helper is constant, can figure out things at compile time.
9564
9565 if (Klass::layout_helper_is_instance(layout_con)) {
9566 // Instance case: layout_con contains the size itself.
9567 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9568 set_result(size);
9569 } else {
9570 // Array case: size is round(header + element_size*arraylength).
9571 // Since arraylength is different for every array instance, we have to
9572 // compute the whole thing at runtime.
9573
9574 Node* arr_length = load_array_length(obj);
9575
9576 int round_mask = MinObjAlignmentInBytes - 1;
9577 int hsize = Klass::layout_helper_header_size(layout_con);
9578 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9579
9580 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9581 round_mask = 0; // strength-reduce it if it goes away completely
9582 }
9583 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9584 Node* header_size = intcon(hsize + round_mask);
9585
9586 Node* lengthx = ConvI2X(arr_length);
9587 Node* headerx = ConvI2X(header_size);
9588
9589 Node* abody = lengthx;
9590 if (eshift != 0) {
9591 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9592 }
9593 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9594 if (round_mask != 0) {
9595 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9596 }
9597 size = ConvX2L(size);
9598 set_result(size);
9599 }
9600 } else {
9601 // Layout helper is not constant, need to test for array-ness at runtime.
9602
9603 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9604 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9605 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9606 record_for_igvn(result_reg);
9607
9608 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9609 if (array_ctl != nullptr) {
9610 // Array case: size is round(header + element_size*arraylength).
9611 // Since arraylength is different for every array instance, we have to
9612 // compute the whole thing at runtime.
9613
9614 PreserveJVMState pjvms(this);
9615 set_control(array_ctl);
9616 Node* arr_length = load_array_length(obj);
9617
9618 int round_mask = MinObjAlignmentInBytes - 1;
9619 Node* mask = intcon(round_mask);
9620
9621 Node* hss = intcon(Klass::_lh_header_size_shift);
9622 Node* hsm = intcon(Klass::_lh_header_size_mask);
9623 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9624 header_size = _gvn.transform(new AndINode(header_size, hsm));
9625 header_size = _gvn.transform(new AddINode(header_size, mask));
9626
9627 // There is no need to mask or shift this value.
9628 // The semantics of LShiftINode include an implicit mask to 0x1F.
9629 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9630 Node* elem_shift = layout_val;
9631
9632 Node* lengthx = ConvI2X(arr_length);
9633 Node* headerx = ConvI2X(header_size);
9634
9635 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9636 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9637 if (round_mask != 0) {
9638 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9639 }
9640 size = ConvX2L(size);
9641
9642 result_reg->init_req(_array_path, control());
9643 result_val->init_req(_array_path, size);
9644 }
9645
9646 if (!stopped()) {
9647 // Instance case: the layout helper gives us instance size almost directly,
9648 // but we need to mask out the _lh_instance_slow_path_bit.
9649 Node* size = ConvI2X(layout_val);
9650 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9651 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9652 size = _gvn.transform(new AndXNode(size, mask));
9653 size = ConvX2L(size);
9654
9655 result_reg->init_req(_instance_path, control());
9656 result_val->init_req(_instance_path, size);
9657 }
9658
9659 set_result(result_reg, result_val);
9660 }
9661
9662 return true;
9663 }
9664
9665 //------------------------------- inline_blackhole --------------------------------------
9666 //
9667 // Make sure all arguments to this node are alive.
9668 // This matches methods that were requested to be blackholed through compile commands.
9669 //
9670 bool LibraryCallKit::inline_blackhole() {
9671 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9672 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9673 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9674
9675 // Blackhole node pinches only the control, not memory. This allows
9676 // the blackhole to be pinned in the loop that computes blackholed
9677 // values, but have no other side effects, like breaking the optimizations
9678 // across the blackhole.
9679
9680 Node* bh = _gvn.transform(new BlackholeNode(control()));
9681 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9682
9683 // Bind call arguments as blackhole arguments to keep them alive
9684 uint nargs = callee()->arg_size();
9685 for (uint i = 0; i < nargs; i++) {
9686 bh->add_req(argument(i));
9687 }
9688
9689 return true;
9690 }
9691
9692 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9693 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9694 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9695 return nullptr; // box klass is not Float16
9696 }
9697
9698 // Null check; get notnull casted pointer
9699 Node* null_ctl = top();
9700 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9701 // If not_null_box is dead, only null-path is taken
9702 if (stopped()) {
9703 set_control(null_ctl);
9704 return nullptr;
9705 }
9706 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9707 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9708 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9709 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9710 }
9711
9712 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9713 PreserveReexecuteState preexecs(this);
9714 jvms()->set_should_reexecute(true);
9715
9716 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9717 Node* klass_node = makecon(klass_type);
9718 Node* box = new_instance(klass_node);
9719
9720 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9721 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9722
9723 Node* field_store = _gvn.transform(access_store_at(box,
9724 value_field,
9725 value_adr_type,
9726 value,
9727 TypeInt::SHORT,
9728 T_SHORT,
9729 IN_HEAP));
9730 set_memory(field_store, value_adr_type);
9731 return box;
9732 }
9733
9734 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9735 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9736 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9737 return false;
9738 }
9739
9740 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9741 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9742 return false;
9743 }
9744
9745 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9746 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9747 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9748 ciSymbols::short_signature(),
9749 false);
9750 assert(field != nullptr, "");
9751
9752 // Transformed nodes
9753 Node* fld1 = nullptr;
9754 Node* fld2 = nullptr;
9755 Node* fld3 = nullptr;
9756 switch(num_args) {
9757 case 3:
9758 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9759 if (fld3 == nullptr) {
9760 return false;
9761 }
9762 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9763 // fall-through
9764 case 2:
9765 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9766 if (fld2 == nullptr) {
9767 return false;
9768 }
9769 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9770 // fall-through
9771 case 1:
9772 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9773 if (fld1 == nullptr) {
9774 return false;
9775 }
9776 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9777 break;
9778 default: fatal("Unsupported number of arguments %d", num_args);
9779 }
9780
9781 Node* result = nullptr;
9782 switch (id) {
9783 // Unary operations
9784 case vmIntrinsics::_sqrt_float16:
9785 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9786 break;
9787 // Ternary operations
9788 case vmIntrinsics::_fma_float16:
9789 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9790 break;
9791 default:
9792 fatal_unexpected_iid(id);
9793 break;
9794 }
9795 result = _gvn.transform(new ReinterpretHF2SNode(result));
9796 set_result(box_fp16_value(float16_box_type, field, result));
9797 return true;
9798 }
9799