1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciUtilities.inline.hpp"
27 #include "ci/ciSymbols.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
125 if (C->log()) {
126 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
127 vmIntrinsics::name_at(intrinsic_id()),
128 (is_virtual() ? " virtual='1'" : ""),
129 C->unique() - nodes);
130 }
131 // Push the result from the inlined method onto the stack.
132 kit.push_result();
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 if (jvms->has_method()) {
139 // Not a root compile.
140 const char* msg;
141 if (callee->intrinsic_candidate()) {
142 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
143 } else {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
145 : "failed to inline (intrinsic), method not annotated";
146 }
147 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
148 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
149 } else {
150 // Root compile
151 ResourceMark rm;
152 stringStream msg_stream;
153 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
154 vmIntrinsics::name_at(intrinsic_id()),
155 is_virtual() ? " (virtual)" : "", bci);
156 const char *msg = msg_stream.freeze();
157 log_debug(jit, inlining)("%s", msg);
158 if (C->print_intrinsics() || C->print_inlining()) {
159 tty->print("%s", msg);
160 }
161 }
162 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
163
164 return nullptr;
165 }
166
167 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
168 LibraryCallKit kit(jvms, this);
169 Compile* C = kit.C;
170 int nodes = C->unique();
171 _last_predicate = predicate;
172 #ifndef PRODUCT
173 assert(is_predicated() && predicate < predicates_count(), "sanity");
174 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
175 char buf[1000];
176 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
177 tty->print_cr("Predicate for intrinsic %s", str);
178 }
179 #endif
180 ciMethod* callee = kit.callee();
181 const int bci = kit.bci();
182
183 Node* slow_ctl = kit.try_to_predicate(predicate);
184 if (!kit.failing()) {
185 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
186 : "(intrinsic, predicate)";
187 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
188 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
189
190 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
191 if (C->log()) {
192 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
193 vmIntrinsics::name_at(intrinsic_id()),
194 (is_virtual() ? " virtual='1'" : ""),
195 C->unique() - nodes);
196 }
197 return slow_ctl; // Could be null if the check folds.
198 }
199
200 // The intrinsic bailed out
201 if (jvms->has_method()) {
202 // Not a root compile.
203 const char* msg = "failed to generate predicate for intrinsic";
204 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
205 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
206 } else {
207 // Root compile
208 ResourceMark rm;
209 stringStream msg_stream;
210 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
211 vmIntrinsics::name_at(intrinsic_id()),
212 is_virtual() ? " (virtual)" : "", bci);
213 const char *msg = msg_stream.freeze();
214 log_debug(jit, inlining)("%s", msg);
215 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
216 }
217 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
218 return nullptr;
219 }
220
221 bool LibraryCallKit::try_to_inline(int predicate) {
222 // Handle symbolic names for otherwise undistinguished boolean switches:
223 const bool is_store = true;
224 const bool is_compress = true;
225 const bool is_static = true;
226 const bool is_volatile = true;
227
228 if (!jvms()->has_method()) {
229 // Root JVMState has a null method.
230 assert(map()->memory()->Opcode() == Op_Parm, "");
231 // Insert the memory aliasing node
232 set_all_memory(reset_memory());
233 }
234 assert(merged_memory(), "");
235
236 switch (intrinsic_id()) {
237 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
238 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
239 case vmIntrinsics::_getClass: return inline_native_getClass();
240
241 case vmIntrinsics::_ceil:
242 case vmIntrinsics::_floor:
243 case vmIntrinsics::_rint:
244 case vmIntrinsics::_dsin:
245 case vmIntrinsics::_dcos:
246 case vmIntrinsics::_dtan:
247 case vmIntrinsics::_dtanh:
248 case vmIntrinsics::_dabs:
249 case vmIntrinsics::_fabs:
250 case vmIntrinsics::_iabs:
251 case vmIntrinsics::_labs:
252 case vmIntrinsics::_datan2:
253 case vmIntrinsics::_dsqrt:
254 case vmIntrinsics::_dsqrt_strict:
255 case vmIntrinsics::_dexp:
256 case vmIntrinsics::_dlog:
257 case vmIntrinsics::_dlog10:
258 case vmIntrinsics::_dpow:
259 case vmIntrinsics::_dcopySign:
260 case vmIntrinsics::_fcopySign:
261 case vmIntrinsics::_dsignum:
262 case vmIntrinsics::_roundF:
263 case vmIntrinsics::_roundD:
264 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
265
266 case vmIntrinsics::_notify:
267 case vmIntrinsics::_notifyAll:
268 return inline_notify(intrinsic_id());
269
270 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
271 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
272 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
273 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
274 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
275 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
276 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
277 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
278 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
279 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
280 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
281 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
282 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
283 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
284
285 case vmIntrinsics::_arraycopy: return inline_arraycopy();
286
287 case vmIntrinsics::_arraySort: return inline_array_sort();
288 case vmIntrinsics::_arrayPartition: return inline_array_partition();
289
290 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
291 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
292 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
293 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
294
295 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
296 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
297 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
298 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
302 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
303
304 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
305
306 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
307
308 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
309 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
310 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
311 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
312
313 case vmIntrinsics::_compressStringC:
314 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
315 case vmIntrinsics::_inflateStringC:
316 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
317
318 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
319 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
320 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
321 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
322 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
323 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
324 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
325 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
326 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
327
328 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
339 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
340 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
341 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
342 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
343 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
344 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
345 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
346 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
347
348 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
359 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
360 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
361 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
362
363 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
369 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
370 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
371 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
372 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
373 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
374 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
375 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
376 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
377
378 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
379 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
380 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
381 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
382 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
383 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
384 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
385 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
386 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
387
388 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
389 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
390 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
391 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
392 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
393 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
394 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
395 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
396 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
397
398 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
409 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
413
414 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
415 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
416 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
417 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
418 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
434
435 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
436 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
437 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
438 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
439 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
440 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
441 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
450
451 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
452 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
455
456 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
457 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
461
462 case vmIntrinsics::_loadFence:
463 case vmIntrinsics::_storeFence:
464 case vmIntrinsics::_storeStoreFence:
465 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
466
467 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
468
469 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
470 case vmIntrinsics::_currentThread: return inline_native_currentThread();
471 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
472
473 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
474 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
475
476 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
477 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
478
479 #if INCLUDE_JVMTI
480 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
481 "notifyJvmtiStart", true, false);
482 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
483 "notifyJvmtiEnd", false, true);
484 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
485 "notifyJvmtiMount", false, false);
486 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
487 "notifyJvmtiUnmount", false, false);
488 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
489 #endif
490
491 #ifdef JFR_HAVE_INTRINSICS
492 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
493 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
494 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
495 #endif
496 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
497 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
498 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
499 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
500 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
501 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
502 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
503 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
504 case vmIntrinsics::_getLength: return inline_native_getLength();
505 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
506 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
507 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
508 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
509 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
510 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
511 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
512
513 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
514 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
515
516 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
517
518 case vmIntrinsics::_isInstance:
519 case vmIntrinsics::_isHidden:
520 case vmIntrinsics::_getSuperclass:
521 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
522
523 case vmIntrinsics::_floatToRawIntBits:
524 case vmIntrinsics::_floatToIntBits:
525 case vmIntrinsics::_intBitsToFloat:
526 case vmIntrinsics::_doubleToRawLongBits:
527 case vmIntrinsics::_doubleToLongBits:
528 case vmIntrinsics::_longBitsToDouble:
529 case vmIntrinsics::_floatToFloat16:
530 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
531 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
532 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
533 case vmIntrinsics::_floatIsFinite:
534 case vmIntrinsics::_floatIsInfinite:
535 case vmIntrinsics::_doubleIsFinite:
536 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
537
538 case vmIntrinsics::_numberOfLeadingZeros_i:
539 case vmIntrinsics::_numberOfLeadingZeros_l:
540 case vmIntrinsics::_numberOfTrailingZeros_i:
541 case vmIntrinsics::_numberOfTrailingZeros_l:
542 case vmIntrinsics::_bitCount_i:
543 case vmIntrinsics::_bitCount_l:
544 case vmIntrinsics::_reverse_i:
545 case vmIntrinsics::_reverse_l:
546 case vmIntrinsics::_reverseBytes_i:
547 case vmIntrinsics::_reverseBytes_l:
548 case vmIntrinsics::_reverseBytes_s:
549 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
550
551 case vmIntrinsics::_compress_i:
552 case vmIntrinsics::_compress_l:
553 case vmIntrinsics::_expand_i:
554 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
555
556 case vmIntrinsics::_compareUnsigned_i:
557 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
558
559 case vmIntrinsics::_divideUnsigned_i:
560 case vmIntrinsics::_divideUnsigned_l:
561 case vmIntrinsics::_remainderUnsigned_i:
562 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
563
564 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
565
566 case vmIntrinsics::_Reference_get: return inline_reference_get();
567 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
568 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
569 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
570 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
571
572 case vmIntrinsics::_Class_cast: return inline_Class_cast();
573
574 case vmIntrinsics::_aescrypt_encryptBlock:
575 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
576
577 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
578 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
579 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
580
581 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
582 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
583 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_counterMode_AESCrypt:
586 return inline_counterMode_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_galoisCounterMode_AESCrypt:
589 return inline_galoisCounterMode_AESCrypt();
590
591 case vmIntrinsics::_md5_implCompress:
592 case vmIntrinsics::_sha_implCompress:
593 case vmIntrinsics::_sha2_implCompress:
594 case vmIntrinsics::_sha5_implCompress:
595 case vmIntrinsics::_sha3_implCompress:
596 return inline_digestBase_implCompress(intrinsic_id());
597 case vmIntrinsics::_double_keccak:
598 return inline_double_keccak();
599
600 case vmIntrinsics::_digestBase_implCompressMB:
601 return inline_digestBase_implCompressMB(predicate);
602
603 case vmIntrinsics::_multiplyToLen:
604 return inline_multiplyToLen();
605
606 case vmIntrinsics::_squareToLen:
607 return inline_squareToLen();
608
609 case vmIntrinsics::_mulAdd:
610 return inline_mulAdd();
611
612 case vmIntrinsics::_montgomeryMultiply:
613 return inline_montgomeryMultiply();
614 case vmIntrinsics::_montgomerySquare:
615 return inline_montgomerySquare();
616
617 case vmIntrinsics::_bigIntegerRightShiftWorker:
618 return inline_bigIntegerShift(true);
619 case vmIntrinsics::_bigIntegerLeftShiftWorker:
620 return inline_bigIntegerShift(false);
621
622 case vmIntrinsics::_vectorizedMismatch:
623 return inline_vectorizedMismatch();
624
625 case vmIntrinsics::_ghash_processBlocks:
626 return inline_ghash_processBlocks();
627 case vmIntrinsics::_chacha20Block:
628 return inline_chacha20Block();
629 case vmIntrinsics::_kyberNtt:
630 return inline_kyberNtt();
631 case vmIntrinsics::_kyberInverseNtt:
632 return inline_kyberInverseNtt();
633 case vmIntrinsics::_kyberNttMult:
634 return inline_kyberNttMult();
635 case vmIntrinsics::_kyberAddPoly_2:
636 return inline_kyberAddPoly_2();
637 case vmIntrinsics::_kyberAddPoly_3:
638 return inline_kyberAddPoly_3();
639 case vmIntrinsics::_kyber12To16:
640 return inline_kyber12To16();
641 case vmIntrinsics::_kyberBarrettReduce:
642 return inline_kyberBarrettReduce();
643 case vmIntrinsics::_dilithiumAlmostNtt:
644 return inline_dilithiumAlmostNtt();
645 case vmIntrinsics::_dilithiumAlmostInverseNtt:
646 return inline_dilithiumAlmostInverseNtt();
647 case vmIntrinsics::_dilithiumNttMult:
648 return inline_dilithiumNttMult();
649 case vmIntrinsics::_dilithiumMontMulByConstant:
650 return inline_dilithiumMontMulByConstant();
651 case vmIntrinsics::_dilithiumDecomposePoly:
652 return inline_dilithiumDecomposePoly();
653 case vmIntrinsics::_base64_encodeBlock:
654 return inline_base64_encodeBlock();
655 case vmIntrinsics::_base64_decodeBlock:
656 return inline_base64_decodeBlock();
657 case vmIntrinsics::_poly1305_processBlocks:
658 return inline_poly1305_processBlocks();
659 case vmIntrinsics::_intpoly_montgomeryMult_P256:
660 return inline_intpoly_montgomeryMult_P256();
661 case vmIntrinsics::_intpoly_assign:
662 return inline_intpoly_assign();
663 case vmIntrinsics::_encodeISOArray:
664 case vmIntrinsics::_encodeByteISOArray:
665 return inline_encodeISOArray(false);
666 case vmIntrinsics::_encodeAsciiArray:
667 return inline_encodeISOArray(true);
668
669 case vmIntrinsics::_updateCRC32:
670 return inline_updateCRC32();
671 case vmIntrinsics::_updateBytesCRC32:
672 return inline_updateBytesCRC32();
673 case vmIntrinsics::_updateByteBufferCRC32:
674 return inline_updateByteBufferCRC32();
675
676 case vmIntrinsics::_updateBytesCRC32C:
677 return inline_updateBytesCRC32C();
678 case vmIntrinsics::_updateDirectByteBufferCRC32C:
679 return inline_updateDirectByteBufferCRC32C();
680
681 case vmIntrinsics::_updateBytesAdler32:
682 return inline_updateBytesAdler32();
683 case vmIntrinsics::_updateByteBufferAdler32:
684 return inline_updateByteBufferAdler32();
685
686 case vmIntrinsics::_profileBoolean:
687 return inline_profileBoolean();
688 case vmIntrinsics::_isCompileConstant:
689 return inline_isCompileConstant();
690
691 case vmIntrinsics::_countPositives:
692 return inline_countPositives();
693
694 case vmIntrinsics::_fmaD:
695 case vmIntrinsics::_fmaF:
696 return inline_fma(intrinsic_id());
697
698 case vmIntrinsics::_isDigit:
699 case vmIntrinsics::_isLowerCase:
700 case vmIntrinsics::_isUpperCase:
701 case vmIntrinsics::_isWhitespace:
702 return inline_character_compare(intrinsic_id());
703
704 case vmIntrinsics::_min:
705 case vmIntrinsics::_max:
706 case vmIntrinsics::_min_strict:
707 case vmIntrinsics::_max_strict:
708 case vmIntrinsics::_minL:
709 case vmIntrinsics::_maxL:
710 case vmIntrinsics::_minF:
711 case vmIntrinsics::_maxF:
712 case vmIntrinsics::_minD:
713 case vmIntrinsics::_maxD:
714 case vmIntrinsics::_minF_strict:
715 case vmIntrinsics::_maxF_strict:
716 case vmIntrinsics::_minD_strict:
717 case vmIntrinsics::_maxD_strict:
718 return inline_min_max(intrinsic_id());
719
720 case vmIntrinsics::_VectorUnaryOp:
721 return inline_vector_nary_operation(1);
722 case vmIntrinsics::_VectorBinaryOp:
723 return inline_vector_nary_operation(2);
724 case vmIntrinsics::_VectorTernaryOp:
725 return inline_vector_nary_operation(3);
726 case vmIntrinsics::_VectorFromBitsCoerced:
727 return inline_vector_frombits_coerced();
728 case vmIntrinsics::_VectorMaskOp:
729 return inline_vector_mask_operation();
730 case vmIntrinsics::_VectorLoadOp:
731 return inline_vector_mem_operation(/*is_store=*/false);
732 case vmIntrinsics::_VectorLoadMaskedOp:
733 return inline_vector_mem_masked_operation(/*is_store*/false);
734 case vmIntrinsics::_VectorStoreOp:
735 return inline_vector_mem_operation(/*is_store=*/true);
736 case vmIntrinsics::_VectorStoreMaskedOp:
737 return inline_vector_mem_masked_operation(/*is_store=*/true);
738 case vmIntrinsics::_VectorGatherOp:
739 return inline_vector_gather_scatter(/*is_scatter*/ false);
740 case vmIntrinsics::_VectorScatterOp:
741 return inline_vector_gather_scatter(/*is_scatter*/ true);
742 case vmIntrinsics::_VectorReductionCoerced:
743 return inline_vector_reduction();
744 case vmIntrinsics::_VectorTest:
745 return inline_vector_test();
746 case vmIntrinsics::_VectorBlend:
747 return inline_vector_blend();
748 case vmIntrinsics::_VectorRearrange:
749 return inline_vector_rearrange();
750 case vmIntrinsics::_VectorSelectFrom:
751 return inline_vector_select_from();
752 case vmIntrinsics::_VectorCompare:
753 return inline_vector_compare();
754 case vmIntrinsics::_VectorBroadcastInt:
755 return inline_vector_broadcast_int();
756 case vmIntrinsics::_VectorConvert:
757 return inline_vector_convert();
758 case vmIntrinsics::_VectorInsert:
759 return inline_vector_insert();
760 case vmIntrinsics::_VectorExtract:
761 return inline_vector_extract();
762 case vmIntrinsics::_VectorCompressExpand:
763 return inline_vector_compress_expand();
764 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
765 return inline_vector_select_from_two_vectors();
766 case vmIntrinsics::_IndexVector:
767 return inline_index_vector();
768 case vmIntrinsics::_IndexPartiallyInUpperRange:
769 return inline_index_partially_in_upper_range();
770
771 case vmIntrinsics::_getObjectSize:
772 return inline_getObjectSize();
773
774 case vmIntrinsics::_blackhole:
775 return inline_blackhole();
776
777 default:
778 // If you get here, it may be that someone has added a new intrinsic
779 // to the list in vmIntrinsics.hpp without implementing it here.
780 #ifndef PRODUCT
781 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
782 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
783 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
784 }
785 #endif
786 return false;
787 }
788 }
789
790 Node* LibraryCallKit::try_to_predicate(int predicate) {
791 if (!jvms()->has_method()) {
792 // Root JVMState has a null method.
793 assert(map()->memory()->Opcode() == Op_Parm, "");
794 // Insert the memory aliasing node
795 set_all_memory(reset_memory());
796 }
797 assert(merged_memory(), "");
798
799 switch (intrinsic_id()) {
800 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
801 return inline_cipherBlockChaining_AESCrypt_predicate(false);
802 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
803 return inline_cipherBlockChaining_AESCrypt_predicate(true);
804 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
805 return inline_electronicCodeBook_AESCrypt_predicate(false);
806 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
807 return inline_electronicCodeBook_AESCrypt_predicate(true);
808 case vmIntrinsics::_counterMode_AESCrypt:
809 return inline_counterMode_AESCrypt_predicate();
810 case vmIntrinsics::_digestBase_implCompressMB:
811 return inline_digestBase_implCompressMB_predicate(predicate);
812 case vmIntrinsics::_galoisCounterMode_AESCrypt:
813 return inline_galoisCounterMode_AESCrypt_predicate();
814
815 default:
816 // If you get here, it may be that someone has added a new intrinsic
817 // to the list in vmIntrinsics.hpp without implementing it here.
818 #ifndef PRODUCT
819 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
820 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
821 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
822 }
823 #endif
824 Node* slow_ctl = control();
825 set_control(top()); // No fast path intrinsic
826 return slow_ctl;
827 }
828 }
829
830 //------------------------------set_result-------------------------------
831 // Helper function for finishing intrinsics.
832 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
833 record_for_igvn(region);
834 set_control(_gvn.transform(region));
835 set_result( _gvn.transform(value));
836 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
837 }
838
839 //------------------------------generate_guard---------------------------
840 // Helper function for generating guarded fast-slow graph structures.
841 // The given 'test', if true, guards a slow path. If the test fails
842 // then a fast path can be taken. (We generally hope it fails.)
843 // In all cases, GraphKit::control() is updated to the fast path.
844 // The returned value represents the control for the slow path.
845 // The return value is never 'top'; it is either a valid control
846 // or null if it is obvious that the slow path can never be taken.
847 // Also, if region and the slow control are not null, the slow edge
848 // is appended to the region.
849 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
850 if (stopped()) {
851 // Already short circuited.
852 return nullptr;
853 }
854
855 // Build an if node and its projections.
856 // If test is true we take the slow path, which we assume is uncommon.
857 if (_gvn.type(test) == TypeInt::ZERO) {
858 // The slow branch is never taken. No need to build this guard.
859 return nullptr;
860 }
861
862 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
863
864 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
865 if (if_slow == top()) {
866 // The slow branch is never taken. No need to build this guard.
867 return nullptr;
868 }
869
870 if (region != nullptr)
871 region->add_req(if_slow);
872
873 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
874 set_control(if_fast);
875
876 return if_slow;
877 }
878
879 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
880 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
881 }
882 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
883 return generate_guard(test, region, PROB_FAIR);
884 }
885
886 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
887 Node* *pos_index) {
888 if (stopped())
889 return nullptr; // already stopped
890 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
891 return nullptr; // index is already adequately typed
892 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
893 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
894 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
895 if (is_neg != nullptr && pos_index != nullptr) {
896 // Emulate effect of Parse::adjust_map_after_if.
897 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
898 (*pos_index) = _gvn.transform(ccast);
899 }
900 return is_neg;
901 }
902
903 // Make sure that 'position' is a valid limit index, in [0..length].
904 // There are two equivalent plans for checking this:
905 // A. (offset + copyLength) unsigned<= arrayLength
906 // B. offset <= (arrayLength - copyLength)
907 // We require that all of the values above, except for the sum and
908 // difference, are already known to be non-negative.
909 // Plan A is robust in the face of overflow, if offset and copyLength
910 // are both hugely positive.
911 //
912 // Plan B is less direct and intuitive, but it does not overflow at
913 // all, since the difference of two non-negatives is always
914 // representable. Whenever Java methods must perform the equivalent
915 // check they generally use Plan B instead of Plan A.
916 // For the moment we use Plan A.
917 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
918 Node* subseq_length,
919 Node* array_length,
920 RegionNode* region) {
921 if (stopped())
922 return nullptr; // already stopped
923 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
924 if (zero_offset && subseq_length->eqv_uncast(array_length))
925 return nullptr; // common case of whole-array copy
926 Node* last = subseq_length;
927 if (!zero_offset) // last += offset
928 last = _gvn.transform(new AddINode(last, offset));
929 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
930 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
931 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
932 return is_over;
933 }
934
935 // Emit range checks for the given String.value byte array
936 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
937 if (stopped()) {
938 return; // already stopped
939 }
940 RegionNode* bailout = new RegionNode(1);
941 record_for_igvn(bailout);
942 if (char_count) {
943 // Convert char count to byte count
944 count = _gvn.transform(new LShiftINode(count, intcon(1)));
945 }
946
947 // Offset and count must not be negative
948 generate_negative_guard(offset, bailout);
949 generate_negative_guard(count, bailout);
950 // Offset + count must not exceed length of array
951 generate_limit_guard(offset, count, load_array_length(array), bailout);
952
953 if (bailout->req() > 1) {
954 PreserveJVMState pjvms(this);
955 set_control(_gvn.transform(bailout));
956 uncommon_trap(Deoptimization::Reason_intrinsic,
957 Deoptimization::Action_maybe_recompile);
958 }
959 }
960
961 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
962 bool is_immutable) {
963 ciKlass* thread_klass = env()->Thread_klass();
964 const Type* thread_type
965 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
966
967 Node* thread = _gvn.transform(new ThreadLocalNode());
968 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
969 tls_output = thread;
970
971 Node* thread_obj_handle
972 = (is_immutable
973 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
974 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
975 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
976 thread_obj_handle = _gvn.transform(thread_obj_handle);
977
978 DecoratorSet decorators = IN_NATIVE;
979 if (is_immutable) {
980 decorators |= C2_IMMUTABLE_MEMORY;
981 }
982 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
983 }
984
985 //--------------------------generate_current_thread--------------------
986 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
987 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
988 /*is_immutable*/false);
989 }
990
991 //--------------------------generate_virtual_thread--------------------
992 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
993 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
994 !C->method()->changes_current_thread());
995 }
996
997 //------------------------------make_string_method_node------------------------
998 // Helper method for String intrinsic functions. This version is called with
999 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1000 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1001 // containing the lengths of str1 and str2.
1002 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1003 Node* result = nullptr;
1004 switch (opcode) {
1005 case Op_StrIndexOf:
1006 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1007 str1_start, cnt1, str2_start, cnt2, ae);
1008 break;
1009 case Op_StrComp:
1010 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1011 str1_start, cnt1, str2_start, cnt2, ae);
1012 break;
1013 case Op_StrEquals:
1014 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1015 // Use the constant length if there is one because optimized match rule may exist.
1016 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1017 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1018 break;
1019 default:
1020 ShouldNotReachHere();
1021 return nullptr;
1022 }
1023
1024 // All these intrinsics have checks.
1025 C->set_has_split_ifs(true); // Has chance for split-if optimization
1026 clear_upper_avx();
1027
1028 return _gvn.transform(result);
1029 }
1030
1031 //------------------------------inline_string_compareTo------------------------
1032 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1033 Node* arg1 = argument(0);
1034 Node* arg2 = argument(1);
1035
1036 arg1 = must_be_not_null(arg1, true);
1037 arg2 = must_be_not_null(arg2, true);
1038
1039 // Get start addr and length of first argument
1040 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1041 Node* arg1_cnt = load_array_length(arg1);
1042
1043 // Get start addr and length of second argument
1044 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1045 Node* arg2_cnt = load_array_length(arg2);
1046
1047 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1048 set_result(result);
1049 return true;
1050 }
1051
1052 //------------------------------inline_string_equals------------------------
1053 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1054 Node* arg1 = argument(0);
1055 Node* arg2 = argument(1);
1056
1057 // paths (plus control) merge
1058 RegionNode* region = new RegionNode(3);
1059 Node* phi = new PhiNode(region, TypeInt::BOOL);
1060
1061 if (!stopped()) {
1062
1063 arg1 = must_be_not_null(arg1, true);
1064 arg2 = must_be_not_null(arg2, true);
1065
1066 // Get start addr and length of first argument
1067 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1068 Node* arg1_cnt = load_array_length(arg1);
1069
1070 // Get start addr and length of second argument
1071 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1072 Node* arg2_cnt = load_array_length(arg2);
1073
1074 // Check for arg1_cnt != arg2_cnt
1075 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1076 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1077 Node* if_ne = generate_slow_guard(bol, nullptr);
1078 if (if_ne != nullptr) {
1079 phi->init_req(2, intcon(0));
1080 region->init_req(2, if_ne);
1081 }
1082
1083 // Check for count == 0 is done by assembler code for StrEquals.
1084
1085 if (!stopped()) {
1086 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1087 phi->init_req(1, equals);
1088 region->init_req(1, control());
1089 }
1090 }
1091
1092 // post merge
1093 set_control(_gvn.transform(region));
1094 record_for_igvn(region);
1095
1096 set_result(_gvn.transform(phi));
1097 return true;
1098 }
1099
1100 //------------------------------inline_array_equals----------------------------
1101 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1102 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1103 Node* arg1 = argument(0);
1104 Node* arg2 = argument(1);
1105
1106 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1107 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1108 clear_upper_avx();
1109
1110 return true;
1111 }
1112
1113
1114 //------------------------------inline_countPositives------------------------------
1115 bool LibraryCallKit::inline_countPositives() {
1116 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1117 return false;
1118 }
1119
1120 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1121 // no receiver since it is static method
1122 Node* ba = argument(0);
1123 Node* offset = argument(1);
1124 Node* len = argument(2);
1125
1126 ba = must_be_not_null(ba, true);
1127
1128 // Range checks
1129 generate_string_range_check(ba, offset, len, false);
1130 if (stopped()) {
1131 return true;
1132 }
1133 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1134 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1135 set_result(_gvn.transform(result));
1136 clear_upper_avx();
1137 return true;
1138 }
1139
1140 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1141 Node* index = argument(0);
1142 Node* length = bt == T_INT ? argument(1) : argument(2);
1143 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1144 return false;
1145 }
1146
1147 // check that length is positive
1148 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1149 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1150
1151 {
1152 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1153 uncommon_trap(Deoptimization::Reason_intrinsic,
1154 Deoptimization::Action_make_not_entrant);
1155 }
1156
1157 if (stopped()) {
1158 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1159 return true;
1160 }
1161
1162 // length is now known positive, add a cast node to make this explicit
1163 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1164 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1165 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1166 ConstraintCastNode::RegularDependency, bt);
1167 casted_length = _gvn.transform(casted_length);
1168 replace_in_map(length, casted_length);
1169 length = casted_length;
1170
1171 // Use an unsigned comparison for the range check itself
1172 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1173 BoolTest::mask btest = BoolTest::lt;
1174 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1175 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1176 _gvn.set_type(rc, rc->Value(&_gvn));
1177 if (!rc_bool->is_Con()) {
1178 record_for_igvn(rc);
1179 }
1180 set_control(_gvn.transform(new IfTrueNode(rc)));
1181 {
1182 PreserveJVMState pjvms(this);
1183 set_control(_gvn.transform(new IfFalseNode(rc)));
1184 uncommon_trap(Deoptimization::Reason_range_check,
1185 Deoptimization::Action_make_not_entrant);
1186 }
1187
1188 if (stopped()) {
1189 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1190 return true;
1191 }
1192
1193 // index is now known to be >= 0 and < length, cast it
1194 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1195 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1196 ConstraintCastNode::RegularDependency, bt);
1197 result = _gvn.transform(result);
1198 set_result(result);
1199 replace_in_map(index, result);
1200 return true;
1201 }
1202
1203 //------------------------------inline_string_indexOf------------------------
1204 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1205 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1206 return false;
1207 }
1208 Node* src = argument(0);
1209 Node* tgt = argument(1);
1210
1211 // Make the merge point
1212 RegionNode* result_rgn = new RegionNode(4);
1213 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1214
1215 src = must_be_not_null(src, true);
1216 tgt = must_be_not_null(tgt, true);
1217
1218 // Get start addr and length of source string
1219 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1220 Node* src_count = load_array_length(src);
1221
1222 // Get start addr and length of substring
1223 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1224 Node* tgt_count = load_array_length(tgt);
1225
1226 Node* result = nullptr;
1227 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1228
1229 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1230 // Divide src size by 2 if String is UTF16 encoded
1231 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1232 }
1233 if (ae == StrIntrinsicNode::UU) {
1234 // Divide substring size by 2 if String is UTF16 encoded
1235 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1236 }
1237
1238 if (call_opt_stub) {
1239 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1240 StubRoutines::_string_indexof_array[ae],
1241 "stringIndexOf", TypePtr::BOTTOM, src_start,
1242 src_count, tgt_start, tgt_count);
1243 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1244 } else {
1245 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1246 result_rgn, result_phi, ae);
1247 }
1248 if (result != nullptr) {
1249 result_phi->init_req(3, result);
1250 result_rgn->init_req(3, control());
1251 }
1252 set_control(_gvn.transform(result_rgn));
1253 record_for_igvn(result_rgn);
1254 set_result(_gvn.transform(result_phi));
1255
1256 return true;
1257 }
1258
1259 //-----------------------------inline_string_indexOfI-----------------------
1260 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1261 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1262 return false;
1263 }
1264 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1265 return false;
1266 }
1267
1268 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1269 Node* src = argument(0); // byte[]
1270 Node* src_count = argument(1); // char count
1271 Node* tgt = argument(2); // byte[]
1272 Node* tgt_count = argument(3); // char count
1273 Node* from_index = argument(4); // char index
1274
1275 src = must_be_not_null(src, true);
1276 tgt = must_be_not_null(tgt, true);
1277
1278 // Multiply byte array index by 2 if String is UTF16 encoded
1279 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1280 src_count = _gvn.transform(new SubINode(src_count, from_index));
1281 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1282 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1283
1284 // Range checks
1285 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1286 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1287 if (stopped()) {
1288 return true;
1289 }
1290
1291 RegionNode* region = new RegionNode(5);
1292 Node* phi = new PhiNode(region, TypeInt::INT);
1293 Node* result = nullptr;
1294
1295 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1296
1297 if (call_opt_stub) {
1298 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1299 StubRoutines::_string_indexof_array[ae],
1300 "stringIndexOf", TypePtr::BOTTOM, src_start,
1301 src_count, tgt_start, tgt_count);
1302 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1303 } else {
1304 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1305 region, phi, ae);
1306 }
1307 if (result != nullptr) {
1308 // The result is index relative to from_index if substring was found, -1 otherwise.
1309 // Generate code which will fold into cmove.
1310 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1311 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1312
1313 Node* if_lt = generate_slow_guard(bol, nullptr);
1314 if (if_lt != nullptr) {
1315 // result == -1
1316 phi->init_req(3, result);
1317 region->init_req(3, if_lt);
1318 }
1319 if (!stopped()) {
1320 result = _gvn.transform(new AddINode(result, from_index));
1321 phi->init_req(4, result);
1322 region->init_req(4, control());
1323 }
1324 }
1325
1326 set_control(_gvn.transform(region));
1327 record_for_igvn(region);
1328 set_result(_gvn.transform(phi));
1329 clear_upper_avx();
1330
1331 return true;
1332 }
1333
1334 // Create StrIndexOfNode with fast path checks
1335 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1336 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1337 // Check for substr count > string count
1338 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1339 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1340 Node* if_gt = generate_slow_guard(bol, nullptr);
1341 if (if_gt != nullptr) {
1342 phi->init_req(1, intcon(-1));
1343 region->init_req(1, if_gt);
1344 }
1345 if (!stopped()) {
1346 // Check for substr count == 0
1347 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1348 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1349 Node* if_zero = generate_slow_guard(bol, nullptr);
1350 if (if_zero != nullptr) {
1351 phi->init_req(2, intcon(0));
1352 region->init_req(2, if_zero);
1353 }
1354 }
1355 if (!stopped()) {
1356 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1357 }
1358 return nullptr;
1359 }
1360
1361 //-----------------------------inline_string_indexOfChar-----------------------
1362 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1363 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1364 return false;
1365 }
1366 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1367 return false;
1368 }
1369 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1370 Node* src = argument(0); // byte[]
1371 Node* int_ch = argument(1);
1372 Node* from_index = argument(2);
1373 Node* max = argument(3);
1374
1375 src = must_be_not_null(src, true);
1376
1377 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1378 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1379 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1380
1381 // Range checks
1382 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1383
1384 // Check for int_ch >= 0
1385 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1386 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1387 {
1388 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1389 uncommon_trap(Deoptimization::Reason_intrinsic,
1390 Deoptimization::Action_maybe_recompile);
1391 }
1392 if (stopped()) {
1393 return true;
1394 }
1395
1396 RegionNode* region = new RegionNode(3);
1397 Node* phi = new PhiNode(region, TypeInt::INT);
1398
1399 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1400 C->set_has_split_ifs(true); // Has chance for split-if optimization
1401 _gvn.transform(result);
1402
1403 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1404 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1405
1406 Node* if_lt = generate_slow_guard(bol, nullptr);
1407 if (if_lt != nullptr) {
1408 // result == -1
1409 phi->init_req(2, result);
1410 region->init_req(2, if_lt);
1411 }
1412 if (!stopped()) {
1413 result = _gvn.transform(new AddINode(result, from_index));
1414 phi->init_req(1, result);
1415 region->init_req(1, control());
1416 }
1417 set_control(_gvn.transform(region));
1418 record_for_igvn(region);
1419 set_result(_gvn.transform(phi));
1420 clear_upper_avx();
1421
1422 return true;
1423 }
1424 //---------------------------inline_string_copy---------------------
1425 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1426 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1427 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1428 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1429 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1430 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1431 bool LibraryCallKit::inline_string_copy(bool compress) {
1432 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1433 return false;
1434 }
1435 int nargs = 5; // 2 oops, 3 ints
1436 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1437
1438 Node* src = argument(0);
1439 Node* src_offset = argument(1);
1440 Node* dst = argument(2);
1441 Node* dst_offset = argument(3);
1442 Node* length = argument(4);
1443
1444 // Check for allocation before we add nodes that would confuse
1445 // tightly_coupled_allocation()
1446 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1447
1448 // Figure out the size and type of the elements we will be copying.
1449 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1450 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1451 if (src_type == nullptr || dst_type == nullptr) {
1452 return false;
1453 }
1454 BasicType src_elem = src_type->elem()->array_element_basic_type();
1455 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1456 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1457 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1458 "Unsupported array types for inline_string_copy");
1459
1460 src = must_be_not_null(src, true);
1461 dst = must_be_not_null(dst, true);
1462
1463 // Convert char[] offsets to byte[] offsets
1464 bool convert_src = (compress && src_elem == T_BYTE);
1465 bool convert_dst = (!compress && dst_elem == T_BYTE);
1466 if (convert_src) {
1467 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1468 } else if (convert_dst) {
1469 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1470 }
1471
1472 // Range checks
1473 generate_string_range_check(src, src_offset, length, convert_src);
1474 generate_string_range_check(dst, dst_offset, length, convert_dst);
1475 if (stopped()) {
1476 return true;
1477 }
1478
1479 Node* src_start = array_element_address(src, src_offset, src_elem);
1480 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1481 // 'src_start' points to src array + scaled offset
1482 // 'dst_start' points to dst array + scaled offset
1483 Node* count = nullptr;
1484 if (compress) {
1485 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1486 } else {
1487 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1488 }
1489
1490 if (alloc != nullptr) {
1491 if (alloc->maybe_set_complete(&_gvn)) {
1492 // "You break it, you buy it."
1493 InitializeNode* init = alloc->initialization();
1494 assert(init->is_complete(), "we just did this");
1495 init->set_complete_with_arraycopy();
1496 assert(dst->is_CheckCastPP(), "sanity");
1497 assert(dst->in(0)->in(0) == init, "dest pinned");
1498 }
1499 // Do not let stores that initialize this object be reordered with
1500 // a subsequent store that would make this object accessible by
1501 // other threads.
1502 // Record what AllocateNode this StoreStore protects so that
1503 // escape analysis can go from the MemBarStoreStoreNode to the
1504 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1505 // based on the escape status of the AllocateNode.
1506 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1507 }
1508 if (compress) {
1509 set_result(_gvn.transform(count));
1510 }
1511 clear_upper_avx();
1512
1513 return true;
1514 }
1515
1516 #ifdef _LP64
1517 #define XTOP ,top() /*additional argument*/
1518 #else //_LP64
1519 #define XTOP /*no additional argument*/
1520 #endif //_LP64
1521
1522 //------------------------inline_string_toBytesU--------------------------
1523 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1524 bool LibraryCallKit::inline_string_toBytesU() {
1525 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1526 return false;
1527 }
1528 // Get the arguments.
1529 Node* value = argument(0);
1530 Node* offset = argument(1);
1531 Node* length = argument(2);
1532
1533 Node* newcopy = nullptr;
1534
1535 // Set the original stack and the reexecute bit for the interpreter to reexecute
1536 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1537 { PreserveReexecuteState preexecs(this);
1538 jvms()->set_should_reexecute(true);
1539
1540 // Check if a null path was taken unconditionally.
1541 value = null_check(value);
1542
1543 RegionNode* bailout = new RegionNode(1);
1544 record_for_igvn(bailout);
1545
1546 // Range checks
1547 generate_negative_guard(offset, bailout);
1548 generate_negative_guard(length, bailout);
1549 generate_limit_guard(offset, length, load_array_length(value), bailout);
1550 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1551 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1552
1553 if (bailout->req() > 1) {
1554 PreserveJVMState pjvms(this);
1555 set_control(_gvn.transform(bailout));
1556 uncommon_trap(Deoptimization::Reason_intrinsic,
1557 Deoptimization::Action_maybe_recompile);
1558 }
1559 if (stopped()) {
1560 return true;
1561 }
1562
1563 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1564 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1565 newcopy = new_array(klass_node, size, 0); // no arguments to push
1566 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1567 guarantee(alloc != nullptr, "created above");
1568
1569 // Calculate starting addresses.
1570 Node* src_start = array_element_address(value, offset, T_CHAR);
1571 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1572
1573 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1574 const TypeInt* toffset = gvn().type(offset)->is_int();
1575 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1576
1577 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1578 const char* copyfunc_name = "arraycopy";
1579 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1580 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1581 OptoRuntime::fast_arraycopy_Type(),
1582 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1583 src_start, dst_start, ConvI2X(length) XTOP);
1584 // Do not let reads from the cloned object float above the arraycopy.
1585 if (alloc->maybe_set_complete(&_gvn)) {
1586 // "You break it, you buy it."
1587 InitializeNode* init = alloc->initialization();
1588 assert(init->is_complete(), "we just did this");
1589 init->set_complete_with_arraycopy();
1590 assert(newcopy->is_CheckCastPP(), "sanity");
1591 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1592 }
1593 // Do not let stores that initialize this object be reordered with
1594 // a subsequent store that would make this object accessible by
1595 // other threads.
1596 // Record what AllocateNode this StoreStore protects so that
1597 // escape analysis can go from the MemBarStoreStoreNode to the
1598 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1599 // based on the escape status of the AllocateNode.
1600 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1601 } // original reexecute is set back here
1602
1603 C->set_has_split_ifs(true); // Has chance for split-if optimization
1604 if (!stopped()) {
1605 set_result(newcopy);
1606 }
1607 clear_upper_avx();
1608
1609 return true;
1610 }
1611
1612 //------------------------inline_string_getCharsU--------------------------
1613 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1614 bool LibraryCallKit::inline_string_getCharsU() {
1615 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1616 return false;
1617 }
1618
1619 // Get the arguments.
1620 Node* src = argument(0);
1621 Node* src_begin = argument(1);
1622 Node* src_end = argument(2); // exclusive offset (i < src_end)
1623 Node* dst = argument(3);
1624 Node* dst_begin = argument(4);
1625
1626 // Check for allocation before we add nodes that would confuse
1627 // tightly_coupled_allocation()
1628 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1629
1630 // Check if a null path was taken unconditionally.
1631 src = null_check(src);
1632 dst = null_check(dst);
1633 if (stopped()) {
1634 return true;
1635 }
1636
1637 // Get length and convert char[] offset to byte[] offset
1638 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1639 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1640
1641 // Range checks
1642 generate_string_range_check(src, src_begin, length, true);
1643 generate_string_range_check(dst, dst_begin, length, false);
1644 if (stopped()) {
1645 return true;
1646 }
1647
1648 if (!stopped()) {
1649 // Calculate starting addresses.
1650 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1651 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1652
1653 // Check if array addresses are aligned to HeapWordSize
1654 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1655 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1656 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1657 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1658
1659 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1660 const char* copyfunc_name = "arraycopy";
1661 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1662 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1663 OptoRuntime::fast_arraycopy_Type(),
1664 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1665 src_start, dst_start, ConvI2X(length) XTOP);
1666 // Do not let reads from the cloned object float above the arraycopy.
1667 if (alloc != nullptr) {
1668 if (alloc->maybe_set_complete(&_gvn)) {
1669 // "You break it, you buy it."
1670 InitializeNode* init = alloc->initialization();
1671 assert(init->is_complete(), "we just did this");
1672 init->set_complete_with_arraycopy();
1673 assert(dst->is_CheckCastPP(), "sanity");
1674 assert(dst->in(0)->in(0) == init, "dest pinned");
1675 }
1676 // Do not let stores that initialize this object be reordered with
1677 // a subsequent store that would make this object accessible by
1678 // other threads.
1679 // Record what AllocateNode this StoreStore protects so that
1680 // escape analysis can go from the MemBarStoreStoreNode to the
1681 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1682 // based on the escape status of the AllocateNode.
1683 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1684 } else {
1685 insert_mem_bar(Op_MemBarCPUOrder);
1686 }
1687 }
1688
1689 C->set_has_split_ifs(true); // Has chance for split-if optimization
1690 return true;
1691 }
1692
1693 //----------------------inline_string_char_access----------------------------
1694 // Store/Load char to/from byte[] array.
1695 // static void StringUTF16.putChar(byte[] val, int index, int c)
1696 // static char StringUTF16.getChar(byte[] val, int index)
1697 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1698 Node* value = argument(0);
1699 Node* index = argument(1);
1700 Node* ch = is_store ? argument(2) : nullptr;
1701
1702 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1703 // correctly requires matched array shapes.
1704 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1705 "sanity: byte[] and char[] bases agree");
1706 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1707 "sanity: byte[] and char[] scales agree");
1708
1709 // Bail when getChar over constants is requested: constant folding would
1710 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1711 // Java method would constant fold nicely instead.
1712 if (!is_store && value->is_Con() && index->is_Con()) {
1713 return false;
1714 }
1715
1716 // Save state and restore on bailout
1717 uint old_sp = sp();
1718 SafePointNode* old_map = clone_map();
1719
1720 value = must_be_not_null(value, true);
1721
1722 Node* adr = array_element_address(value, index, T_CHAR);
1723 if (adr->is_top()) {
1724 set_map(old_map);
1725 set_sp(old_sp);
1726 return false;
1727 }
1728 destruct_map_clone(old_map);
1729 if (is_store) {
1730 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1731 } else {
1732 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);
1733 set_result(ch);
1734 }
1735 return true;
1736 }
1737
1738
1739 //------------------------------inline_math-----------------------------------
1740 // public static double Math.abs(double)
1741 // public static double Math.sqrt(double)
1742 // public static double Math.log(double)
1743 // public static double Math.log10(double)
1744 // public static double Math.round(double)
1745 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1746 Node* arg = argument(0);
1747 Node* n = nullptr;
1748 switch (id) {
1749 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1750 case vmIntrinsics::_dsqrt:
1751 case vmIntrinsics::_dsqrt_strict:
1752 n = new SqrtDNode(C, control(), arg); break;
1753 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1754 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1755 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1756 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1757 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1758 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1759 default: fatal_unexpected_iid(id); break;
1760 }
1761 set_result(_gvn.transform(n));
1762 return true;
1763 }
1764
1765 //------------------------------inline_math-----------------------------------
1766 // public static float Math.abs(float)
1767 // public static int Math.abs(int)
1768 // public static long Math.abs(long)
1769 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1770 Node* arg = argument(0);
1771 Node* n = nullptr;
1772 switch (id) {
1773 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1774 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1775 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1776 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1777 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1778 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1779 default: fatal_unexpected_iid(id); break;
1780 }
1781 set_result(_gvn.transform(n));
1782 return true;
1783 }
1784
1785 //------------------------------runtime_math-----------------------------
1786 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1787 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1788 "must be (DD)D or (D)D type");
1789
1790 // Inputs
1791 Node* a = argument(0);
1792 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1793
1794 const TypePtr* no_memory_effects = nullptr;
1795 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1796 no_memory_effects,
1797 a, top(), b, b ? top() : nullptr);
1798 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1799 #ifdef ASSERT
1800 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1801 assert(value_top == top(), "second value must be top");
1802 #endif
1803
1804 set_result(value);
1805 return true;
1806 }
1807
1808 //------------------------------inline_math_pow-----------------------------
1809 bool LibraryCallKit::inline_math_pow() {
1810 Node* exp = argument(2);
1811 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1812 if (d != nullptr) {
1813 if (d->getd() == 2.0) {
1814 // Special case: pow(x, 2.0) => x * x
1815 Node* base = argument(0);
1816 set_result(_gvn.transform(new MulDNode(base, base)));
1817 return true;
1818 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1819 // Special case: pow(x, 0.5) => sqrt(x)
1820 Node* base = argument(0);
1821 Node* zero = _gvn.zerocon(T_DOUBLE);
1822
1823 RegionNode* region = new RegionNode(3);
1824 Node* phi = new PhiNode(region, Type::DOUBLE);
1825
1826 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1827 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1828 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1829 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1830 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1831
1832 Node* if_pow = generate_slow_guard(test, nullptr);
1833 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1834 phi->init_req(1, value_sqrt);
1835 region->init_req(1, control());
1836
1837 if (if_pow != nullptr) {
1838 set_control(if_pow);
1839 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1840 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1841 const TypePtr* no_memory_effects = nullptr;
1842 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1843 no_memory_effects, base, top(), exp, top());
1844 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1845 #ifdef ASSERT
1846 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1847 assert(value_top == top(), "second value must be top");
1848 #endif
1849 phi->init_req(2, value_pow);
1850 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1851 }
1852
1853 C->set_has_split_ifs(true); // Has chance for split-if optimization
1854 set_control(_gvn.transform(region));
1855 record_for_igvn(region);
1856 set_result(_gvn.transform(phi));
1857
1858 return true;
1859 }
1860 }
1861
1862 return StubRoutines::dpow() != nullptr ?
1863 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1864 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1865 }
1866
1867 //------------------------------inline_math_native-----------------------------
1868 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1869 switch (id) {
1870 case vmIntrinsics::_dsin:
1871 return StubRoutines::dsin() != nullptr ?
1872 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1873 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1874 case vmIntrinsics::_dcos:
1875 return StubRoutines::dcos() != nullptr ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1878 case vmIntrinsics::_dtan:
1879 return StubRoutines::dtan() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1882 case vmIntrinsics::_dtanh:
1883 return StubRoutines::dtanh() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1885 case vmIntrinsics::_dexp:
1886 return StubRoutines::dexp() != nullptr ?
1887 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1888 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1889 case vmIntrinsics::_dlog:
1890 return StubRoutines::dlog() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1893 case vmIntrinsics::_dlog10:
1894 return StubRoutines::dlog10() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1897
1898 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1899 case vmIntrinsics::_ceil:
1900 case vmIntrinsics::_floor:
1901 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1902
1903 case vmIntrinsics::_dsqrt:
1904 case vmIntrinsics::_dsqrt_strict:
1905 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1906 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1908 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1909 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1910
1911 case vmIntrinsics::_dpow: return inline_math_pow();
1912 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1913 case vmIntrinsics::_fcopySign: return inline_math(id);
1914 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1916 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1917
1918 // These intrinsics are not yet correctly implemented
1919 case vmIntrinsics::_datan2:
1920 return false;
1921
1922 default:
1923 fatal_unexpected_iid(id);
1924 return false;
1925 }
1926 }
1927
1928 //----------------------------inline_notify-----------------------------------*
1929 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1930 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1931 address func;
1932 if (id == vmIntrinsics::_notify) {
1933 func = OptoRuntime::monitor_notify_Java();
1934 } else {
1935 func = OptoRuntime::monitor_notifyAll_Java();
1936 }
1937 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1938 make_slow_call_ex(call, env()->Throwable_klass(), false);
1939 return true;
1940 }
1941
1942
1943 //----------------------------inline_min_max-----------------------------------
1944 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1945 Node* a = nullptr;
1946 Node* b = nullptr;
1947 Node* n = nullptr;
1948 switch (id) {
1949 case vmIntrinsics::_min:
1950 case vmIntrinsics::_max:
1951 case vmIntrinsics::_minF:
1952 case vmIntrinsics::_maxF:
1953 case vmIntrinsics::_minF_strict:
1954 case vmIntrinsics::_maxF_strict:
1955 case vmIntrinsics::_min_strict:
1956 case vmIntrinsics::_max_strict:
1957 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1958 a = argument(0);
1959 b = argument(1);
1960 break;
1961 case vmIntrinsics::_minD:
1962 case vmIntrinsics::_maxD:
1963 case vmIntrinsics::_minD_strict:
1964 case vmIntrinsics::_maxD_strict:
1965 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1966 a = argument(0);
1967 b = argument(2);
1968 break;
1969 case vmIntrinsics::_minL:
1970 case vmIntrinsics::_maxL:
1971 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1972 a = argument(0);
1973 b = argument(2);
1974 break;
1975 default:
1976 fatal_unexpected_iid(id);
1977 break;
1978 }
1979
1980 switch (id) {
1981 case vmIntrinsics::_min:
1982 case vmIntrinsics::_min_strict:
1983 n = new MinINode(a, b);
1984 break;
1985 case vmIntrinsics::_max:
1986 case vmIntrinsics::_max_strict:
1987 n = new MaxINode(a, b);
1988 break;
1989 case vmIntrinsics::_minF:
1990 case vmIntrinsics::_minF_strict:
1991 n = new MinFNode(a, b);
1992 break;
1993 case vmIntrinsics::_maxF:
1994 case vmIntrinsics::_maxF_strict:
1995 n = new MaxFNode(a, b);
1996 break;
1997 case vmIntrinsics::_minD:
1998 case vmIntrinsics::_minD_strict:
1999 n = new MinDNode(a, b);
2000 break;
2001 case vmIntrinsics::_maxD:
2002 case vmIntrinsics::_maxD_strict:
2003 n = new MaxDNode(a, b);
2004 break;
2005 case vmIntrinsics::_minL:
2006 n = new MinLNode(_gvn.C, a, b);
2007 break;
2008 case vmIntrinsics::_maxL:
2009 n = new MaxLNode(_gvn.C, a, b);
2010 break;
2011 default:
2012 fatal_unexpected_iid(id);
2013 break;
2014 }
2015
2016 set_result(_gvn.transform(n));
2017 return true;
2018 }
2019
2020 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2021 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2022 env()->ArithmeticException_instance())) {
2023 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2024 // so let's bail out intrinsic rather than risking deopting again.
2025 return false;
2026 }
2027
2028 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2029 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2030 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2031 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2032
2033 {
2034 PreserveJVMState pjvms(this);
2035 PreserveReexecuteState preexecs(this);
2036 jvms()->set_should_reexecute(true);
2037
2038 set_control(slow_path);
2039 set_i_o(i_o());
2040
2041 builtin_throw(Deoptimization::Reason_intrinsic,
2042 env()->ArithmeticException_instance(),
2043 /*allow_too_many_traps*/ false);
2044 }
2045
2046 set_control(fast_path);
2047 set_result(math);
2048 return true;
2049 }
2050
2051 template <typename OverflowOp>
2052 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2053 typedef typename OverflowOp::MathOp MathOp;
2054
2055 MathOp* mathOp = new MathOp(arg1, arg2);
2056 Node* operation = _gvn.transform( mathOp );
2057 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2058 return inline_math_mathExact(operation, ofcheck);
2059 }
2060
2061 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2062 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2063 }
2064
2065 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2066 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2067 }
2068
2069 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2070 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2071 }
2072
2073 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2074 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2075 }
2076
2077 bool LibraryCallKit::inline_math_negateExactI() {
2078 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2079 }
2080
2081 bool LibraryCallKit::inline_math_negateExactL() {
2082 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2083 }
2084
2085 bool LibraryCallKit::inline_math_multiplyExactI() {
2086 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2087 }
2088
2089 bool LibraryCallKit::inline_math_multiplyExactL() {
2090 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2091 }
2092
2093 bool LibraryCallKit::inline_math_multiplyHigh() {
2094 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2095 return true;
2096 }
2097
2098 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2099 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2100 return true;
2101 }
2102
2103 inline int
2104 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2105 const TypePtr* base_type = TypePtr::NULL_PTR;
2106 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2107 if (base_type == nullptr) {
2108 // Unknown type.
2109 return Type::AnyPtr;
2110 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2111 // Since this is a null+long form, we have to switch to a rawptr.
2112 base = _gvn.transform(new CastX2PNode(offset));
2113 offset = MakeConX(0);
2114 return Type::RawPtr;
2115 } else if (base_type->base() == Type::RawPtr) {
2116 return Type::RawPtr;
2117 } else if (base_type->isa_oopptr()) {
2118 // Base is never null => always a heap address.
2119 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2120 return Type::OopPtr;
2121 }
2122 // Offset is small => always a heap address.
2123 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2124 if (offset_type != nullptr &&
2125 base_type->offset() == 0 && // (should always be?)
2126 offset_type->_lo >= 0 &&
2127 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2128 return Type::OopPtr;
2129 } else if (type == T_OBJECT) {
2130 // off heap access to an oop doesn't make any sense. Has to be on
2131 // heap.
2132 return Type::OopPtr;
2133 }
2134 // Otherwise, it might either be oop+off or null+addr.
2135 return Type::AnyPtr;
2136 } else {
2137 // No information:
2138 return Type::AnyPtr;
2139 }
2140 }
2141
2142 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2143 Node* uncasted_base = base;
2144 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2145 if (kind == Type::RawPtr) {
2146 return basic_plus_adr(top(), uncasted_base, offset);
2147 } else if (kind == Type::AnyPtr) {
2148 assert(base == uncasted_base, "unexpected base change");
2149 if (can_cast) {
2150 if (!_gvn.type(base)->speculative_maybe_null() &&
2151 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2152 // According to profiling, this access is always on
2153 // heap. Casting the base to not null and thus avoiding membars
2154 // around the access should allow better optimizations
2155 Node* null_ctl = top();
2156 base = null_check_oop(base, &null_ctl, true, true, true);
2157 assert(null_ctl->is_top(), "no null control here");
2158 return basic_plus_adr(base, offset);
2159 } else if (_gvn.type(base)->speculative_always_null() &&
2160 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2161 // According to profiling, this access is always off
2162 // heap.
2163 base = null_assert(base);
2164 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2165 offset = MakeConX(0);
2166 return basic_plus_adr(top(), raw_base, offset);
2167 }
2168 }
2169 // We don't know if it's an on heap or off heap access. Fall back
2170 // to raw memory access.
2171 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2172 return basic_plus_adr(top(), raw, offset);
2173 } else {
2174 assert(base == uncasted_base, "unexpected base change");
2175 // We know it's an on heap access so base can't be null
2176 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2177 base = must_be_not_null(base, true);
2178 }
2179 return basic_plus_adr(base, offset);
2180 }
2181 }
2182
2183 //--------------------------inline_number_methods-----------------------------
2184 // inline int Integer.numberOfLeadingZeros(int)
2185 // inline int Long.numberOfLeadingZeros(long)
2186 //
2187 // inline int Integer.numberOfTrailingZeros(int)
2188 // inline int Long.numberOfTrailingZeros(long)
2189 //
2190 // inline int Integer.bitCount(int)
2191 // inline int Long.bitCount(long)
2192 //
2193 // inline char Character.reverseBytes(char)
2194 // inline short Short.reverseBytes(short)
2195 // inline int Integer.reverseBytes(int)
2196 // inline long Long.reverseBytes(long)
2197 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2198 Node* arg = argument(0);
2199 Node* n = nullptr;
2200 switch (id) {
2201 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2202 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2203 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2204 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2205 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2206 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2207 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2208 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2209 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2210 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2211 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2212 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2213 default: fatal_unexpected_iid(id); break;
2214 }
2215 set_result(_gvn.transform(n));
2216 return true;
2217 }
2218
2219 //--------------------------inline_bitshuffle_methods-----------------------------
2220 // inline int Integer.compress(int, int)
2221 // inline int Integer.expand(int, int)
2222 // inline long Long.compress(long, long)
2223 // inline long Long.expand(long, long)
2224 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2225 Node* n = nullptr;
2226 switch (id) {
2227 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2228 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2229 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2230 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2231 default: fatal_unexpected_iid(id); break;
2232 }
2233 set_result(_gvn.transform(n));
2234 return true;
2235 }
2236
2237 //--------------------------inline_number_methods-----------------------------
2238 // inline int Integer.compareUnsigned(int, int)
2239 // inline int Long.compareUnsigned(long, long)
2240 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2241 Node* arg1 = argument(0);
2242 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2243 Node* n = nullptr;
2244 switch (id) {
2245 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2246 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2247 default: fatal_unexpected_iid(id); break;
2248 }
2249 set_result(_gvn.transform(n));
2250 return true;
2251 }
2252
2253 //--------------------------inline_unsigned_divmod_methods-----------------------------
2254 // inline int Integer.divideUnsigned(int, int)
2255 // inline int Integer.remainderUnsigned(int, int)
2256 // inline long Long.divideUnsigned(long, long)
2257 // inline long Long.remainderUnsigned(long, long)
2258 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2259 Node* n = nullptr;
2260 switch (id) {
2261 case vmIntrinsics::_divideUnsigned_i: {
2262 zero_check_int(argument(1));
2263 // Compile-time detect of null-exception
2264 if (stopped()) {
2265 return true; // keep the graph constructed so far
2266 }
2267 n = new UDivINode(control(), argument(0), argument(1));
2268 break;
2269 }
2270 case vmIntrinsics::_divideUnsigned_l: {
2271 zero_check_long(argument(2));
2272 // Compile-time detect of null-exception
2273 if (stopped()) {
2274 return true; // keep the graph constructed so far
2275 }
2276 n = new UDivLNode(control(), argument(0), argument(2));
2277 break;
2278 }
2279 case vmIntrinsics::_remainderUnsigned_i: {
2280 zero_check_int(argument(1));
2281 // Compile-time detect of null-exception
2282 if (stopped()) {
2283 return true; // keep the graph constructed so far
2284 }
2285 n = new UModINode(control(), argument(0), argument(1));
2286 break;
2287 }
2288 case vmIntrinsics::_remainderUnsigned_l: {
2289 zero_check_long(argument(2));
2290 // Compile-time detect of null-exception
2291 if (stopped()) {
2292 return true; // keep the graph constructed so far
2293 }
2294 n = new UModLNode(control(), argument(0), argument(2));
2295 break;
2296 }
2297 default: fatal_unexpected_iid(id); break;
2298 }
2299 set_result(_gvn.transform(n));
2300 return true;
2301 }
2302
2303 //----------------------------inline_unsafe_access----------------------------
2304
2305 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2306 // Attempt to infer a sharper value type from the offset and base type.
2307 ciKlass* sharpened_klass = nullptr;
2308
2309 // See if it is an instance field, with an object type.
2310 if (alias_type->field() != nullptr) {
2311 if (alias_type->field()->type()->is_klass()) {
2312 sharpened_klass = alias_type->field()->type()->as_klass();
2313 }
2314 }
2315
2316 const TypeOopPtr* result = nullptr;
2317 // See if it is a narrow oop array.
2318 if (adr_type->isa_aryptr()) {
2319 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2320 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2321 if (elem_type != nullptr && elem_type->is_loaded()) {
2322 // Sharpen the value type.
2323 result = elem_type;
2324 }
2325 }
2326 }
2327
2328 // The sharpened class might be unloaded if there is no class loader
2329 // contraint in place.
2330 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2331 // Sharpen the value type.
2332 result = TypeOopPtr::make_from_klass(sharpened_klass);
2333 }
2334 if (result != nullptr) {
2335 #ifndef PRODUCT
2336 if (C->print_intrinsics() || C->print_inlining()) {
2337 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2338 tty->print(" sharpened value: "); result->dump(); tty->cr();
2339 }
2340 #endif
2341 }
2342 return result;
2343 }
2344
2345 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2346 switch (kind) {
2347 case Relaxed:
2348 return MO_UNORDERED;
2349 case Opaque:
2350 return MO_RELAXED;
2351 case Acquire:
2352 return MO_ACQUIRE;
2353 case Release:
2354 return MO_RELEASE;
2355 case Volatile:
2356 return MO_SEQ_CST;
2357 default:
2358 ShouldNotReachHere();
2359 return 0;
2360 }
2361 }
2362
2363 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2364 if (callee()->is_static()) return false; // caller must have the capability!
2365 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2366 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2367 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2368 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2369
2370 if (is_reference_type(type)) {
2371 decorators |= ON_UNKNOWN_OOP_REF;
2372 }
2373
2374 if (unaligned) {
2375 decorators |= C2_UNALIGNED;
2376 }
2377
2378 #ifndef PRODUCT
2379 {
2380 ResourceMark rm;
2381 // Check the signatures.
2382 ciSignature* sig = callee()->signature();
2383 #ifdef ASSERT
2384 if (!is_store) {
2385 // Object getReference(Object base, int/long offset), etc.
2386 BasicType rtype = sig->return_type()->basic_type();
2387 assert(rtype == type, "getter must return the expected value");
2388 assert(sig->count() == 2, "oop getter has 2 arguments");
2389 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2390 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2391 } else {
2392 // void putReference(Object base, int/long offset, Object x), etc.
2393 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2394 assert(sig->count() == 3, "oop putter has 3 arguments");
2395 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2396 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2397 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2398 assert(vtype == type, "putter must accept the expected value");
2399 }
2400 #endif // ASSERT
2401 }
2402 #endif //PRODUCT
2403
2404 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2405
2406 Node* receiver = argument(0); // type: oop
2407
2408 // Build address expression.
2409 Node* heap_base_oop = top();
2410
2411 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2412 Node* base = argument(1); // type: oop
2413 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2414 Node* offset = argument(2); // type: long
2415 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2416 // to be plain byte offsets, which are also the same as those accepted
2417 // by oopDesc::field_addr.
2418 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2419 "fieldOffset must be byte-scaled");
2420 // 32-bit machines ignore the high half!
2421 offset = ConvL2X(offset);
2422
2423 // Save state and restore on bailout
2424 uint old_sp = sp();
2425 SafePointNode* old_map = clone_map();
2426
2427 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2428 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2429
2430 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2431 if (type != T_OBJECT) {
2432 decorators |= IN_NATIVE; // off-heap primitive access
2433 } else {
2434 set_map(old_map);
2435 set_sp(old_sp);
2436 return false; // off-heap oop accesses are not supported
2437 }
2438 } else {
2439 heap_base_oop = base; // on-heap or mixed access
2440 }
2441
2442 // Can base be null? Otherwise, always on-heap access.
2443 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2444
2445 if (!can_access_non_heap) {
2446 decorators |= IN_HEAP;
2447 }
2448
2449 Node* val = is_store ? argument(4) : nullptr;
2450
2451 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2452 if (adr_type == TypePtr::NULL_PTR) {
2453 set_map(old_map);
2454 set_sp(old_sp);
2455 return false; // off-heap access with zero address
2456 }
2457
2458 // Try to categorize the address.
2459 Compile::AliasType* alias_type = C->alias_type(adr_type);
2460 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2461
2462 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2463 alias_type->adr_type() == TypeAryPtr::RANGE) {
2464 set_map(old_map);
2465 set_sp(old_sp);
2466 return false; // not supported
2467 }
2468
2469 bool mismatched = false;
2470 BasicType bt = alias_type->basic_type();
2471 if (bt != T_ILLEGAL) {
2472 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2473 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2474 // Alias type doesn't differentiate between byte[] and boolean[]).
2475 // Use address type to get the element type.
2476 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2477 }
2478 if (is_reference_type(bt, true)) {
2479 // accessing an array field with getReference is not a mismatch
2480 bt = T_OBJECT;
2481 }
2482 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2483 // Don't intrinsify mismatched object accesses
2484 set_map(old_map);
2485 set_sp(old_sp);
2486 return false;
2487 }
2488 mismatched = (bt != type);
2489 } else if (alias_type->adr_type()->isa_oopptr()) {
2490 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2491 }
2492
2493 destruct_map_clone(old_map);
2494 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2495
2496 if (mismatched) {
2497 decorators |= C2_MISMATCHED;
2498 }
2499
2500 // First guess at the value type.
2501 const Type *value_type = Type::get_const_basic_type(type);
2502
2503 // Figure out the memory ordering.
2504 decorators |= mo_decorator_for_access_kind(kind);
2505
2506 if (!is_store && type == T_OBJECT) {
2507 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2508 if (tjp != nullptr) {
2509 value_type = tjp;
2510 }
2511 }
2512
2513 receiver = null_check(receiver);
2514 if (stopped()) {
2515 return true;
2516 }
2517 // Heap pointers get a null-check from the interpreter,
2518 // as a courtesy. However, this is not guaranteed by Unsafe,
2519 // and it is not possible to fully distinguish unintended nulls
2520 // from intended ones in this API.
2521
2522 if (!is_store) {
2523 Node* p = nullptr;
2524 // Try to constant fold a load from a constant field
2525 ciField* field = alias_type->field();
2526 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2527 // final or stable field
2528 p = make_constant_from_field(field, heap_base_oop);
2529 }
2530
2531 if (p == nullptr) { // Could not constant fold the load
2532 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2533 // Normalize the value returned by getBoolean in the following cases
2534 if (type == T_BOOLEAN &&
2535 (mismatched ||
2536 heap_base_oop == top() || // - heap_base_oop is null or
2537 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2538 // and the unsafe access is made to large offset
2539 // (i.e., larger than the maximum offset necessary for any
2540 // field access)
2541 ) {
2542 IdealKit ideal = IdealKit(this);
2543 #define __ ideal.
2544 IdealVariable normalized_result(ideal);
2545 __ declarations_done();
2546 __ set(normalized_result, p);
2547 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2548 __ set(normalized_result, ideal.ConI(1));
2549 ideal.end_if();
2550 final_sync(ideal);
2551 p = __ value(normalized_result);
2552 #undef __
2553 }
2554 }
2555 if (type == T_ADDRESS) {
2556 p = gvn().transform(new CastP2XNode(nullptr, p));
2557 p = ConvX2UL(p);
2558 }
2559 // The load node has the control of the preceding MemBarCPUOrder. All
2560 // following nodes will have the control of the MemBarCPUOrder inserted at
2561 // the end of this method. So, pushing the load onto the stack at a later
2562 // point is fine.
2563 set_result(p);
2564 } else {
2565 if (bt == T_ADDRESS) {
2566 // Repackage the long as a pointer.
2567 val = ConvL2X(val);
2568 val = gvn().transform(new CastX2PNode(val));
2569 }
2570 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2571 }
2572
2573 return true;
2574 }
2575
2576 //----------------------------inline_unsafe_load_store----------------------------
2577 // This method serves a couple of different customers (depending on LoadStoreKind):
2578 //
2579 // LS_cmp_swap:
2580 //
2581 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2582 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2583 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2584 //
2585 // LS_cmp_swap_weak:
2586 //
2587 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2588 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2589 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2590 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2591 //
2592 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2593 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2594 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2595 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2596 //
2597 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2598 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2599 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2600 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2601 //
2602 // LS_cmp_exchange:
2603 //
2604 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2605 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2606 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2607 //
2608 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2610 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2611 //
2612 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2615 //
2616 // LS_get_add:
2617 //
2618 // int getAndAddInt( Object o, long offset, int delta)
2619 // long getAndAddLong(Object o, long offset, long delta)
2620 //
2621 // LS_get_set:
2622 //
2623 // int getAndSet(Object o, long offset, int newValue)
2624 // long getAndSet(Object o, long offset, long newValue)
2625 // Object getAndSet(Object o, long offset, Object newValue)
2626 //
2627 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2628 // This basic scheme here is the same as inline_unsafe_access, but
2629 // differs in enough details that combining them would make the code
2630 // overly confusing. (This is a true fact! I originally combined
2631 // them, but even I was confused by it!) As much code/comments as
2632 // possible are retained from inline_unsafe_access though to make
2633 // the correspondences clearer. - dl
2634
2635 if (callee()->is_static()) return false; // caller must have the capability!
2636
2637 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2638 decorators |= mo_decorator_for_access_kind(access_kind);
2639
2640 #ifndef PRODUCT
2641 BasicType rtype;
2642 {
2643 ResourceMark rm;
2644 // Check the signatures.
2645 ciSignature* sig = callee()->signature();
2646 rtype = sig->return_type()->basic_type();
2647 switch(kind) {
2648 case LS_get_add:
2649 case LS_get_set: {
2650 // Check the signatures.
2651 #ifdef ASSERT
2652 assert(rtype == type, "get and set must return the expected type");
2653 assert(sig->count() == 3, "get and set has 3 arguments");
2654 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2655 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2656 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2657 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2658 #endif // ASSERT
2659 break;
2660 }
2661 case LS_cmp_swap:
2662 case LS_cmp_swap_weak: {
2663 // Check the signatures.
2664 #ifdef ASSERT
2665 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2666 assert(sig->count() == 4, "CAS has 4 arguments");
2667 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2668 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2669 #endif // ASSERT
2670 break;
2671 }
2672 case LS_cmp_exchange: {
2673 // Check the signatures.
2674 #ifdef ASSERT
2675 assert(rtype == type, "CAS must return the expected type");
2676 assert(sig->count() == 4, "CAS has 4 arguments");
2677 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2678 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2679 #endif // ASSERT
2680 break;
2681 }
2682 default:
2683 ShouldNotReachHere();
2684 }
2685 }
2686 #endif //PRODUCT
2687
2688 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2689
2690 // Get arguments:
2691 Node* receiver = nullptr;
2692 Node* base = nullptr;
2693 Node* offset = nullptr;
2694 Node* oldval = nullptr;
2695 Node* newval = nullptr;
2696 switch(kind) {
2697 case LS_cmp_swap:
2698 case LS_cmp_swap_weak:
2699 case LS_cmp_exchange: {
2700 const bool two_slot_type = type2size[type] == 2;
2701 receiver = argument(0); // type: oop
2702 base = argument(1); // type: oop
2703 offset = argument(2); // type: long
2704 oldval = argument(4); // type: oop, int, or long
2705 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2706 break;
2707 }
2708 case LS_get_add:
2709 case LS_get_set: {
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = nullptr;
2714 newval = argument(4); // type: oop, int, or long
2715 break;
2716 }
2717 default:
2718 ShouldNotReachHere();
2719 }
2720
2721 // Build field offset expression.
2722 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2723 // to be plain byte offsets, which are also the same as those accepted
2724 // by oopDesc::field_addr.
2725 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2726 // 32-bit machines ignore the high half of long offsets
2727 offset = ConvL2X(offset);
2728 // Save state and restore on bailout
2729 uint old_sp = sp();
2730 SafePointNode* old_map = clone_map();
2731 Node* adr = make_unsafe_address(base, offset,type, false);
2732 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2733
2734 Compile::AliasType* alias_type = C->alias_type(adr_type);
2735 BasicType bt = alias_type->basic_type();
2736 if (bt != T_ILLEGAL &&
2737 (is_reference_type(bt) != (type == T_OBJECT))) {
2738 // Don't intrinsify mismatched object accesses.
2739 set_map(old_map);
2740 set_sp(old_sp);
2741 return false;
2742 }
2743
2744 destruct_map_clone(old_map);
2745
2746 // For CAS, unlike inline_unsafe_access, there seems no point in
2747 // trying to refine types. Just use the coarse types here.
2748 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2749 const Type *value_type = Type::get_const_basic_type(type);
2750
2751 switch (kind) {
2752 case LS_get_set:
2753 case LS_cmp_exchange: {
2754 if (type == T_OBJECT) {
2755 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2756 if (tjp != nullptr) {
2757 value_type = tjp;
2758 }
2759 }
2760 break;
2761 }
2762 case LS_cmp_swap:
2763 case LS_cmp_swap_weak:
2764 case LS_get_add:
2765 break;
2766 default:
2767 ShouldNotReachHere();
2768 }
2769
2770 // Null check receiver.
2771 receiver = null_check(receiver);
2772 if (stopped()) {
2773 return true;
2774 }
2775
2776 int alias_idx = C->get_alias_index(adr_type);
2777
2778 if (is_reference_type(type)) {
2779 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2780
2781 // Transformation of a value which could be null pointer (CastPP #null)
2782 // could be delayed during Parse (for example, in adjust_map_after_if()).
2783 // Execute transformation here to avoid barrier generation in such case.
2784 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2785 newval = _gvn.makecon(TypePtr::NULL_PTR);
2786
2787 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2788 // Refine the value to a null constant, when it is known to be null
2789 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2790 }
2791 }
2792
2793 Node* result = nullptr;
2794 switch (kind) {
2795 case LS_cmp_exchange: {
2796 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2797 oldval, newval, value_type, type, decorators);
2798 break;
2799 }
2800 case LS_cmp_swap_weak:
2801 decorators |= C2_WEAK_CMPXCHG;
2802 case LS_cmp_swap: {
2803 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2804 oldval, newval, value_type, type, decorators);
2805 break;
2806 }
2807 case LS_get_set: {
2808 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2809 newval, value_type, type, decorators);
2810 break;
2811 }
2812 case LS_get_add: {
2813 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2814 newval, value_type, type, decorators);
2815 break;
2816 }
2817 default:
2818 ShouldNotReachHere();
2819 }
2820
2821 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2822 set_result(result);
2823 return true;
2824 }
2825
2826 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2827 // Regardless of form, don't allow previous ld/st to move down,
2828 // then issue acquire, release, or volatile mem_bar.
2829 insert_mem_bar(Op_MemBarCPUOrder);
2830 switch(id) {
2831 case vmIntrinsics::_loadFence:
2832 insert_mem_bar(Op_LoadFence);
2833 return true;
2834 case vmIntrinsics::_storeFence:
2835 insert_mem_bar(Op_StoreFence);
2836 return true;
2837 case vmIntrinsics::_storeStoreFence:
2838 insert_mem_bar(Op_StoreStoreFence);
2839 return true;
2840 case vmIntrinsics::_fullFence:
2841 insert_mem_bar(Op_MemBarVolatile);
2842 return true;
2843 default:
2844 fatal_unexpected_iid(id);
2845 return false;
2846 }
2847 }
2848
2849 bool LibraryCallKit::inline_onspinwait() {
2850 insert_mem_bar(Op_OnSpinWait);
2851 return true;
2852 }
2853
2854 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2855 if (!kls->is_Con()) {
2856 return true;
2857 }
2858 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2859 if (klsptr == nullptr) {
2860 return true;
2861 }
2862 ciInstanceKlass* ik = klsptr->instance_klass();
2863 // don't need a guard for a klass that is already initialized
2864 return !ik->is_initialized();
2865 }
2866
2867 //----------------------------inline_unsafe_writeback0-------------------------
2868 // public native void Unsafe.writeback0(long address)
2869 bool LibraryCallKit::inline_unsafe_writeback0() {
2870 if (!Matcher::has_match_rule(Op_CacheWB)) {
2871 return false;
2872 }
2873 #ifndef PRODUCT
2874 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2875 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2876 ciSignature* sig = callee()->signature();
2877 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2878 #endif
2879 null_check_receiver(); // null-check, then ignore
2880 Node *addr = argument(1);
2881 addr = new CastX2PNode(addr);
2882 addr = _gvn.transform(addr);
2883 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2884 flush = _gvn.transform(flush);
2885 set_memory(flush, TypeRawPtr::BOTTOM);
2886 return true;
2887 }
2888
2889 //----------------------------inline_unsafe_writeback0-------------------------
2890 // public native void Unsafe.writeback0(long address)
2891 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2892 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2893 return false;
2894 }
2895 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2896 return false;
2897 }
2898 #ifndef PRODUCT
2899 assert(Matcher::has_match_rule(Op_CacheWB),
2900 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2901 : "found match rule for CacheWBPostSync but not CacheWB"));
2902
2903 #endif
2904 null_check_receiver(); // null-check, then ignore
2905 Node *sync;
2906 if (is_pre) {
2907 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2908 } else {
2909 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2910 }
2911 sync = _gvn.transform(sync);
2912 set_memory(sync, TypeRawPtr::BOTTOM);
2913 return true;
2914 }
2915
2916 //----------------------------inline_unsafe_allocate---------------------------
2917 // public native Object Unsafe.allocateInstance(Class<?> cls);
2918 bool LibraryCallKit::inline_unsafe_allocate() {
2919
2920 #if INCLUDE_JVMTI
2921 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2922 return false;
2923 }
2924 #endif //INCLUDE_JVMTI
2925
2926 if (callee()->is_static()) return false; // caller must have the capability!
2927
2928 null_check_receiver(); // null-check, then ignore
2929 Node* cls = null_check(argument(1));
2930 if (stopped()) return true;
2931
2932 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2933 kls = null_check(kls);
2934 if (stopped()) return true; // argument was like int.class
2935
2936 #if INCLUDE_JVMTI
2937 // Don't try to access new allocated obj in the intrinsic.
2938 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2939 // Deoptimize and allocate in interpreter instead.
2940 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2941 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2942 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2943 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2944 {
2945 BuildCutout unless(this, tst, PROB_MAX);
2946 uncommon_trap(Deoptimization::Reason_intrinsic,
2947 Deoptimization::Action_make_not_entrant);
2948 }
2949 if (stopped()) {
2950 return true;
2951 }
2952 #endif //INCLUDE_JVMTI
2953
2954 Node* test = nullptr;
2955 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2956 // Note: The argument might still be an illegal value like
2957 // Serializable.class or Object[].class. The runtime will handle it.
2958 // But we must make an explicit check for initialization.
2959 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2960 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2961 // can generate code to load it as unsigned byte.
2962 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2963 Node* bits = intcon(InstanceKlass::fully_initialized);
2964 test = _gvn.transform(new SubINode(inst, bits));
2965 // The 'test' is non-zero if we need to take a slow path.
2966 }
2967
2968 Node* obj = new_instance(kls, test);
2969 set_result(obj);
2970 return true;
2971 }
2972
2973 //------------------------inline_native_time_funcs--------------
2974 // inline code for System.currentTimeMillis() and System.nanoTime()
2975 // these have the same type and signature
2976 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2977 const TypeFunc* tf = OptoRuntime::void_long_Type();
2978 const TypePtr* no_memory_effects = nullptr;
2979 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2980 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2981 #ifdef ASSERT
2982 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2983 assert(value_top == top(), "second value must be top");
2984 #endif
2985 set_result(value);
2986 return true;
2987 }
2988
2989
2990 #if INCLUDE_JVMTI
2991
2992 // When notifications are disabled then just update the VTMS transition bit and return.
2993 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2994 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2995 if (!DoJVMTIVirtualThreadTransitions) {
2996 return true;
2997 }
2998 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2999 IdealKit ideal(this);
3000
3001 Node* ONE = ideal.ConI(1);
3002 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3003 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3004 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3005
3006 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3007 sync_kit(ideal);
3008 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3009 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3010 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3011 ideal.sync_kit(this);
3012 } ideal.else_(); {
3013 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3014 Node* thread = ideal.thread();
3015 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3016 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3017
3018 sync_kit(ideal);
3019 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3020 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3021
3022 ideal.sync_kit(this);
3023 } ideal.end_if();
3024 final_sync(ideal);
3025
3026 return true;
3027 }
3028
3029 // Always update the is_disable_suspend bit.
3030 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3031 if (!DoJVMTIVirtualThreadTransitions) {
3032 return true;
3033 }
3034 IdealKit ideal(this);
3035
3036 {
3037 // unconditionally update the is_disable_suspend bit in current JavaThread
3038 Node* thread = ideal.thread();
3039 Node* arg = _gvn.transform(argument(0)); // argument for notification
3040 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3041 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3042
3043 sync_kit(ideal);
3044 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3045 ideal.sync_kit(this);
3046 }
3047 final_sync(ideal);
3048
3049 return true;
3050 }
3051
3052 #endif // INCLUDE_JVMTI
3053
3054 #ifdef JFR_HAVE_INTRINSICS
3055
3056 /**
3057 * if oop->klass != null
3058 * // normal class
3059 * epoch = _epoch_state ? 2 : 1
3060 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3061 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3062 * }
3063 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3064 * else
3065 * // primitive class
3066 * if oop->array_klass != null
3067 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3068 * else
3069 * id = LAST_TYPE_ID + 1 // void class path
3070 * if (!signaled)
3071 * signaled = true
3072 */
3073 bool LibraryCallKit::inline_native_classID() {
3074 Node* cls = argument(0);
3075
3076 IdealKit ideal(this);
3077 #define __ ideal.
3078 IdealVariable result(ideal); __ declarations_done();
3079 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3080 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3081 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3082
3083
3084 __ if_then(kls, BoolTest::ne, null()); {
3085 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3086 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3087
3088 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3089 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3090 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3091 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3092 mask = _gvn.transform(new OrLNode(mask, epoch));
3093 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3094
3095 float unlikely = PROB_UNLIKELY(0.999);
3096 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3097 sync_kit(ideal);
3098 make_runtime_call(RC_LEAF,
3099 OptoRuntime::class_id_load_barrier_Type(),
3100 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3101 "class id load barrier",
3102 TypePtr::BOTTOM,
3103 kls);
3104 ideal.sync_kit(this);
3105 } __ end_if();
3106
3107 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3108 } __ else_(); {
3109 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3110 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3111 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3112 __ if_then(array_kls, BoolTest::ne, null()); {
3113 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3114 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3115 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3116 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3117 } __ else_(); {
3118 // void class case
3119 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3120 } __ end_if();
3121
3122 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3123 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3124 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3125 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3126 } __ end_if();
3127 } __ end_if();
3128
3129 final_sync(ideal);
3130 set_result(ideal.value(result));
3131 #undef __
3132 return true;
3133 }
3134
3135 //------------------------inline_native_jvm_commit------------------
3136 bool LibraryCallKit::inline_native_jvm_commit() {
3137 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3138
3139 // Save input memory and i_o state.
3140 Node* input_memory_state = reset_memory();
3141 set_all_memory(input_memory_state);
3142 Node* input_io_state = i_o();
3143
3144 // TLS.
3145 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3146 // Jfr java buffer.
3147 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3148 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3149 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3150
3151 // Load the current value of the notified field in the JfrThreadLocal.
3152 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3153 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3154
3155 // Test for notification.
3156 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3157 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3158 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3159
3160 // True branch, is notified.
3161 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3162 set_control(is_notified);
3163
3164 // Reset notified state.
3165 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3166 Node* notified_reset_memory = reset_memory();
3167
3168 // 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.
3169 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3170 // Convert the machine-word to a long.
3171 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3172
3173 // False branch, not notified.
3174 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3175 set_control(not_notified);
3176 set_all_memory(input_memory_state);
3177
3178 // Arg is the next position as a long.
3179 Node* arg = argument(0);
3180 // Convert long to machine-word.
3181 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3182
3183 // Store the next_position to the underlying jfr java buffer.
3184 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3185
3186 Node* commit_memory = reset_memory();
3187 set_all_memory(commit_memory);
3188
3189 // 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.
3190 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3191 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3192 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3193
3194 // And flags with lease constant.
3195 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3196
3197 // Branch on lease to conditionalize returning the leased java buffer.
3198 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3199 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3200 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3201
3202 // False branch, not a lease.
3203 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3204
3205 // True branch, is lease.
3206 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3207 set_control(is_lease);
3208
3209 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3210 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3211 OptoRuntime::void_void_Type(),
3212 SharedRuntime::jfr_return_lease(),
3213 "return_lease", TypePtr::BOTTOM);
3214 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3215
3216 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3217 record_for_igvn(lease_compare_rgn);
3218 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3219 record_for_igvn(lease_compare_mem);
3220 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3221 record_for_igvn(lease_compare_io);
3222 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3223 record_for_igvn(lease_result_value);
3224
3225 // Update control and phi nodes.
3226 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3227 lease_compare_rgn->init_req(_false_path, not_lease);
3228
3229 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3230 lease_compare_mem->init_req(_false_path, commit_memory);
3231
3232 lease_compare_io->init_req(_true_path, i_o());
3233 lease_compare_io->init_req(_false_path, input_io_state);
3234
3235 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3236 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3237
3238 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3239 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3240 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3241 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3242
3243 // Update control and phi nodes.
3244 result_rgn->init_req(_true_path, is_notified);
3245 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3246
3247 result_mem->init_req(_true_path, notified_reset_memory);
3248 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3249
3250 result_io->init_req(_true_path, input_io_state);
3251 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3252
3253 result_value->init_req(_true_path, current_pos);
3254 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3255
3256 // Set output state.
3257 set_control(_gvn.transform(result_rgn));
3258 set_all_memory(_gvn.transform(result_mem));
3259 set_i_o(_gvn.transform(result_io));
3260 set_result(result_rgn, result_value);
3261 return true;
3262 }
3263
3264 /*
3265 * The intrinsic is a model of this pseudo-code:
3266 *
3267 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3268 * jobject h_event_writer = tl->java_event_writer();
3269 * if (h_event_writer == nullptr) {
3270 * return nullptr;
3271 * }
3272 * oop threadObj = Thread::threadObj();
3273 * oop vthread = java_lang_Thread::vthread(threadObj);
3274 * traceid tid;
3275 * bool pinVirtualThread;
3276 * bool excluded;
3277 * if (vthread != threadObj) { // i.e. current thread is virtual
3278 * tid = java_lang_Thread::tid(vthread);
3279 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3280 * pinVirtualThread = VMContinuations;
3281 * excluded = vthread_epoch_raw & excluded_mask;
3282 * if (!excluded) {
3283 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3284 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3285 * if (vthread_epoch != current_epoch) {
3286 * write_checkpoint();
3287 * }
3288 * }
3289 * } else {
3290 * tid = java_lang_Thread::tid(threadObj);
3291 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3292 * pinVirtualThread = false;
3293 * excluded = thread_epoch_raw & excluded_mask;
3294 * }
3295 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3296 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3297 * if (tid_in_event_writer != tid) {
3298 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3299 * setField(event_writer, "excluded", excluded);
3300 * setField(event_writer, "threadID", tid);
3301 * }
3302 * return event_writer
3303 */
3304 bool LibraryCallKit::inline_native_getEventWriter() {
3305 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3306
3307 // Save input memory and i_o state.
3308 Node* input_memory_state = reset_memory();
3309 set_all_memory(input_memory_state);
3310 Node* input_io_state = i_o();
3311
3312 // The most significant bit of the u2 is used to denote thread exclusion
3313 Node* excluded_shift = _gvn.intcon(15);
3314 Node* excluded_mask = _gvn.intcon(1 << 15);
3315 // The epoch generation is the range [1-32767]
3316 Node* epoch_mask = _gvn.intcon(32767);
3317
3318 // TLS
3319 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3320
3321 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3322 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3323
3324 // Load the eventwriter jobject handle.
3325 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3326
3327 // Null check the jobject handle.
3328 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3329 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3330 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3331
3332 // False path, jobj is null.
3333 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3334
3335 // True path, jobj is not null.
3336 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3337
3338 set_control(jobj_is_not_null);
3339
3340 // Load the threadObj for the CarrierThread.
3341 Node* threadObj = generate_current_thread(tls_ptr);
3342
3343 // Load the vthread.
3344 Node* vthread = generate_virtual_thread(tls_ptr);
3345
3346 // If vthread != threadObj, this is a virtual thread.
3347 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3348 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3349 IfNode* iff_vthread_not_equal_threadObj =
3350 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3351
3352 // False branch, fallback to threadObj.
3353 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3354 set_control(vthread_equal_threadObj);
3355
3356 // Load the tid field from the vthread object.
3357 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3358
3359 // Load the raw epoch value from the threadObj.
3360 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3361 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3362 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3363 TypeInt::CHAR, T_CHAR,
3364 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3365
3366 // Mask off the excluded information from the epoch.
3367 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3368
3369 // True branch, this is a virtual thread.
3370 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3371 set_control(vthread_not_equal_threadObj);
3372
3373 // Load the tid field from the vthread object.
3374 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3375
3376 // Continuation support determines if a virtual thread should be pinned.
3377 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3378 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3379
3380 // Load the raw epoch value from the vthread.
3381 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3382 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3383 TypeInt::CHAR, T_CHAR,
3384 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3385
3386 // Mask off the excluded information from the epoch.
3387 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3388
3389 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3390 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3391 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3392 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3393
3394 // False branch, vthread is excluded, no need to write epoch info.
3395 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3396
3397 // True branch, vthread is included, update epoch info.
3398 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3399 set_control(included);
3400
3401 // Get epoch value.
3402 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3403
3404 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3405 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3406 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3407
3408 // Compare the epoch in the vthread to the current epoch generation.
3409 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3410 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3411 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3412
3413 // False path, epoch is equal, checkpoint information is valid.
3414 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3415
3416 // True path, epoch is not equal, write a checkpoint for the vthread.
3417 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3418
3419 set_control(epoch_is_not_equal);
3420
3421 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3422 // The call also updates the native thread local thread id and the vthread with the current epoch.
3423 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3424 OptoRuntime::jfr_write_checkpoint_Type(),
3425 SharedRuntime::jfr_write_checkpoint(),
3426 "write_checkpoint", TypePtr::BOTTOM);
3427 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3428
3429 // vthread epoch != current epoch
3430 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3431 record_for_igvn(epoch_compare_rgn);
3432 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3433 record_for_igvn(epoch_compare_mem);
3434 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3435 record_for_igvn(epoch_compare_io);
3436
3437 // Update control and phi nodes.
3438 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3439 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3440 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3441 epoch_compare_mem->init_req(_false_path, input_memory_state);
3442 epoch_compare_io->init_req(_true_path, i_o());
3443 epoch_compare_io->init_req(_false_path, input_io_state);
3444
3445 // excluded != true
3446 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3447 record_for_igvn(exclude_compare_rgn);
3448 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3449 record_for_igvn(exclude_compare_mem);
3450 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3451 record_for_igvn(exclude_compare_io);
3452
3453 // Update control and phi nodes.
3454 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3455 exclude_compare_rgn->init_req(_false_path, excluded);
3456 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3457 exclude_compare_mem->init_req(_false_path, input_memory_state);
3458 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3459 exclude_compare_io->init_req(_false_path, input_io_state);
3460
3461 // vthread != threadObj
3462 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3463 record_for_igvn(vthread_compare_rgn);
3464 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3465 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3466 record_for_igvn(vthread_compare_io);
3467 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3468 record_for_igvn(tid);
3469 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3470 record_for_igvn(exclusion);
3471 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3472 record_for_igvn(pinVirtualThread);
3473
3474 // Update control and phi nodes.
3475 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3476 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3477 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3478 vthread_compare_mem->init_req(_false_path, input_memory_state);
3479 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3480 vthread_compare_io->init_req(_false_path, input_io_state);
3481 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3482 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3483 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3484 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3485 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3486 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3487
3488 // Update branch state.
3489 set_control(_gvn.transform(vthread_compare_rgn));
3490 set_all_memory(_gvn.transform(vthread_compare_mem));
3491 set_i_o(_gvn.transform(vthread_compare_io));
3492
3493 // Load the event writer oop by dereferencing the jobject handle.
3494 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3495 assert(klass_EventWriter->is_loaded(), "invariant");
3496 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3497 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3498 const TypeOopPtr* const xtype = aklass->as_instance_type();
3499 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3500 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3501
3502 // Load the current thread id from the event writer object.
3503 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3504 // Get the field offset to, conditionally, store an updated tid value later.
3505 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3506 // Get the field offset to, conditionally, store an updated exclusion value later.
3507 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3508 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3509 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3510
3511 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3512 record_for_igvn(event_writer_tid_compare_rgn);
3513 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3514 record_for_igvn(event_writer_tid_compare_mem);
3515 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3516 record_for_igvn(event_writer_tid_compare_io);
3517
3518 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3519 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3520 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3521 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3522
3523 // False path, tids are the same.
3524 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3525
3526 // True path, tid is not equal, need to update the tid in the event writer.
3527 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3528 record_for_igvn(tid_is_not_equal);
3529
3530 // Store the pin state to the event writer.
3531 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3532
3533 // Store the exclusion state to the event writer.
3534 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3535 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3536
3537 // Store the tid to the event writer.
3538 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3539
3540 // Update control and phi nodes.
3541 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3542 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3543 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3544 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3545 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3546 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3547
3548 // Result of top level CFG, Memory, IO and Value.
3549 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3550 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3551 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3552 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3553
3554 // Result control.
3555 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3556 result_rgn->init_req(_false_path, jobj_is_null);
3557
3558 // Result memory.
3559 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3560 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3561
3562 // Result IO.
3563 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3564 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3565
3566 // Result value.
3567 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3568 result_value->init_req(_false_path, null()); // return null
3569
3570 // Set output state.
3571 set_control(_gvn.transform(result_rgn));
3572 set_all_memory(_gvn.transform(result_mem));
3573 set_i_o(_gvn.transform(result_io));
3574 set_result(result_rgn, result_value);
3575 return true;
3576 }
3577
3578 /*
3579 * The intrinsic is a model of this pseudo-code:
3580 *
3581 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3582 * if (carrierThread != thread) { // is virtual thread
3583 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3584 * bool excluded = vthread_epoch_raw & excluded_mask;
3585 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3586 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3587 * if (!excluded) {
3588 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3589 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3590 * }
3591 * Atomic::release_store(&tl->_vthread, true);
3592 * return;
3593 * }
3594 * Atomic::release_store(&tl->_vthread, false);
3595 */
3596 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3597 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3598
3599 Node* input_memory_state = reset_memory();
3600 set_all_memory(input_memory_state);
3601
3602 // The most significant bit of the u2 is used to denote thread exclusion
3603 Node* excluded_mask = _gvn.intcon(1 << 15);
3604 // The epoch generation is the range [1-32767]
3605 Node* epoch_mask = _gvn.intcon(32767);
3606
3607 Node* const carrierThread = generate_current_thread(jt);
3608 // If thread != carrierThread, this is a virtual thread.
3609 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3610 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3611 IfNode* iff_thread_not_equal_carrierThread =
3612 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3613
3614 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3615
3616 // False branch, is carrierThread.
3617 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3618 // Store release
3619 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3620
3621 set_all_memory(input_memory_state);
3622
3623 // True branch, is virtual thread.
3624 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3625 set_control(thread_not_equal_carrierThread);
3626
3627 // Load the raw epoch value from the vthread.
3628 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3629 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3630 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3631
3632 // Mask off the excluded information from the epoch.
3633 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3634
3635 // Load the tid field from the thread.
3636 Node* tid = load_field_from_object(thread, "tid", "J");
3637
3638 // Store the vthread tid to the jfr thread local.
3639 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3640 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3641
3642 // Branch is_excluded to conditionalize updating the epoch .
3643 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3644 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3645 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3646
3647 // True branch, vthread is excluded, no need to write epoch info.
3648 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3649 set_control(excluded);
3650 Node* vthread_is_excluded = _gvn.intcon(1);
3651
3652 // False branch, vthread is included, update epoch info.
3653 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3654 set_control(included);
3655 Node* vthread_is_included = _gvn.intcon(0);
3656
3657 // Get epoch value.
3658 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3659
3660 // Store the vthread epoch to the jfr thread local.
3661 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3662 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3663
3664 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3665 record_for_igvn(excluded_rgn);
3666 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3667 record_for_igvn(excluded_mem);
3668 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3669 record_for_igvn(exclusion);
3670
3671 // Merge the excluded control and memory.
3672 excluded_rgn->init_req(_true_path, excluded);
3673 excluded_rgn->init_req(_false_path, included);
3674 excluded_mem->init_req(_true_path, tid_memory);
3675 excluded_mem->init_req(_false_path, included_memory);
3676 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3677 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3678
3679 // Set intermediate state.
3680 set_control(_gvn.transform(excluded_rgn));
3681 set_all_memory(excluded_mem);
3682
3683 // Store the vthread exclusion state to the jfr thread local.
3684 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3685 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3686
3687 // Store release
3688 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3689
3690 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3691 record_for_igvn(thread_compare_rgn);
3692 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3693 record_for_igvn(thread_compare_mem);
3694 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3695 record_for_igvn(vthread);
3696
3697 // Merge the thread_compare control and memory.
3698 thread_compare_rgn->init_req(_true_path, control());
3699 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3700 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3701 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3702
3703 // Set output state.
3704 set_control(_gvn.transform(thread_compare_rgn));
3705 set_all_memory(_gvn.transform(thread_compare_mem));
3706 }
3707
3708 #endif // JFR_HAVE_INTRINSICS
3709
3710 //------------------------inline_native_currentCarrierThread------------------
3711 bool LibraryCallKit::inline_native_currentCarrierThread() {
3712 Node* junk = nullptr;
3713 set_result(generate_current_thread(junk));
3714 return true;
3715 }
3716
3717 //------------------------inline_native_currentThread------------------
3718 bool LibraryCallKit::inline_native_currentThread() {
3719 Node* junk = nullptr;
3720 set_result(generate_virtual_thread(junk));
3721 return true;
3722 }
3723
3724 //------------------------inline_native_setVthread------------------
3725 bool LibraryCallKit::inline_native_setCurrentThread() {
3726 assert(C->method()->changes_current_thread(),
3727 "method changes current Thread but is not annotated ChangesCurrentThread");
3728 Node* arr = argument(1);
3729 Node* thread = _gvn.transform(new ThreadLocalNode());
3730 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3731 Node* thread_obj_handle
3732 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3733 thread_obj_handle = _gvn.transform(thread_obj_handle);
3734 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3735 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3736
3737 // Change the _monitor_owner_id of the JavaThread
3738 Node* tid = load_field_from_object(arr, "tid", "J");
3739 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3740 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3741
3742 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3743 return true;
3744 }
3745
3746 const Type* LibraryCallKit::scopedValueCache_type() {
3747 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3748 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3749 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3750
3751 // Because we create the scopedValue cache lazily we have to make the
3752 // type of the result BotPTR.
3753 bool xk = etype->klass_is_exact();
3754 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3755 return objects_type;
3756 }
3757
3758 Node* LibraryCallKit::scopedValueCache_helper() {
3759 Node* thread = _gvn.transform(new ThreadLocalNode());
3760 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3761 // We cannot use immutable_memory() because we might flip onto a
3762 // different carrier thread, at which point we'll need to use that
3763 // carrier thread's cache.
3764 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3765 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3766 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3767 }
3768
3769 //------------------------inline_native_scopedValueCache------------------
3770 bool LibraryCallKit::inline_native_scopedValueCache() {
3771 Node* cache_obj_handle = scopedValueCache_helper();
3772 const Type* objects_type = scopedValueCache_type();
3773 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3774
3775 return true;
3776 }
3777
3778 //------------------------inline_native_setScopedValueCache------------------
3779 bool LibraryCallKit::inline_native_setScopedValueCache() {
3780 Node* arr = argument(0);
3781 Node* cache_obj_handle = scopedValueCache_helper();
3782 const Type* objects_type = scopedValueCache_type();
3783
3784 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3785 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3786
3787 return true;
3788 }
3789
3790 //------------------------inline_native_Continuation_pin and unpin-----------
3791
3792 // Shared implementation routine for both pin and unpin.
3793 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3794 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3795
3796 // Save input memory.
3797 Node* input_memory_state = reset_memory();
3798 set_all_memory(input_memory_state);
3799
3800 // TLS
3801 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3802 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3803 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3804
3805 // Null check the last continuation object.
3806 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3807 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3808 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3809
3810 // False path, last continuation is null.
3811 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3812
3813 // True path, last continuation is not null.
3814 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3815
3816 set_control(continuation_is_not_null);
3817
3818 // Load the pin count from the last continuation.
3819 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3820 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3821
3822 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3823 Node* pin_count_rhs;
3824 if (unpin) {
3825 pin_count_rhs = _gvn.intcon(0);
3826 } else {
3827 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3828 }
3829 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3830 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3831 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3832
3833 // True branch, pin count over/underflow.
3834 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3835 {
3836 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3837 // which will throw IllegalStateException for pin count over/underflow.
3838 // No memory changed so far - we can use memory create by reset_memory()
3839 // at the beginning of this intrinsic. No need to call reset_memory() again.
3840 PreserveJVMState pjvms(this);
3841 set_control(pin_count_over_underflow);
3842 uncommon_trap(Deoptimization::Reason_intrinsic,
3843 Deoptimization::Action_none);
3844 assert(stopped(), "invariant");
3845 }
3846
3847 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3848 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3849 set_control(valid_pin_count);
3850
3851 Node* next_pin_count;
3852 if (unpin) {
3853 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3854 } else {
3855 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3856 }
3857
3858 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3859
3860 // Result of top level CFG and Memory.
3861 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3862 record_for_igvn(result_rgn);
3863 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3864 record_for_igvn(result_mem);
3865
3866 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3867 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3868 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3869 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3870
3871 // Set output state.
3872 set_control(_gvn.transform(result_rgn));
3873 set_all_memory(_gvn.transform(result_mem));
3874
3875 return true;
3876 }
3877
3878 //---------------------------load_mirror_from_klass----------------------------
3879 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3880 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3881 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3882 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3883 // mirror = ((OopHandle)mirror)->resolve();
3884 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3885 }
3886
3887 //-----------------------load_klass_from_mirror_common-------------------------
3888 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3889 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3890 // and branch to the given path on the region.
3891 // If never_see_null, take an uncommon trap on null, so we can optimistically
3892 // compile for the non-null case.
3893 // If the region is null, force never_see_null = true.
3894 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3895 bool never_see_null,
3896 RegionNode* region,
3897 int null_path,
3898 int offset) {
3899 if (region == nullptr) never_see_null = true;
3900 Node* p = basic_plus_adr(mirror, offset);
3901 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3902 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3903 Node* null_ctl = top();
3904 kls = null_check_oop(kls, &null_ctl, never_see_null);
3905 if (region != nullptr) {
3906 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3907 region->init_req(null_path, null_ctl);
3908 } else {
3909 assert(null_ctl == top(), "no loose ends");
3910 }
3911 return kls;
3912 }
3913
3914 //--------------------(inline_native_Class_query helpers)---------------------
3915 // Use this for JVM_ACC_INTERFACE.
3916 // Fall through if (mods & mask) == bits, take the guard otherwise.
3917 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3918 ByteSize offset, const Type* type, BasicType bt) {
3919 // Branch around if the given klass has the given modifier bit set.
3920 // Like generate_guard, adds a new path onto the region.
3921 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3922 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3923 Node* mask = intcon(modifier_mask);
3924 Node* bits = intcon(modifier_bits);
3925 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3926 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3927 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3928 return generate_fair_guard(bol, region);
3929 }
3930 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3931 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3932 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3933 }
3934
3935 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3936 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3937 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3938 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3939 }
3940
3941 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3942 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3943 }
3944
3945 //-------------------------inline_native_Class_query-------------------
3946 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3947 const Type* return_type = TypeInt::BOOL;
3948 Node* prim_return_value = top(); // what happens if it's a primitive class?
3949 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3950 bool expect_prim = false; // most of these guys expect to work on refs
3951
3952 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3953
3954 Node* mirror = argument(0);
3955 Node* obj = top();
3956
3957 switch (id) {
3958 case vmIntrinsics::_isInstance:
3959 // nothing is an instance of a primitive type
3960 prim_return_value = intcon(0);
3961 obj = argument(1);
3962 break;
3963 case vmIntrinsics::_isHidden:
3964 prim_return_value = intcon(0);
3965 break;
3966 case vmIntrinsics::_getSuperclass:
3967 prim_return_value = null();
3968 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3969 break;
3970 case vmIntrinsics::_getClassAccessFlags:
3971 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3972 return_type = TypeInt::CHAR;
3973 break;
3974 default:
3975 fatal_unexpected_iid(id);
3976 break;
3977 }
3978
3979 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3980 if (mirror_con == nullptr) return false; // cannot happen?
3981
3982 #ifndef PRODUCT
3983 if (C->print_intrinsics() || C->print_inlining()) {
3984 ciType* k = mirror_con->java_mirror_type();
3985 if (k) {
3986 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3987 k->print_name();
3988 tty->cr();
3989 }
3990 }
3991 #endif
3992
3993 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3994 RegionNode* region = new RegionNode(PATH_LIMIT);
3995 record_for_igvn(region);
3996 PhiNode* phi = new PhiNode(region, return_type);
3997
3998 // The mirror will never be null of Reflection.getClassAccessFlags, however
3999 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4000 // if it is. See bug 4774291.
4001
4002 // For Reflection.getClassAccessFlags(), the null check occurs in
4003 // the wrong place; see inline_unsafe_access(), above, for a similar
4004 // situation.
4005 mirror = null_check(mirror);
4006 // If mirror or obj is dead, only null-path is taken.
4007 if (stopped()) return true;
4008
4009 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4010
4011 // Now load the mirror's klass metaobject, and null-check it.
4012 // Side-effects region with the control path if the klass is null.
4013 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4014 // If kls is null, we have a primitive mirror.
4015 phi->init_req(_prim_path, prim_return_value);
4016 if (stopped()) { set_result(region, phi); return true; }
4017 bool safe_for_replace = (region->in(_prim_path) == top());
4018
4019 Node* p; // handy temp
4020 Node* null_ctl;
4021
4022 // Now that we have the non-null klass, we can perform the real query.
4023 // For constant classes, the query will constant-fold in LoadNode::Value.
4024 Node* query_value = top();
4025 switch (id) {
4026 case vmIntrinsics::_isInstance:
4027 // nothing is an instance of a primitive type
4028 query_value = gen_instanceof(obj, kls, safe_for_replace);
4029 break;
4030
4031 case vmIntrinsics::_isHidden:
4032 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4033 if (generate_hidden_class_guard(kls, region) != nullptr)
4034 // A guard was added. If the guard is taken, it was an hidden class.
4035 phi->add_req(intcon(1));
4036 // If we fall through, it's a plain class.
4037 query_value = intcon(0);
4038 break;
4039
4040
4041 case vmIntrinsics::_getSuperclass:
4042 // The rules here are somewhat unfortunate, but we can still do better
4043 // with random logic than with a JNI call.
4044 // Interfaces store null or Object as _super, but must report null.
4045 // Arrays store an intermediate super as _super, but must report Object.
4046 // Other types can report the actual _super.
4047 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4048 if (generate_interface_guard(kls, region) != nullptr)
4049 // A guard was added. If the guard is taken, it was an interface.
4050 phi->add_req(null());
4051 if (generate_array_guard(kls, region) != nullptr)
4052 // A guard was added. If the guard is taken, it was an array.
4053 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4054 // If we fall through, it's a plain class. Get its _super.
4055 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4056 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4057 null_ctl = top();
4058 kls = null_check_oop(kls, &null_ctl);
4059 if (null_ctl != top()) {
4060 // If the guard is taken, Object.superClass is null (both klass and mirror).
4061 region->add_req(null_ctl);
4062 phi ->add_req(null());
4063 }
4064 if (!stopped()) {
4065 query_value = load_mirror_from_klass(kls);
4066 }
4067 break;
4068
4069 case vmIntrinsics::_getClassAccessFlags:
4070 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4071 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4072 break;
4073
4074 default:
4075 fatal_unexpected_iid(id);
4076 break;
4077 }
4078
4079 // Fall-through is the normal case of a query to a real class.
4080 phi->init_req(1, query_value);
4081 region->init_req(1, control());
4082
4083 C->set_has_split_ifs(true); // Has chance for split-if optimization
4084 set_result(region, phi);
4085 return true;
4086 }
4087
4088 //-------------------------inline_Class_cast-------------------
4089 bool LibraryCallKit::inline_Class_cast() {
4090 Node* mirror = argument(0); // Class
4091 Node* obj = argument(1);
4092 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4093 if (mirror_con == nullptr) {
4094 return false; // dead path (mirror->is_top()).
4095 }
4096 if (obj == nullptr || obj->is_top()) {
4097 return false; // dead path
4098 }
4099 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4100
4101 // First, see if Class.cast() can be folded statically.
4102 // java_mirror_type() returns non-null for compile-time Class constants.
4103 ciType* tm = mirror_con->java_mirror_type();
4104 if (tm != nullptr && tm->is_klass() &&
4105 tp != nullptr) {
4106 if (!tp->is_loaded()) {
4107 // Don't use intrinsic when class is not loaded.
4108 return false;
4109 } else {
4110 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4111 if (static_res == Compile::SSC_always_true) {
4112 // isInstance() is true - fold the code.
4113 set_result(obj);
4114 return true;
4115 } else if (static_res == Compile::SSC_always_false) {
4116 // Don't use intrinsic, have to throw ClassCastException.
4117 // If the reference is null, the non-intrinsic bytecode will
4118 // be optimized appropriately.
4119 return false;
4120 }
4121 }
4122 }
4123
4124 // Bailout intrinsic and do normal inlining if exception path is frequent.
4125 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4126 return false;
4127 }
4128
4129 // Generate dynamic checks.
4130 // Class.cast() is java implementation of _checkcast bytecode.
4131 // Do checkcast (Parse::do_checkcast()) optimizations here.
4132
4133 mirror = null_check(mirror);
4134 // If mirror is dead, only null-path is taken.
4135 if (stopped()) {
4136 return true;
4137 }
4138
4139 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4140 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4141 RegionNode* region = new RegionNode(PATH_LIMIT);
4142 record_for_igvn(region);
4143
4144 // Now load the mirror's klass metaobject, and null-check it.
4145 // If kls is null, we have a primitive mirror and
4146 // nothing is an instance of a primitive type.
4147 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4148
4149 Node* res = top();
4150 if (!stopped()) {
4151 Node* bad_type_ctrl = top();
4152 // Do checkcast optimizations.
4153 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4154 region->init_req(_bad_type_path, bad_type_ctrl);
4155 }
4156 if (region->in(_prim_path) != top() ||
4157 region->in(_bad_type_path) != top()) {
4158 // Let Interpreter throw ClassCastException.
4159 PreserveJVMState pjvms(this);
4160 set_control(_gvn.transform(region));
4161 uncommon_trap(Deoptimization::Reason_intrinsic,
4162 Deoptimization::Action_maybe_recompile);
4163 }
4164 if (!stopped()) {
4165 set_result(res);
4166 }
4167 return true;
4168 }
4169
4170
4171 //--------------------------inline_native_subtype_check------------------------
4172 // This intrinsic takes the JNI calls out of the heart of
4173 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4174 bool LibraryCallKit::inline_native_subtype_check() {
4175 // Pull both arguments off the stack.
4176 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4177 args[0] = argument(0);
4178 args[1] = argument(1);
4179 Node* klasses[2]; // corresponding Klasses: superk, subk
4180 klasses[0] = klasses[1] = top();
4181
4182 enum {
4183 // A full decision tree on {superc is prim, subc is prim}:
4184 _prim_0_path = 1, // {P,N} => false
4185 // {P,P} & superc!=subc => false
4186 _prim_same_path, // {P,P} & superc==subc => true
4187 _prim_1_path, // {N,P} => false
4188 _ref_subtype_path, // {N,N} & subtype check wins => true
4189 _both_ref_path, // {N,N} & subtype check loses => false
4190 PATH_LIMIT
4191 };
4192
4193 RegionNode* region = new RegionNode(PATH_LIMIT);
4194 Node* phi = new PhiNode(region, TypeInt::BOOL);
4195 record_for_igvn(region);
4196
4197 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4198 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4199 int class_klass_offset = java_lang_Class::klass_offset();
4200
4201 // First null-check both mirrors and load each mirror's klass metaobject.
4202 int which_arg;
4203 for (which_arg = 0; which_arg <= 1; which_arg++) {
4204 Node* arg = args[which_arg];
4205 arg = null_check(arg);
4206 if (stopped()) break;
4207 args[which_arg] = arg;
4208
4209 Node* p = basic_plus_adr(arg, class_klass_offset);
4210 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4211 klasses[which_arg] = _gvn.transform(kls);
4212 }
4213
4214 // Having loaded both klasses, test each for null.
4215 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4216 for (which_arg = 0; which_arg <= 1; which_arg++) {
4217 Node* kls = klasses[which_arg];
4218 Node* null_ctl = top();
4219 kls = null_check_oop(kls, &null_ctl, never_see_null);
4220 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4221 region->init_req(prim_path, null_ctl);
4222 if (stopped()) break;
4223 klasses[which_arg] = kls;
4224 }
4225
4226 if (!stopped()) {
4227 // now we have two reference types, in klasses[0..1]
4228 Node* subk = klasses[1]; // the argument to isAssignableFrom
4229 Node* superk = klasses[0]; // the receiver
4230 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4231 // now we have a successful reference subtype check
4232 region->set_req(_ref_subtype_path, control());
4233 }
4234
4235 // If both operands are primitive (both klasses null), then
4236 // we must return true when they are identical primitives.
4237 // It is convenient to test this after the first null klass check.
4238 set_control(region->in(_prim_0_path)); // go back to first null check
4239 if (!stopped()) {
4240 // Since superc is primitive, make a guard for the superc==subc case.
4241 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4242 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4243 generate_guard(bol_eq, region, PROB_FAIR);
4244 if (region->req() == PATH_LIMIT+1) {
4245 // A guard was added. If the added guard is taken, superc==subc.
4246 region->swap_edges(PATH_LIMIT, _prim_same_path);
4247 region->del_req(PATH_LIMIT);
4248 }
4249 region->set_req(_prim_0_path, control()); // Not equal after all.
4250 }
4251
4252 // these are the only paths that produce 'true':
4253 phi->set_req(_prim_same_path, intcon(1));
4254 phi->set_req(_ref_subtype_path, intcon(1));
4255
4256 // pull together the cases:
4257 assert(region->req() == PATH_LIMIT, "sane region");
4258 for (uint i = 1; i < region->req(); i++) {
4259 Node* ctl = region->in(i);
4260 if (ctl == nullptr || ctl == top()) {
4261 region->set_req(i, top());
4262 phi ->set_req(i, top());
4263 } else if (phi->in(i) == nullptr) {
4264 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4265 }
4266 }
4267
4268 set_control(_gvn.transform(region));
4269 set_result(_gvn.transform(phi));
4270 return true;
4271 }
4272
4273 //---------------------generate_array_guard_common------------------------
4274 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4275 bool obj_array, bool not_array, Node** obj) {
4276
4277 if (stopped()) {
4278 return nullptr;
4279 }
4280
4281 // If obj_array/non_array==false/false:
4282 // Branch around if the given klass is in fact an array (either obj or prim).
4283 // If obj_array/non_array==false/true:
4284 // Branch around if the given klass is not an array klass of any kind.
4285 // If obj_array/non_array==true/true:
4286 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4287 // If obj_array/non_array==true/false:
4288 // Branch around if the kls is an oop array (Object[] or subtype)
4289 //
4290 // Like generate_guard, adds a new path onto the region.
4291 jint layout_con = 0;
4292 Node* layout_val = get_layout_helper(kls, layout_con);
4293 if (layout_val == nullptr) {
4294 bool query = (obj_array
4295 ? Klass::layout_helper_is_objArray(layout_con)
4296 : Klass::layout_helper_is_array(layout_con));
4297 if (query == not_array) {
4298 return nullptr; // never a branch
4299 } else { // always a branch
4300 Node* always_branch = control();
4301 if (region != nullptr)
4302 region->add_req(always_branch);
4303 set_control(top());
4304 return always_branch;
4305 }
4306 }
4307 // Now test the correct condition.
4308 jint nval = (obj_array
4309 ? (jint)(Klass::_lh_array_tag_type_value
4310 << Klass::_lh_array_tag_shift)
4311 : Klass::_lh_neutral_value);
4312 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4313 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4314 // invert the test if we are looking for a non-array
4315 if (not_array) btest = BoolTest(btest).negate();
4316 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4317 Node* ctrl = generate_fair_guard(bol, region);
4318 Node* is_array_ctrl = not_array ? control() : ctrl;
4319 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4320 // Keep track of the fact that 'obj' is an array to prevent
4321 // array specific accesses from floating above the guard.
4322 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4323 }
4324 return ctrl;
4325 }
4326
4327
4328 //-----------------------inline_native_newArray--------------------------
4329 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4330 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4331 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4332 Node* mirror;
4333 Node* count_val;
4334 if (uninitialized) {
4335 null_check_receiver();
4336 mirror = argument(1);
4337 count_val = argument(2);
4338 } else {
4339 mirror = argument(0);
4340 count_val = argument(1);
4341 }
4342
4343 mirror = null_check(mirror);
4344 // If mirror or obj is dead, only null-path is taken.
4345 if (stopped()) return true;
4346
4347 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4348 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4349 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4350 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4351 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4352
4353 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4354 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4355 result_reg, _slow_path);
4356 Node* normal_ctl = control();
4357 Node* no_array_ctl = result_reg->in(_slow_path);
4358
4359 // Generate code for the slow case. We make a call to newArray().
4360 set_control(no_array_ctl);
4361 if (!stopped()) {
4362 // Either the input type is void.class, or else the
4363 // array klass has not yet been cached. Either the
4364 // ensuing call will throw an exception, or else it
4365 // will cache the array klass for next time.
4366 PreserveJVMState pjvms(this);
4367 CallJavaNode* slow_call = nullptr;
4368 if (uninitialized) {
4369 // Generate optimized virtual call (holder class 'Unsafe' is final)
4370 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4371 } else {
4372 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4373 }
4374 Node* slow_result = set_results_for_java_call(slow_call);
4375 // this->control() comes from set_results_for_java_call
4376 result_reg->set_req(_slow_path, control());
4377 result_val->set_req(_slow_path, slow_result);
4378 result_io ->set_req(_slow_path, i_o());
4379 result_mem->set_req(_slow_path, reset_memory());
4380 }
4381
4382 set_control(normal_ctl);
4383 if (!stopped()) {
4384 // Normal case: The array type has been cached in the java.lang.Class.
4385 // The following call works fine even if the array type is polymorphic.
4386 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4387 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4388 result_reg->init_req(_normal_path, control());
4389 result_val->init_req(_normal_path, obj);
4390 result_io ->init_req(_normal_path, i_o());
4391 result_mem->init_req(_normal_path, reset_memory());
4392
4393 if (uninitialized) {
4394 // Mark the allocation so that zeroing is skipped
4395 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4396 alloc->maybe_set_complete(&_gvn);
4397 }
4398 }
4399
4400 // Return the combined state.
4401 set_i_o( _gvn.transform(result_io) );
4402 set_all_memory( _gvn.transform(result_mem));
4403
4404 C->set_has_split_ifs(true); // Has chance for split-if optimization
4405 set_result(result_reg, result_val);
4406 return true;
4407 }
4408
4409 //----------------------inline_native_getLength--------------------------
4410 // public static native int java.lang.reflect.Array.getLength(Object array);
4411 bool LibraryCallKit::inline_native_getLength() {
4412 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4413
4414 Node* array = null_check(argument(0));
4415 // If array is dead, only null-path is taken.
4416 if (stopped()) return true;
4417
4418 // Deoptimize if it is a non-array.
4419 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4420
4421 if (non_array != nullptr) {
4422 PreserveJVMState pjvms(this);
4423 set_control(non_array);
4424 uncommon_trap(Deoptimization::Reason_intrinsic,
4425 Deoptimization::Action_maybe_recompile);
4426 }
4427
4428 // If control is dead, only non-array-path is taken.
4429 if (stopped()) return true;
4430
4431 // The works fine even if the array type is polymorphic.
4432 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4433 Node* result = load_array_length(array);
4434
4435 C->set_has_split_ifs(true); // Has chance for split-if optimization
4436 set_result(result);
4437 return true;
4438 }
4439
4440 //------------------------inline_array_copyOf----------------------------
4441 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4442 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4443 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4444 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4445
4446 // Get the arguments.
4447 Node* original = argument(0);
4448 Node* start = is_copyOfRange? argument(1): intcon(0);
4449 Node* end = is_copyOfRange? argument(2): argument(1);
4450 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4451
4452 Node* newcopy = nullptr;
4453
4454 // Set the original stack and the reexecute bit for the interpreter to reexecute
4455 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4456 { PreserveReexecuteState preexecs(this);
4457 jvms()->set_should_reexecute(true);
4458
4459 array_type_mirror = null_check(array_type_mirror);
4460 original = null_check(original);
4461
4462 // Check if a null path was taken unconditionally.
4463 if (stopped()) return true;
4464
4465 Node* orig_length = load_array_length(original);
4466
4467 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4468 klass_node = null_check(klass_node);
4469
4470 RegionNode* bailout = new RegionNode(1);
4471 record_for_igvn(bailout);
4472
4473 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4474 // Bail out if that is so.
4475 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4476 if (not_objArray != nullptr) {
4477 // Improve the klass node's type from the new optimistic assumption:
4478 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4479 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4480 Node* cast = new CastPPNode(control(), klass_node, akls);
4481 klass_node = _gvn.transform(cast);
4482 }
4483
4484 // Bail out if either start or end is negative.
4485 generate_negative_guard(start, bailout, &start);
4486 generate_negative_guard(end, bailout, &end);
4487
4488 Node* length = end;
4489 if (_gvn.type(start) != TypeInt::ZERO) {
4490 length = _gvn.transform(new SubINode(end, start));
4491 }
4492
4493 // Bail out if length is negative (i.e., if start > end).
4494 // Without this the new_array would throw
4495 // NegativeArraySizeException but IllegalArgumentException is what
4496 // should be thrown
4497 generate_negative_guard(length, bailout, &length);
4498
4499 // Bail out if start is larger than the original length
4500 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4501 generate_negative_guard(orig_tail, bailout, &orig_tail);
4502
4503 if (bailout->req() > 1) {
4504 PreserveJVMState pjvms(this);
4505 set_control(_gvn.transform(bailout));
4506 uncommon_trap(Deoptimization::Reason_intrinsic,
4507 Deoptimization::Action_maybe_recompile);
4508 }
4509
4510 if (!stopped()) {
4511 // How many elements will we copy from the original?
4512 // The answer is MinI(orig_tail, length).
4513 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4514
4515 // Generate a direct call to the right arraycopy function(s).
4516 // We know the copy is disjoint but we might not know if the
4517 // oop stores need checking.
4518 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4519 // This will fail a store-check if x contains any non-nulls.
4520
4521 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4522 // loads/stores but it is legal only if we're sure the
4523 // Arrays.copyOf would succeed. So we need all input arguments
4524 // to the copyOf to be validated, including that the copy to the
4525 // new array won't trigger an ArrayStoreException. That subtype
4526 // check can be optimized if we know something on the type of
4527 // the input array from type speculation.
4528 if (_gvn.type(klass_node)->singleton()) {
4529 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4530 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4531
4532 int test = C->static_subtype_check(superk, subk);
4533 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4534 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4535 if (t_original->speculative_type() != nullptr) {
4536 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4537 }
4538 }
4539 }
4540
4541 bool validated = false;
4542 // Reason_class_check rather than Reason_intrinsic because we
4543 // want to intrinsify even if this traps.
4544 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4545 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4546
4547 if (not_subtype_ctrl != top()) {
4548 PreserveJVMState pjvms(this);
4549 set_control(not_subtype_ctrl);
4550 uncommon_trap(Deoptimization::Reason_class_check,
4551 Deoptimization::Action_make_not_entrant);
4552 assert(stopped(), "Should be stopped");
4553 }
4554 validated = true;
4555 }
4556
4557 if (!stopped()) {
4558 newcopy = new_array(klass_node, length, 0); // no arguments to push
4559
4560 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4561 load_object_klass(original), klass_node);
4562 if (!is_copyOfRange) {
4563 ac->set_copyof(validated);
4564 } else {
4565 ac->set_copyofrange(validated);
4566 }
4567 Node* n = _gvn.transform(ac);
4568 if (n == ac) {
4569 ac->connect_outputs(this);
4570 } else {
4571 assert(validated, "shouldn't transform if all arguments not validated");
4572 set_all_memory(n);
4573 }
4574 }
4575 }
4576 } // original reexecute is set back here
4577
4578 C->set_has_split_ifs(true); // Has chance for split-if optimization
4579 if (!stopped()) {
4580 set_result(newcopy);
4581 }
4582 return true;
4583 }
4584
4585
4586 //----------------------generate_virtual_guard---------------------------
4587 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4588 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4589 RegionNode* slow_region) {
4590 ciMethod* method = callee();
4591 int vtable_index = method->vtable_index();
4592 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4593 "bad index %d", vtable_index);
4594 // Get the Method* out of the appropriate vtable entry.
4595 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4596 vtable_index*vtableEntry::size_in_bytes() +
4597 in_bytes(vtableEntry::method_offset());
4598 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4599 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4600
4601 // Compare the target method with the expected method (e.g., Object.hashCode).
4602 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4603
4604 Node* native_call = makecon(native_call_addr);
4605 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4606 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4607
4608 return generate_slow_guard(test_native, slow_region);
4609 }
4610
4611 //-----------------------generate_method_call----------------------------
4612 // Use generate_method_call to make a slow-call to the real
4613 // method if the fast path fails. An alternative would be to
4614 // use a stub like OptoRuntime::slow_arraycopy_Java.
4615 // This only works for expanding the current library call,
4616 // not another intrinsic. (E.g., don't use this for making an
4617 // arraycopy call inside of the copyOf intrinsic.)
4618 CallJavaNode*
4619 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4620 // When compiling the intrinsic method itself, do not use this technique.
4621 guarantee(callee() != C->method(), "cannot make slow-call to self");
4622
4623 ciMethod* method = callee();
4624 // ensure the JVMS we have will be correct for this call
4625 guarantee(method_id == method->intrinsic_id(), "must match");
4626
4627 const TypeFunc* tf = TypeFunc::make(method);
4628 if (res_not_null) {
4629 assert(tf->return_type() == T_OBJECT, "");
4630 const TypeTuple* range = tf->range();
4631 const Type** fields = TypeTuple::fields(range->cnt());
4632 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4633 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4634 tf = TypeFunc::make(tf->domain(), new_range);
4635 }
4636 CallJavaNode* slow_call;
4637 if (is_static) {
4638 assert(!is_virtual, "");
4639 slow_call = new CallStaticJavaNode(C, tf,
4640 SharedRuntime::get_resolve_static_call_stub(), method);
4641 } else if (is_virtual) {
4642 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4643 int vtable_index = Method::invalid_vtable_index;
4644 if (UseInlineCaches) {
4645 // Suppress the vtable call
4646 } else {
4647 // hashCode and clone are not a miranda methods,
4648 // so the vtable index is fixed.
4649 // No need to use the linkResolver to get it.
4650 vtable_index = method->vtable_index();
4651 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4652 "bad index %d", vtable_index);
4653 }
4654 slow_call = new CallDynamicJavaNode(tf,
4655 SharedRuntime::get_resolve_virtual_call_stub(),
4656 method, vtable_index);
4657 } else { // neither virtual nor static: opt_virtual
4658 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4659 slow_call = new CallStaticJavaNode(C, tf,
4660 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4661 slow_call->set_optimized_virtual(true);
4662 }
4663 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4664 // To be able to issue a direct call (optimized virtual or virtual)
4665 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4666 // about the method being invoked should be attached to the call site to
4667 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4668 slow_call->set_override_symbolic_info(true);
4669 }
4670 set_arguments_for_java_call(slow_call);
4671 set_edges_for_java_call(slow_call);
4672 return slow_call;
4673 }
4674
4675
4676 /**
4677 * Build special case code for calls to hashCode on an object. This call may
4678 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4679 * slightly different code.
4680 */
4681 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4682 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4683 assert(!(is_virtual && is_static), "either virtual, special, or static");
4684
4685 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4686
4687 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4688 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4689 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4690 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4691 Node* obj = nullptr;
4692 if (!is_static) {
4693 // Check for hashing null object
4694 obj = null_check_receiver();
4695 if (stopped()) return true; // unconditionally null
4696 result_reg->init_req(_null_path, top());
4697 result_val->init_req(_null_path, top());
4698 } else {
4699 // Do a null check, and return zero if null.
4700 // System.identityHashCode(null) == 0
4701 obj = argument(0);
4702 Node* null_ctl = top();
4703 obj = null_check_oop(obj, &null_ctl);
4704 result_reg->init_req(_null_path, null_ctl);
4705 result_val->init_req(_null_path, _gvn.intcon(0));
4706 }
4707
4708 // Unconditionally null? Then return right away.
4709 if (stopped()) {
4710 set_control( result_reg->in(_null_path));
4711 if (!stopped())
4712 set_result(result_val->in(_null_path));
4713 return true;
4714 }
4715
4716 // We only go to the fast case code if we pass a number of guards. The
4717 // paths which do not pass are accumulated in the slow_region.
4718 RegionNode* slow_region = new RegionNode(1);
4719 record_for_igvn(slow_region);
4720
4721 // If this is a virtual call, we generate a funny guard. We pull out
4722 // the vtable entry corresponding to hashCode() from the target object.
4723 // If the target method which we are calling happens to be the native
4724 // Object hashCode() method, we pass the guard. We do not need this
4725 // guard for non-virtual calls -- the caller is known to be the native
4726 // Object hashCode().
4727 if (is_virtual) {
4728 // After null check, get the object's klass.
4729 Node* obj_klass = load_object_klass(obj);
4730 generate_virtual_guard(obj_klass, slow_region);
4731 }
4732
4733 // Get the header out of the object, use LoadMarkNode when available
4734 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4735 // The control of the load must be null. Otherwise, the load can move before
4736 // the null check after castPP removal.
4737 Node* no_ctrl = nullptr;
4738 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4739
4740 if (!UseObjectMonitorTable) {
4741 // Test the header to see if it is safe to read w.r.t. locking.
4742 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4743 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4744 if (LockingMode == LM_LIGHTWEIGHT) {
4745 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4746 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4747 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4748
4749 generate_slow_guard(test_monitor, slow_region);
4750 } else {
4751 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4752 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4753 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4754
4755 generate_slow_guard(test_not_unlocked, slow_region);
4756 }
4757 }
4758
4759 // Get the hash value and check to see that it has been properly assigned.
4760 // We depend on hash_mask being at most 32 bits and avoid the use of
4761 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4762 // vm: see markWord.hpp.
4763 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4764 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4765 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4766 // This hack lets the hash bits live anywhere in the mark object now, as long
4767 // as the shift drops the relevant bits into the low 32 bits. Note that
4768 // Java spec says that HashCode is an int so there's no point in capturing
4769 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4770 hshifted_header = ConvX2I(hshifted_header);
4771 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4772
4773 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4774 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4775 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4776
4777 generate_slow_guard(test_assigned, slow_region);
4778
4779 Node* init_mem = reset_memory();
4780 // fill in the rest of the null path:
4781 result_io ->init_req(_null_path, i_o());
4782 result_mem->init_req(_null_path, init_mem);
4783
4784 result_val->init_req(_fast_path, hash_val);
4785 result_reg->init_req(_fast_path, control());
4786 result_io ->init_req(_fast_path, i_o());
4787 result_mem->init_req(_fast_path, init_mem);
4788
4789 // Generate code for the slow case. We make a call to hashCode().
4790 set_control(_gvn.transform(slow_region));
4791 if (!stopped()) {
4792 // No need for PreserveJVMState, because we're using up the present state.
4793 set_all_memory(init_mem);
4794 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4795 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4796 Node* slow_result = set_results_for_java_call(slow_call);
4797 // this->control() comes from set_results_for_java_call
4798 result_reg->init_req(_slow_path, control());
4799 result_val->init_req(_slow_path, slow_result);
4800 result_io ->set_req(_slow_path, i_o());
4801 result_mem ->set_req(_slow_path, reset_memory());
4802 }
4803
4804 // Return the combined state.
4805 set_i_o( _gvn.transform(result_io) );
4806 set_all_memory( _gvn.transform(result_mem));
4807
4808 set_result(result_reg, result_val);
4809 return true;
4810 }
4811
4812 //---------------------------inline_native_getClass----------------------------
4813 // public final native Class<?> java.lang.Object.getClass();
4814 //
4815 // Build special case code for calls to getClass on an object.
4816 bool LibraryCallKit::inline_native_getClass() {
4817 Node* obj = null_check_receiver();
4818 if (stopped()) return true;
4819 set_result(load_mirror_from_klass(load_object_klass(obj)));
4820 return true;
4821 }
4822
4823 //-----------------inline_native_Reflection_getCallerClass---------------------
4824 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4825 //
4826 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4827 //
4828 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4829 // in that it must skip particular security frames and checks for
4830 // caller sensitive methods.
4831 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4832 #ifndef PRODUCT
4833 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4834 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4835 }
4836 #endif
4837
4838 if (!jvms()->has_method()) {
4839 #ifndef PRODUCT
4840 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4841 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4842 }
4843 #endif
4844 return false;
4845 }
4846
4847 // Walk back up the JVM state to find the caller at the required
4848 // depth.
4849 JVMState* caller_jvms = jvms();
4850
4851 // Cf. JVM_GetCallerClass
4852 // NOTE: Start the loop at depth 1 because the current JVM state does
4853 // not include the Reflection.getCallerClass() frame.
4854 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4855 ciMethod* m = caller_jvms->method();
4856 switch (n) {
4857 case 0:
4858 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4859 break;
4860 case 1:
4861 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4862 if (!m->caller_sensitive()) {
4863 #ifndef PRODUCT
4864 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4865 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4866 }
4867 #endif
4868 return false; // bail-out; let JVM_GetCallerClass do the work
4869 }
4870 break;
4871 default:
4872 if (!m->is_ignored_by_security_stack_walk()) {
4873 // We have reached the desired frame; return the holder class.
4874 // Acquire method holder as java.lang.Class and push as constant.
4875 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4876 ciInstance* caller_mirror = caller_klass->java_mirror();
4877 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4878
4879 #ifndef PRODUCT
4880 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4881 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());
4882 tty->print_cr(" JVM state at this point:");
4883 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4884 ciMethod* m = jvms()->of_depth(i)->method();
4885 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4886 }
4887 }
4888 #endif
4889 return true;
4890 }
4891 break;
4892 }
4893 }
4894
4895 #ifndef PRODUCT
4896 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4897 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4898 tty->print_cr(" JVM state at this point:");
4899 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4900 ciMethod* m = jvms()->of_depth(i)->method();
4901 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4902 }
4903 }
4904 #endif
4905
4906 return false; // bail-out; let JVM_GetCallerClass do the work
4907 }
4908
4909 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4910 Node* arg = argument(0);
4911 Node* result = nullptr;
4912
4913 switch (id) {
4914 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4915 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4916 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4917 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4918 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4919 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4920
4921 case vmIntrinsics::_doubleToLongBits: {
4922 // two paths (plus control) merge in a wood
4923 RegionNode *r = new RegionNode(3);
4924 Node *phi = new PhiNode(r, TypeLong::LONG);
4925
4926 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4927 // Build the boolean node
4928 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4929
4930 // Branch either way.
4931 // NaN case is less traveled, which makes all the difference.
4932 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4933 Node *opt_isnan = _gvn.transform(ifisnan);
4934 assert( opt_isnan->is_If(), "Expect an IfNode");
4935 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4936 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4937
4938 set_control(iftrue);
4939
4940 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4941 Node *slow_result = longcon(nan_bits); // return NaN
4942 phi->init_req(1, _gvn.transform( slow_result ));
4943 r->init_req(1, iftrue);
4944
4945 // Else fall through
4946 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4947 set_control(iffalse);
4948
4949 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4950 r->init_req(2, iffalse);
4951
4952 // Post merge
4953 set_control(_gvn.transform(r));
4954 record_for_igvn(r);
4955
4956 C->set_has_split_ifs(true); // Has chance for split-if optimization
4957 result = phi;
4958 assert(result->bottom_type()->isa_long(), "must be");
4959 break;
4960 }
4961
4962 case vmIntrinsics::_floatToIntBits: {
4963 // two paths (plus control) merge in a wood
4964 RegionNode *r = new RegionNode(3);
4965 Node *phi = new PhiNode(r, TypeInt::INT);
4966
4967 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4968 // Build the boolean node
4969 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4970
4971 // Branch either way.
4972 // NaN case is less traveled, which makes all the difference.
4973 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4974 Node *opt_isnan = _gvn.transform(ifisnan);
4975 assert( opt_isnan->is_If(), "Expect an IfNode");
4976 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4977 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4978
4979 set_control(iftrue);
4980
4981 static const jint nan_bits = 0x7fc00000;
4982 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4983 phi->init_req(1, _gvn.transform( slow_result ));
4984 r->init_req(1, iftrue);
4985
4986 // Else fall through
4987 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4988 set_control(iffalse);
4989
4990 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4991 r->init_req(2, iffalse);
4992
4993 // Post merge
4994 set_control(_gvn.transform(r));
4995 record_for_igvn(r);
4996
4997 C->set_has_split_ifs(true); // Has chance for split-if optimization
4998 result = phi;
4999 assert(result->bottom_type()->isa_int(), "must be");
5000 break;
5001 }
5002
5003 default:
5004 fatal_unexpected_iid(id);
5005 break;
5006 }
5007 set_result(_gvn.transform(result));
5008 return true;
5009 }
5010
5011 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5012 Node* arg = argument(0);
5013 Node* result = nullptr;
5014
5015 switch (id) {
5016 case vmIntrinsics::_floatIsInfinite:
5017 result = new IsInfiniteFNode(arg);
5018 break;
5019 case vmIntrinsics::_floatIsFinite:
5020 result = new IsFiniteFNode(arg);
5021 break;
5022 case vmIntrinsics::_doubleIsInfinite:
5023 result = new IsInfiniteDNode(arg);
5024 break;
5025 case vmIntrinsics::_doubleIsFinite:
5026 result = new IsFiniteDNode(arg);
5027 break;
5028 default:
5029 fatal_unexpected_iid(id);
5030 break;
5031 }
5032 set_result(_gvn.transform(result));
5033 return true;
5034 }
5035
5036 //----------------------inline_unsafe_copyMemory-------------------------
5037 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5038
5039 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5040 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5041 const Type* base_t = gvn.type(base);
5042
5043 bool in_native = (base_t == TypePtr::NULL_PTR);
5044 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5045 bool is_mixed = !in_heap && !in_native;
5046
5047 if (is_mixed) {
5048 return true; // mixed accesses can touch both on-heap and off-heap memory
5049 }
5050 if (in_heap) {
5051 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5052 if (!is_prim_array) {
5053 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5054 // there's not enough type information available to determine proper memory slice for it.
5055 return true;
5056 }
5057 }
5058 return false;
5059 }
5060
5061 bool LibraryCallKit::inline_unsafe_copyMemory() {
5062 if (callee()->is_static()) return false; // caller must have the capability!
5063 null_check_receiver(); // null-check receiver
5064 if (stopped()) return true;
5065
5066 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5067
5068 Node* src_base = argument(1); // type: oop
5069 Node* src_off = ConvL2X(argument(2)); // type: long
5070 Node* dst_base = argument(4); // type: oop
5071 Node* dst_off = ConvL2X(argument(5)); // type: long
5072 Node* size = ConvL2X(argument(7)); // type: long
5073
5074 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5075 "fieldOffset must be byte-scaled");
5076
5077 Node* src_addr = make_unsafe_address(src_base, src_off);
5078 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5079
5080 Node* thread = _gvn.transform(new ThreadLocalNode());
5081 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5082 BasicType doing_unsafe_access_bt = T_BYTE;
5083 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5084
5085 // update volatile field
5086 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5087
5088 int flags = RC_LEAF | RC_NO_FP;
5089
5090 const TypePtr* dst_type = TypePtr::BOTTOM;
5091
5092 // Adjust memory effects of the runtime call based on input values.
5093 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5094 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5095 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5096
5097 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5098 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5099 flags |= RC_NARROW_MEM; // narrow in memory
5100 }
5101 }
5102
5103 // Call it. Note that the length argument is not scaled.
5104 make_runtime_call(flags,
5105 OptoRuntime::fast_arraycopy_Type(),
5106 StubRoutines::unsafe_arraycopy(),
5107 "unsafe_arraycopy",
5108 dst_type,
5109 src_addr, dst_addr, size XTOP);
5110
5111 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5112
5113 return true;
5114 }
5115
5116 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5117 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5118 bool LibraryCallKit::inline_unsafe_setMemory() {
5119 if (callee()->is_static()) return false; // caller must have the capability!
5120 null_check_receiver(); // null-check receiver
5121 if (stopped()) return true;
5122
5123 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5124
5125 Node* dst_base = argument(1); // type: oop
5126 Node* dst_off = ConvL2X(argument(2)); // type: long
5127 Node* size = ConvL2X(argument(4)); // type: long
5128 Node* byte = argument(6); // type: byte
5129
5130 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5131 "fieldOffset must be byte-scaled");
5132
5133 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5134
5135 Node* thread = _gvn.transform(new ThreadLocalNode());
5136 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5137 BasicType doing_unsafe_access_bt = T_BYTE;
5138 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5139
5140 // update volatile field
5141 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5142
5143 int flags = RC_LEAF | RC_NO_FP;
5144
5145 const TypePtr* dst_type = TypePtr::BOTTOM;
5146
5147 // Adjust memory effects of the runtime call based on input values.
5148 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5149 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5150
5151 flags |= RC_NARROW_MEM; // narrow in memory
5152 }
5153
5154 // Call it. Note that the length argument is not scaled.
5155 make_runtime_call(flags,
5156 OptoRuntime::unsafe_setmemory_Type(),
5157 StubRoutines::unsafe_setmemory(),
5158 "unsafe_setmemory",
5159 dst_type,
5160 dst_addr, size XTOP, byte);
5161
5162 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5163
5164 return true;
5165 }
5166
5167 #undef XTOP
5168
5169 //------------------------clone_coping-----------------------------------
5170 // Helper function for inline_native_clone.
5171 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5172 assert(obj_size != nullptr, "");
5173 Node* raw_obj = alloc_obj->in(1);
5174 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5175
5176 AllocateNode* alloc = nullptr;
5177 if (ReduceBulkZeroing &&
5178 // If we are implementing an array clone without knowing its source type
5179 // (can happen when compiling the array-guarded branch of a reflective
5180 // Object.clone() invocation), initialize the array within the allocation.
5181 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5182 // to a runtime clone call that assumes fully initialized source arrays.
5183 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5184 // We will be completely responsible for initializing this object -
5185 // mark Initialize node as complete.
5186 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5187 // The object was just allocated - there should be no any stores!
5188 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5189 // Mark as complete_with_arraycopy so that on AllocateNode
5190 // expansion, we know this AllocateNode is initialized by an array
5191 // copy and a StoreStore barrier exists after the array copy.
5192 alloc->initialization()->set_complete_with_arraycopy();
5193 }
5194
5195 Node* size = _gvn.transform(obj_size);
5196 access_clone(obj, alloc_obj, size, is_array);
5197
5198 // Do not let reads from the cloned object float above the arraycopy.
5199 if (alloc != nullptr) {
5200 // Do not let stores that initialize this object be reordered with
5201 // a subsequent store that would make this object accessible by
5202 // other threads.
5203 // Record what AllocateNode this StoreStore protects so that
5204 // escape analysis can go from the MemBarStoreStoreNode to the
5205 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5206 // based on the escape status of the AllocateNode.
5207 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5208 } else {
5209 insert_mem_bar(Op_MemBarCPUOrder);
5210 }
5211 }
5212
5213 //------------------------inline_native_clone----------------------------
5214 // protected native Object java.lang.Object.clone();
5215 //
5216 // Here are the simple edge cases:
5217 // null receiver => normal trap
5218 // virtual and clone was overridden => slow path to out-of-line clone
5219 // not cloneable or finalizer => slow path to out-of-line Object.clone
5220 //
5221 // The general case has two steps, allocation and copying.
5222 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5223 //
5224 // Copying also has two cases, oop arrays and everything else.
5225 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5226 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5227 //
5228 // These steps fold up nicely if and when the cloned object's klass
5229 // can be sharply typed as an object array, a type array, or an instance.
5230 //
5231 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5232 PhiNode* result_val;
5233
5234 // Set the reexecute bit for the interpreter to reexecute
5235 // the bytecode that invokes Object.clone if deoptimization happens.
5236 { PreserveReexecuteState preexecs(this);
5237 jvms()->set_should_reexecute(true);
5238
5239 Node* obj = null_check_receiver();
5240 if (stopped()) return true;
5241
5242 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5243
5244 // If we are going to clone an instance, we need its exact type to
5245 // know the number and types of fields to convert the clone to
5246 // loads/stores. Maybe a speculative type can help us.
5247 if (!obj_type->klass_is_exact() &&
5248 obj_type->speculative_type() != nullptr &&
5249 obj_type->speculative_type()->is_instance_klass()) {
5250 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5251 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5252 !spec_ik->has_injected_fields()) {
5253 if (!obj_type->isa_instptr() ||
5254 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5255 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5256 }
5257 }
5258 }
5259
5260 // Conservatively insert a memory barrier on all memory slices.
5261 // Do not let writes into the original float below the clone.
5262 insert_mem_bar(Op_MemBarCPUOrder);
5263
5264 // paths into result_reg:
5265 enum {
5266 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5267 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5268 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5269 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5270 PATH_LIMIT
5271 };
5272 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5273 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5274 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5275 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5276 record_for_igvn(result_reg);
5277
5278 Node* obj_klass = load_object_klass(obj);
5279 Node* array_obj = obj;
5280 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5281 if (array_ctl != nullptr) {
5282 // It's an array.
5283 PreserveJVMState pjvms(this);
5284 set_control(array_ctl);
5285 Node* obj_length = load_array_length(array_obj);
5286 Node* array_size = nullptr; // Size of the array without object alignment padding.
5287 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5288
5289 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5290 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5291 // If it is an oop array, it requires very special treatment,
5292 // because gc barriers are required when accessing the array.
5293 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5294 if (is_obja != nullptr) {
5295 PreserveJVMState pjvms2(this);
5296 set_control(is_obja);
5297 // Generate a direct call to the right arraycopy function(s).
5298 // Clones are always tightly coupled.
5299 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5300 ac->set_clone_oop_array();
5301 Node* n = _gvn.transform(ac);
5302 assert(n == ac, "cannot disappear");
5303 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5304
5305 result_reg->init_req(_objArray_path, control());
5306 result_val->init_req(_objArray_path, alloc_obj);
5307 result_i_o ->set_req(_objArray_path, i_o());
5308 result_mem ->set_req(_objArray_path, reset_memory());
5309 }
5310 }
5311 // Otherwise, there are no barriers to worry about.
5312 // (We can dispense with card marks if we know the allocation
5313 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5314 // causes the non-eden paths to take compensating steps to
5315 // simulate a fresh allocation, so that no further
5316 // card marks are required in compiled code to initialize
5317 // the object.)
5318
5319 if (!stopped()) {
5320 copy_to_clone(array_obj, alloc_obj, array_size, true);
5321
5322 // Present the results of the copy.
5323 result_reg->init_req(_array_path, control());
5324 result_val->init_req(_array_path, alloc_obj);
5325 result_i_o ->set_req(_array_path, i_o());
5326 result_mem ->set_req(_array_path, reset_memory());
5327 }
5328 }
5329
5330 // We only go to the instance fast case code if we pass a number of guards.
5331 // The paths which do not pass are accumulated in the slow_region.
5332 RegionNode* slow_region = new RegionNode(1);
5333 record_for_igvn(slow_region);
5334 if (!stopped()) {
5335 // It's an instance (we did array above). Make the slow-path tests.
5336 // If this is a virtual call, we generate a funny guard. We grab
5337 // the vtable entry corresponding to clone() from the target object.
5338 // If the target method which we are calling happens to be the
5339 // Object clone() method, we pass the guard. We do not need this
5340 // guard for non-virtual calls; the caller is known to be the native
5341 // Object clone().
5342 if (is_virtual) {
5343 generate_virtual_guard(obj_klass, slow_region);
5344 }
5345
5346 // The object must be easily cloneable and must not have a finalizer.
5347 // Both of these conditions may be checked in a single test.
5348 // We could optimize the test further, but we don't care.
5349 generate_misc_flags_guard(obj_klass,
5350 // Test both conditions:
5351 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5352 // Must be cloneable but not finalizer:
5353 KlassFlags::_misc_is_cloneable_fast,
5354 slow_region);
5355 }
5356
5357 if (!stopped()) {
5358 // It's an instance, and it passed the slow-path tests.
5359 PreserveJVMState pjvms(this);
5360 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5361 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5362 // is reexecuted if deoptimization occurs and there could be problems when merging
5363 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5364 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5365
5366 copy_to_clone(obj, alloc_obj, obj_size, false);
5367
5368 // Present the results of the slow call.
5369 result_reg->init_req(_instance_path, control());
5370 result_val->init_req(_instance_path, alloc_obj);
5371 result_i_o ->set_req(_instance_path, i_o());
5372 result_mem ->set_req(_instance_path, reset_memory());
5373 }
5374
5375 // Generate code for the slow case. We make a call to clone().
5376 set_control(_gvn.transform(slow_region));
5377 if (!stopped()) {
5378 PreserveJVMState pjvms(this);
5379 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5380 // We need to deoptimize on exception (see comment above)
5381 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5382 // this->control() comes from set_results_for_java_call
5383 result_reg->init_req(_slow_path, control());
5384 result_val->init_req(_slow_path, slow_result);
5385 result_i_o ->set_req(_slow_path, i_o());
5386 result_mem ->set_req(_slow_path, reset_memory());
5387 }
5388
5389 // Return the combined state.
5390 set_control( _gvn.transform(result_reg));
5391 set_i_o( _gvn.transform(result_i_o));
5392 set_all_memory( _gvn.transform(result_mem));
5393 } // original reexecute is set back here
5394
5395 set_result(_gvn.transform(result_val));
5396 return true;
5397 }
5398
5399 // If we have a tightly coupled allocation, the arraycopy may take care
5400 // of the array initialization. If one of the guards we insert between
5401 // the allocation and the arraycopy causes a deoptimization, an
5402 // uninitialized array will escape the compiled method. To prevent that
5403 // we set the JVM state for uncommon traps between the allocation and
5404 // the arraycopy to the state before the allocation so, in case of
5405 // deoptimization, we'll reexecute the allocation and the
5406 // initialization.
5407 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5408 if (alloc != nullptr) {
5409 ciMethod* trap_method = alloc->jvms()->method();
5410 int trap_bci = alloc->jvms()->bci();
5411
5412 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5413 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5414 // Make sure there's no store between the allocation and the
5415 // arraycopy otherwise visible side effects could be rexecuted
5416 // in case of deoptimization and cause incorrect execution.
5417 bool no_interfering_store = true;
5418 Node* mem = alloc->in(TypeFunc::Memory);
5419 if (mem->is_MergeMem()) {
5420 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5421 Node* n = mms.memory();
5422 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5423 assert(n->is_Store(), "what else?");
5424 no_interfering_store = false;
5425 break;
5426 }
5427 }
5428 } else {
5429 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5430 Node* n = mms.memory();
5431 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5432 assert(n->is_Store(), "what else?");
5433 no_interfering_store = false;
5434 break;
5435 }
5436 }
5437 }
5438
5439 if (no_interfering_store) {
5440 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5441
5442 JVMState* saved_jvms = jvms();
5443 saved_reexecute_sp = _reexecute_sp;
5444
5445 set_jvms(sfpt->jvms());
5446 _reexecute_sp = jvms()->sp();
5447
5448 return saved_jvms;
5449 }
5450 }
5451 }
5452 return nullptr;
5453 }
5454
5455 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5456 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5457 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5458 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5459 uint size = alloc->req();
5460 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5461 old_jvms->set_map(sfpt);
5462 for (uint i = 0; i < size; i++) {
5463 sfpt->init_req(i, alloc->in(i));
5464 }
5465 // re-push array length for deoptimization
5466 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5467 old_jvms->set_sp(old_jvms->sp()+1);
5468 old_jvms->set_monoff(old_jvms->monoff()+1);
5469 old_jvms->set_scloff(old_jvms->scloff()+1);
5470 old_jvms->set_endoff(old_jvms->endoff()+1);
5471 old_jvms->set_should_reexecute(true);
5472
5473 sfpt->set_i_o(map()->i_o());
5474 sfpt->set_memory(map()->memory());
5475 sfpt->set_control(map()->control());
5476 return sfpt;
5477 }
5478
5479 // In case of a deoptimization, we restart execution at the
5480 // allocation, allocating a new array. We would leave an uninitialized
5481 // array in the heap that GCs wouldn't expect. Move the allocation
5482 // after the traps so we don't allocate the array if we
5483 // deoptimize. This is possible because tightly_coupled_allocation()
5484 // guarantees there's no observer of the allocated array at this point
5485 // and the control flow is simple enough.
5486 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5487 int saved_reexecute_sp, uint new_idx) {
5488 if (saved_jvms_before_guards != nullptr && !stopped()) {
5489 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5490
5491 assert(alloc != nullptr, "only with a tightly coupled allocation");
5492 // restore JVM state to the state at the arraycopy
5493 saved_jvms_before_guards->map()->set_control(map()->control());
5494 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5495 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5496 // If we've improved the types of some nodes (null check) while
5497 // emitting the guards, propagate them to the current state
5498 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5499 set_jvms(saved_jvms_before_guards);
5500 _reexecute_sp = saved_reexecute_sp;
5501
5502 // Remove the allocation from above the guards
5503 CallProjections callprojs;
5504 alloc->extract_projections(&callprojs, true);
5505 InitializeNode* init = alloc->initialization();
5506 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5507 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5508 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5509
5510 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5511 // the allocation (i.e. is only valid if the allocation succeeds):
5512 // 1) replace CastIINode with AllocateArrayNode's length here
5513 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5514 //
5515 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5516 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5517 Node* init_control = init->proj_out(TypeFunc::Control);
5518 Node* alloc_length = alloc->Ideal_length();
5519 #ifdef ASSERT
5520 Node* prev_cast = nullptr;
5521 #endif
5522 for (uint i = 0; i < init_control->outcnt(); i++) {
5523 Node* init_out = init_control->raw_out(i);
5524 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5525 #ifdef ASSERT
5526 if (prev_cast == nullptr) {
5527 prev_cast = init_out;
5528 } else {
5529 if (prev_cast->cmp(*init_out) == false) {
5530 prev_cast->dump();
5531 init_out->dump();
5532 assert(false, "not equal CastIINode");
5533 }
5534 }
5535 #endif
5536 C->gvn_replace_by(init_out, alloc_length);
5537 }
5538 }
5539 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5540
5541 // move the allocation here (after the guards)
5542 _gvn.hash_delete(alloc);
5543 alloc->set_req(TypeFunc::Control, control());
5544 alloc->set_req(TypeFunc::I_O, i_o());
5545 Node *mem = reset_memory();
5546 set_all_memory(mem);
5547 alloc->set_req(TypeFunc::Memory, mem);
5548 set_control(init->proj_out_or_null(TypeFunc::Control));
5549 set_i_o(callprojs.fallthrough_ioproj);
5550
5551 // Update memory as done in GraphKit::set_output_for_allocation()
5552 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5553 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5554 if (ary_type->isa_aryptr() && length_type != nullptr) {
5555 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5556 }
5557 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5558 int elemidx = C->get_alias_index(telemref);
5559 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5560 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5561
5562 Node* allocx = _gvn.transform(alloc);
5563 assert(allocx == alloc, "where has the allocation gone?");
5564 assert(dest->is_CheckCastPP(), "not an allocation result?");
5565
5566 _gvn.hash_delete(dest);
5567 dest->set_req(0, control());
5568 Node* destx = _gvn.transform(dest);
5569 assert(destx == dest, "where has the allocation result gone?");
5570
5571 array_ideal_length(alloc, ary_type, true);
5572 }
5573 }
5574
5575 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5576 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5577 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5578 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5579 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5580 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5581 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5582 JVMState* saved_jvms_before_guards) {
5583 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5584 // There is at least one unrelated uncommon trap which needs to be replaced.
5585 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5586
5587 JVMState* saved_jvms = jvms();
5588 const int saved_reexecute_sp = _reexecute_sp;
5589 set_jvms(sfpt->jvms());
5590 _reexecute_sp = jvms()->sp();
5591
5592 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5593
5594 // Restore state
5595 set_jvms(saved_jvms);
5596 _reexecute_sp = saved_reexecute_sp;
5597 }
5598 }
5599
5600 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5601 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5602 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5603 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5604 while (if_proj->is_IfProj()) {
5605 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5606 if (uncommon_trap != nullptr) {
5607 create_new_uncommon_trap(uncommon_trap);
5608 }
5609 assert(if_proj->in(0)->is_If(), "must be If");
5610 if_proj = if_proj->in(0)->in(0);
5611 }
5612 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5613 "must have reached control projection of init node");
5614 }
5615
5616 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5617 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5618 assert(trap_request != 0, "no valid UCT trap request");
5619 PreserveJVMState pjvms(this);
5620 set_control(uncommon_trap_call->in(0));
5621 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5622 Deoptimization::trap_request_action(trap_request));
5623 assert(stopped(), "Should be stopped");
5624 _gvn.hash_delete(uncommon_trap_call);
5625 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5626 }
5627
5628 // Common checks for array sorting intrinsics arguments.
5629 // Returns `true` if checks passed.
5630 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5631 // check address of the class
5632 if (elementType == nullptr || elementType->is_top()) {
5633 return false; // dead path
5634 }
5635 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5636 if (elem_klass == nullptr) {
5637 return false; // dead path
5638 }
5639 // java_mirror_type() returns non-null for compile-time Class constants only
5640 ciType* elem_type = elem_klass->java_mirror_type();
5641 if (elem_type == nullptr) {
5642 return false;
5643 }
5644 bt = elem_type->basic_type();
5645 // Disable the intrinsic if the CPU does not support SIMD sort
5646 if (!Matcher::supports_simd_sort(bt)) {
5647 return false;
5648 }
5649 // check address of the array
5650 if (obj == nullptr || obj->is_top()) {
5651 return false; // dead path
5652 }
5653 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5654 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5655 return false; // failed input validation
5656 }
5657 return true;
5658 }
5659
5660 //------------------------------inline_array_partition-----------------------
5661 bool LibraryCallKit::inline_array_partition() {
5662 address stubAddr = StubRoutines::select_array_partition_function();
5663 if (stubAddr == nullptr) {
5664 return false; // Intrinsic's stub is not implemented on this platform
5665 }
5666 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5667
5668 // no receiver because it is a static method
5669 Node* elementType = argument(0);
5670 Node* obj = argument(1);
5671 Node* offset = argument(2); // long
5672 Node* fromIndex = argument(4);
5673 Node* toIndex = argument(5);
5674 Node* indexPivot1 = argument(6);
5675 Node* indexPivot2 = argument(7);
5676 // PartitionOperation: argument(8) is ignored
5677
5678 Node* pivotIndices = nullptr;
5679 BasicType bt = T_ILLEGAL;
5680
5681 if (!check_array_sort_arguments(elementType, obj, bt)) {
5682 return false;
5683 }
5684 null_check(obj);
5685 // If obj is dead, only null-path is taken.
5686 if (stopped()) {
5687 return true;
5688 }
5689 // Set the original stack and the reexecute bit for the interpreter to reexecute
5690 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5691 { PreserveReexecuteState preexecs(this);
5692 jvms()->set_should_reexecute(true);
5693
5694 Node* obj_adr = make_unsafe_address(obj, offset);
5695
5696 // create the pivotIndices array of type int and size = 2
5697 Node* size = intcon(2);
5698 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5699 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5700 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5701 guarantee(alloc != nullptr, "created above");
5702 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5703
5704 // pass the basic type enum to the stub
5705 Node* elemType = intcon(bt);
5706
5707 // Call the stub
5708 const char *stubName = "array_partition_stub";
5709 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5710 stubAddr, stubName, TypePtr::BOTTOM,
5711 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5712 indexPivot1, indexPivot2);
5713
5714 } // original reexecute is set back here
5715
5716 if (!stopped()) {
5717 set_result(pivotIndices);
5718 }
5719
5720 return true;
5721 }
5722
5723
5724 //------------------------------inline_array_sort-----------------------
5725 bool LibraryCallKit::inline_array_sort() {
5726 address stubAddr = StubRoutines::select_arraysort_function();
5727 if (stubAddr == nullptr) {
5728 return false; // Intrinsic's stub is not implemented on this platform
5729 }
5730 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5731
5732 // no receiver because it is a static method
5733 Node* elementType = argument(0);
5734 Node* obj = argument(1);
5735 Node* offset = argument(2); // long
5736 Node* fromIndex = argument(4);
5737 Node* toIndex = argument(5);
5738 // SortOperation: argument(6) is ignored
5739
5740 BasicType bt = T_ILLEGAL;
5741
5742 if (!check_array_sort_arguments(elementType, obj, bt)) {
5743 return false;
5744 }
5745 null_check(obj);
5746 // If obj is dead, only null-path is taken.
5747 if (stopped()) {
5748 return true;
5749 }
5750 Node* obj_adr = make_unsafe_address(obj, offset);
5751
5752 // pass the basic type enum to the stub
5753 Node* elemType = intcon(bt);
5754
5755 // Call the stub.
5756 const char *stubName = "arraysort_stub";
5757 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5758 stubAddr, stubName, TypePtr::BOTTOM,
5759 obj_adr, elemType, fromIndex, toIndex);
5760
5761 return true;
5762 }
5763
5764
5765 //------------------------------inline_arraycopy-----------------------
5766 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5767 // Object dest, int destPos,
5768 // int length);
5769 bool LibraryCallKit::inline_arraycopy() {
5770 // Get the arguments.
5771 Node* src = argument(0); // type: oop
5772 Node* src_offset = argument(1); // type: int
5773 Node* dest = argument(2); // type: oop
5774 Node* dest_offset = argument(3); // type: int
5775 Node* length = argument(4); // type: int
5776
5777 uint new_idx = C->unique();
5778
5779 // Check for allocation before we add nodes that would confuse
5780 // tightly_coupled_allocation()
5781 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5782
5783 int saved_reexecute_sp = -1;
5784 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5785 // See arraycopy_restore_alloc_state() comment
5786 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5787 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5788 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5789 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5790
5791 // The following tests must be performed
5792 // (1) src and dest are arrays.
5793 // (2) src and dest arrays must have elements of the same BasicType
5794 // (3) src and dest must not be null.
5795 // (4) src_offset must not be negative.
5796 // (5) dest_offset must not be negative.
5797 // (6) length must not be negative.
5798 // (7) src_offset + length must not exceed length of src.
5799 // (8) dest_offset + length must not exceed length of dest.
5800 // (9) each element of an oop array must be assignable
5801
5802 // (3) src and dest must not be null.
5803 // always do this here because we need the JVM state for uncommon traps
5804 Node* null_ctl = top();
5805 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5806 assert(null_ctl->is_top(), "no null control here");
5807 dest = null_check(dest, T_ARRAY);
5808
5809 if (!can_emit_guards) {
5810 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5811 // guards but the arraycopy node could still take advantage of a
5812 // tightly allocated allocation. tightly_coupled_allocation() is
5813 // called again to make sure it takes the null check above into
5814 // account: the null check is mandatory and if it caused an
5815 // uncommon trap to be emitted then the allocation can't be
5816 // considered tightly coupled in this context.
5817 alloc = tightly_coupled_allocation(dest);
5818 }
5819
5820 bool validated = false;
5821
5822 const Type* src_type = _gvn.type(src);
5823 const Type* dest_type = _gvn.type(dest);
5824 const TypeAryPtr* top_src = src_type->isa_aryptr();
5825 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5826
5827 // Do we have the type of src?
5828 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5829 // Do we have the type of dest?
5830 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5831 // Is the type for src from speculation?
5832 bool src_spec = false;
5833 // Is the type for dest from speculation?
5834 bool dest_spec = false;
5835
5836 if ((!has_src || !has_dest) && can_emit_guards) {
5837 // We don't have sufficient type information, let's see if
5838 // speculative types can help. We need to have types for both src
5839 // and dest so that it pays off.
5840
5841 // Do we already have or could we have type information for src
5842 bool could_have_src = has_src;
5843 // Do we already have or could we have type information for dest
5844 bool could_have_dest = has_dest;
5845
5846 ciKlass* src_k = nullptr;
5847 if (!has_src) {
5848 src_k = src_type->speculative_type_not_null();
5849 if (src_k != nullptr && src_k->is_array_klass()) {
5850 could_have_src = true;
5851 }
5852 }
5853
5854 ciKlass* dest_k = nullptr;
5855 if (!has_dest) {
5856 dest_k = dest_type->speculative_type_not_null();
5857 if (dest_k != nullptr && dest_k->is_array_klass()) {
5858 could_have_dest = true;
5859 }
5860 }
5861
5862 if (could_have_src && could_have_dest) {
5863 // This is going to pay off so emit the required guards
5864 if (!has_src) {
5865 src = maybe_cast_profiled_obj(src, src_k, true);
5866 src_type = _gvn.type(src);
5867 top_src = src_type->isa_aryptr();
5868 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5869 src_spec = true;
5870 }
5871 if (!has_dest) {
5872 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5873 dest_type = _gvn.type(dest);
5874 top_dest = dest_type->isa_aryptr();
5875 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5876 dest_spec = true;
5877 }
5878 }
5879 }
5880
5881 if (has_src && has_dest && can_emit_guards) {
5882 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5883 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5884 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5885 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5886
5887 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5888 // If both arrays are object arrays then having the exact types
5889 // for both will remove the need for a subtype check at runtime
5890 // before the call and may make it possible to pick a faster copy
5891 // routine (without a subtype check on every element)
5892 // Do we have the exact type of src?
5893 bool could_have_src = src_spec;
5894 // Do we have the exact type of dest?
5895 bool could_have_dest = dest_spec;
5896 ciKlass* src_k = nullptr;
5897 ciKlass* dest_k = nullptr;
5898 if (!src_spec) {
5899 src_k = src_type->speculative_type_not_null();
5900 if (src_k != nullptr && src_k->is_array_klass()) {
5901 could_have_src = true;
5902 }
5903 }
5904 if (!dest_spec) {
5905 dest_k = dest_type->speculative_type_not_null();
5906 if (dest_k != nullptr && dest_k->is_array_klass()) {
5907 could_have_dest = true;
5908 }
5909 }
5910 if (could_have_src && could_have_dest) {
5911 // If we can have both exact types, emit the missing guards
5912 if (could_have_src && !src_spec) {
5913 src = maybe_cast_profiled_obj(src, src_k, true);
5914 }
5915 if (could_have_dest && !dest_spec) {
5916 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5917 }
5918 }
5919 }
5920 }
5921
5922 ciMethod* trap_method = method();
5923 int trap_bci = bci();
5924 if (saved_jvms_before_guards != nullptr) {
5925 trap_method = alloc->jvms()->method();
5926 trap_bci = alloc->jvms()->bci();
5927 }
5928
5929 bool negative_length_guard_generated = false;
5930
5931 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5932 can_emit_guards &&
5933 !src->is_top() && !dest->is_top()) {
5934 // validate arguments: enables transformation the ArrayCopyNode
5935 validated = true;
5936
5937 RegionNode* slow_region = new RegionNode(1);
5938 record_for_igvn(slow_region);
5939
5940 // (1) src and dest are arrays.
5941 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5942 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5943
5944 // (2) src and dest arrays must have elements of the same BasicType
5945 // done at macro expansion or at Ideal transformation time
5946
5947 // (4) src_offset must not be negative.
5948 generate_negative_guard(src_offset, slow_region);
5949
5950 // (5) dest_offset must not be negative.
5951 generate_negative_guard(dest_offset, slow_region);
5952
5953 // (7) src_offset + length must not exceed length of src.
5954 generate_limit_guard(src_offset, length,
5955 load_array_length(src),
5956 slow_region);
5957
5958 // (8) dest_offset + length must not exceed length of dest.
5959 generate_limit_guard(dest_offset, length,
5960 load_array_length(dest),
5961 slow_region);
5962
5963 // (6) length must not be negative.
5964 // This is also checked in generate_arraycopy() during macro expansion, but
5965 // we also have to check it here for the case where the ArrayCopyNode will
5966 // be eliminated by Escape Analysis.
5967 if (EliminateAllocations) {
5968 generate_negative_guard(length, slow_region);
5969 negative_length_guard_generated = true;
5970 }
5971
5972 // (9) each element of an oop array must be assignable
5973 Node* dest_klass = load_object_klass(dest);
5974 if (src != dest) {
5975 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5976
5977 if (not_subtype_ctrl != top()) {
5978 PreserveJVMState pjvms(this);
5979 set_control(not_subtype_ctrl);
5980 uncommon_trap(Deoptimization::Reason_intrinsic,
5981 Deoptimization::Action_make_not_entrant);
5982 assert(stopped(), "Should be stopped");
5983 }
5984 }
5985 {
5986 PreserveJVMState pjvms(this);
5987 set_control(_gvn.transform(slow_region));
5988 uncommon_trap(Deoptimization::Reason_intrinsic,
5989 Deoptimization::Action_make_not_entrant);
5990 assert(stopped(), "Should be stopped");
5991 }
5992
5993 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5994 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5995 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5996 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5997 }
5998
5999 if (stopped()) {
6000 return true;
6001 }
6002
6003 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6004 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6005 // so the compiler has a chance to eliminate them: during macro expansion,
6006 // we have to set their control (CastPP nodes are eliminated).
6007 load_object_klass(src), load_object_klass(dest),
6008 load_array_length(src), load_array_length(dest));
6009
6010 ac->set_arraycopy(validated);
6011
6012 Node* n = _gvn.transform(ac);
6013 if (n == ac) {
6014 ac->connect_outputs(this);
6015 } else {
6016 assert(validated, "shouldn't transform if all arguments not validated");
6017 set_all_memory(n);
6018 }
6019 clear_upper_avx();
6020
6021
6022 return true;
6023 }
6024
6025
6026 // Helper function which determines if an arraycopy immediately follows
6027 // an allocation, with no intervening tests or other escapes for the object.
6028 AllocateArrayNode*
6029 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6030 if (stopped()) return nullptr; // no fast path
6031 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6032
6033 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6034 if (alloc == nullptr) return nullptr;
6035
6036 Node* rawmem = memory(Compile::AliasIdxRaw);
6037 // Is the allocation's memory state untouched?
6038 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6039 // Bail out if there have been raw-memory effects since the allocation.
6040 // (Example: There might have been a call or safepoint.)
6041 return nullptr;
6042 }
6043 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6044 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6045 return nullptr;
6046 }
6047
6048 // There must be no unexpected observers of this allocation.
6049 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6050 Node* obs = ptr->fast_out(i);
6051 if (obs != this->map()) {
6052 return nullptr;
6053 }
6054 }
6055
6056 // This arraycopy must unconditionally follow the allocation of the ptr.
6057 Node* alloc_ctl = ptr->in(0);
6058 Node* ctl = control();
6059 while (ctl != alloc_ctl) {
6060 // There may be guards which feed into the slow_region.
6061 // Any other control flow means that we might not get a chance
6062 // to finish initializing the allocated object.
6063 // Various low-level checks bottom out in uncommon traps. These
6064 // are considered safe since we've already checked above that
6065 // there is no unexpected observer of this allocation.
6066 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6067 assert(ctl->in(0)->is_If(), "must be If");
6068 ctl = ctl->in(0)->in(0);
6069 } else {
6070 return nullptr;
6071 }
6072 }
6073
6074 // If we get this far, we have an allocation which immediately
6075 // precedes the arraycopy, and we can take over zeroing the new object.
6076 // The arraycopy will finish the initialization, and provide
6077 // a new control state to which we will anchor the destination pointer.
6078
6079 return alloc;
6080 }
6081
6082 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6083 if (node->is_IfProj()) {
6084 Node* other_proj = node->as_IfProj()->other_if_proj();
6085 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6086 Node* obs = other_proj->fast_out(j);
6087 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6088 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6089 return obs->as_CallStaticJava();
6090 }
6091 }
6092 }
6093 return nullptr;
6094 }
6095
6096 //-------------inline_encodeISOArray-----------------------------------
6097 // encode char[] to byte[] in ISO_8859_1 or ASCII
6098 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6099 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6100 // no receiver since it is static method
6101 Node *src = argument(0);
6102 Node *src_offset = argument(1);
6103 Node *dst = argument(2);
6104 Node *dst_offset = argument(3);
6105 Node *length = argument(4);
6106
6107 src = must_be_not_null(src, true);
6108 dst = must_be_not_null(dst, true);
6109
6110 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6111 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6112 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6113 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6114 // failed array check
6115 return false;
6116 }
6117
6118 // Figure out the size and type of the elements we will be copying.
6119 BasicType src_elem = src_type->elem()->array_element_basic_type();
6120 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6121 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6122 return false;
6123 }
6124
6125 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6126 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6127 // 'src_start' points to src array + scaled offset
6128 // 'dst_start' points to dst array + scaled offset
6129
6130 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6131 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6132 enc = _gvn.transform(enc);
6133 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6134 set_memory(res_mem, mtype);
6135 set_result(enc);
6136 clear_upper_avx();
6137
6138 return true;
6139 }
6140
6141 //-------------inline_multiplyToLen-----------------------------------
6142 bool LibraryCallKit::inline_multiplyToLen() {
6143 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6144
6145 address stubAddr = StubRoutines::multiplyToLen();
6146 if (stubAddr == nullptr) {
6147 return false; // Intrinsic's stub is not implemented on this platform
6148 }
6149 const char* stubName = "multiplyToLen";
6150
6151 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6152
6153 // no receiver because it is a static method
6154 Node* x = argument(0);
6155 Node* xlen = argument(1);
6156 Node* y = argument(2);
6157 Node* ylen = argument(3);
6158 Node* z = argument(4);
6159
6160 x = must_be_not_null(x, true);
6161 y = must_be_not_null(y, true);
6162
6163 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6164 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6165 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6166 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6167 // failed array check
6168 return false;
6169 }
6170
6171 BasicType x_elem = x_type->elem()->array_element_basic_type();
6172 BasicType y_elem = y_type->elem()->array_element_basic_type();
6173 if (x_elem != T_INT || y_elem != T_INT) {
6174 return false;
6175 }
6176
6177 Node* x_start = array_element_address(x, intcon(0), x_elem);
6178 Node* y_start = array_element_address(y, intcon(0), y_elem);
6179 // 'x_start' points to x array + scaled xlen
6180 // 'y_start' points to y array + scaled ylen
6181
6182 Node* z_start = array_element_address(z, intcon(0), T_INT);
6183
6184 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6185 OptoRuntime::multiplyToLen_Type(),
6186 stubAddr, stubName, TypePtr::BOTTOM,
6187 x_start, xlen, y_start, ylen, z_start);
6188
6189 C->set_has_split_ifs(true); // Has chance for split-if optimization
6190 set_result(z);
6191 return true;
6192 }
6193
6194 //-------------inline_squareToLen------------------------------------
6195 bool LibraryCallKit::inline_squareToLen() {
6196 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6197
6198 address stubAddr = StubRoutines::squareToLen();
6199 if (stubAddr == nullptr) {
6200 return false; // Intrinsic's stub is not implemented on this platform
6201 }
6202 const char* stubName = "squareToLen";
6203
6204 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6205
6206 Node* x = argument(0);
6207 Node* len = argument(1);
6208 Node* z = argument(2);
6209 Node* zlen = argument(3);
6210
6211 x = must_be_not_null(x, true);
6212 z = must_be_not_null(z, true);
6213
6214 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6215 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6216 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6217 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6218 // failed array check
6219 return false;
6220 }
6221
6222 BasicType x_elem = x_type->elem()->array_element_basic_type();
6223 BasicType z_elem = z_type->elem()->array_element_basic_type();
6224 if (x_elem != T_INT || z_elem != T_INT) {
6225 return false;
6226 }
6227
6228
6229 Node* x_start = array_element_address(x, intcon(0), x_elem);
6230 Node* z_start = array_element_address(z, intcon(0), z_elem);
6231
6232 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6233 OptoRuntime::squareToLen_Type(),
6234 stubAddr, stubName, TypePtr::BOTTOM,
6235 x_start, len, z_start, zlen);
6236
6237 set_result(z);
6238 return true;
6239 }
6240
6241 //-------------inline_mulAdd------------------------------------------
6242 bool LibraryCallKit::inline_mulAdd() {
6243 assert(UseMulAddIntrinsic, "not implemented on this platform");
6244
6245 address stubAddr = StubRoutines::mulAdd();
6246 if (stubAddr == nullptr) {
6247 return false; // Intrinsic's stub is not implemented on this platform
6248 }
6249 const char* stubName = "mulAdd";
6250
6251 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6252
6253 Node* out = argument(0);
6254 Node* in = argument(1);
6255 Node* offset = argument(2);
6256 Node* len = argument(3);
6257 Node* k = argument(4);
6258
6259 in = must_be_not_null(in, true);
6260 out = must_be_not_null(out, true);
6261
6262 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6263 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6264 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6265 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6266 // failed array check
6267 return false;
6268 }
6269
6270 BasicType out_elem = out_type->elem()->array_element_basic_type();
6271 BasicType in_elem = in_type->elem()->array_element_basic_type();
6272 if (out_elem != T_INT || in_elem != T_INT) {
6273 return false;
6274 }
6275
6276 Node* outlen = load_array_length(out);
6277 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6278 Node* out_start = array_element_address(out, intcon(0), out_elem);
6279 Node* in_start = array_element_address(in, intcon(0), in_elem);
6280
6281 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6282 OptoRuntime::mulAdd_Type(),
6283 stubAddr, stubName, TypePtr::BOTTOM,
6284 out_start,in_start, new_offset, len, k);
6285 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6286 set_result(result);
6287 return true;
6288 }
6289
6290 //-------------inline_montgomeryMultiply-----------------------------------
6291 bool LibraryCallKit::inline_montgomeryMultiply() {
6292 address stubAddr = StubRoutines::montgomeryMultiply();
6293 if (stubAddr == nullptr) {
6294 return false; // Intrinsic's stub is not implemented on this platform
6295 }
6296
6297 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6298 const char* stubName = "montgomery_multiply";
6299
6300 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6301
6302 Node* a = argument(0);
6303 Node* b = argument(1);
6304 Node* n = argument(2);
6305 Node* len = argument(3);
6306 Node* inv = argument(4);
6307 Node* m = argument(6);
6308
6309 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6310 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6311 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6312 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6313 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6314 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6315 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6316 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6317 // failed array check
6318 return false;
6319 }
6320
6321 BasicType a_elem = a_type->elem()->array_element_basic_type();
6322 BasicType b_elem = b_type->elem()->array_element_basic_type();
6323 BasicType n_elem = n_type->elem()->array_element_basic_type();
6324 BasicType m_elem = m_type->elem()->array_element_basic_type();
6325 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6326 return false;
6327 }
6328
6329 // Make the call
6330 {
6331 Node* a_start = array_element_address(a, intcon(0), a_elem);
6332 Node* b_start = array_element_address(b, intcon(0), b_elem);
6333 Node* n_start = array_element_address(n, intcon(0), n_elem);
6334 Node* m_start = array_element_address(m, intcon(0), m_elem);
6335
6336 Node* call = make_runtime_call(RC_LEAF,
6337 OptoRuntime::montgomeryMultiply_Type(),
6338 stubAddr, stubName, TypePtr::BOTTOM,
6339 a_start, b_start, n_start, len, inv, top(),
6340 m_start);
6341 set_result(m);
6342 }
6343
6344 return true;
6345 }
6346
6347 bool LibraryCallKit::inline_montgomerySquare() {
6348 address stubAddr = StubRoutines::montgomerySquare();
6349 if (stubAddr == nullptr) {
6350 return false; // Intrinsic's stub is not implemented on this platform
6351 }
6352
6353 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6354 const char* stubName = "montgomery_square";
6355
6356 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6357
6358 Node* a = argument(0);
6359 Node* n = argument(1);
6360 Node* len = argument(2);
6361 Node* inv = argument(3);
6362 Node* m = argument(5);
6363
6364 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6365 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6366 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6367 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6368 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6369 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6370 // failed array check
6371 return false;
6372 }
6373
6374 BasicType a_elem = a_type->elem()->array_element_basic_type();
6375 BasicType n_elem = n_type->elem()->array_element_basic_type();
6376 BasicType m_elem = m_type->elem()->array_element_basic_type();
6377 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6378 return false;
6379 }
6380
6381 // Make the call
6382 {
6383 Node* a_start = array_element_address(a, intcon(0), a_elem);
6384 Node* n_start = array_element_address(n, intcon(0), n_elem);
6385 Node* m_start = array_element_address(m, intcon(0), m_elem);
6386
6387 Node* call = make_runtime_call(RC_LEAF,
6388 OptoRuntime::montgomerySquare_Type(),
6389 stubAddr, stubName, TypePtr::BOTTOM,
6390 a_start, n_start, len, inv, top(),
6391 m_start);
6392 set_result(m);
6393 }
6394
6395 return true;
6396 }
6397
6398 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6399 address stubAddr = nullptr;
6400 const char* stubName = nullptr;
6401
6402 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6403 if (stubAddr == nullptr) {
6404 return false; // Intrinsic's stub is not implemented on this platform
6405 }
6406
6407 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6408
6409 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6410
6411 Node* newArr = argument(0);
6412 Node* oldArr = argument(1);
6413 Node* newIdx = argument(2);
6414 Node* shiftCount = argument(3);
6415 Node* numIter = argument(4);
6416
6417 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6418 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6419 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6420 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6421 return false;
6422 }
6423
6424 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6425 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6426 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6427 return false;
6428 }
6429
6430 // Make the call
6431 {
6432 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6433 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6434
6435 Node* call = make_runtime_call(RC_LEAF,
6436 OptoRuntime::bigIntegerShift_Type(),
6437 stubAddr,
6438 stubName,
6439 TypePtr::BOTTOM,
6440 newArr_start,
6441 oldArr_start,
6442 newIdx,
6443 shiftCount,
6444 numIter);
6445 }
6446
6447 return true;
6448 }
6449
6450 //-------------inline_vectorizedMismatch------------------------------
6451 bool LibraryCallKit::inline_vectorizedMismatch() {
6452 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6453
6454 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6455 Node* obja = argument(0); // Object
6456 Node* aoffset = argument(1); // long
6457 Node* objb = argument(3); // Object
6458 Node* boffset = argument(4); // long
6459 Node* length = argument(6); // int
6460 Node* scale = argument(7); // int
6461
6462 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6463 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6464 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6465 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6466 scale == top()) {
6467 return false; // failed input validation
6468 }
6469
6470 Node* obja_adr = make_unsafe_address(obja, aoffset);
6471 Node* objb_adr = make_unsafe_address(objb, boffset);
6472
6473 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6474 //
6475 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6476 // if (length <= inline_limit) {
6477 // inline_path:
6478 // vmask = VectorMaskGen length
6479 // vload1 = LoadVectorMasked obja, vmask
6480 // vload2 = LoadVectorMasked objb, vmask
6481 // result1 = VectorCmpMasked vload1, vload2, vmask
6482 // } else {
6483 // call_stub_path:
6484 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6485 // }
6486 // exit_block:
6487 // return Phi(result1, result2);
6488 //
6489 enum { inline_path = 1, // input is small enough to process it all at once
6490 stub_path = 2, // input is too large; call into the VM
6491 PATH_LIMIT = 3
6492 };
6493
6494 Node* exit_block = new RegionNode(PATH_LIMIT);
6495 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6496 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6497
6498 Node* call_stub_path = control();
6499
6500 BasicType elem_bt = T_ILLEGAL;
6501
6502 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6503 if (scale_t->is_con()) {
6504 switch (scale_t->get_con()) {
6505 case 0: elem_bt = T_BYTE; break;
6506 case 1: elem_bt = T_SHORT; break;
6507 case 2: elem_bt = T_INT; break;
6508 case 3: elem_bt = T_LONG; break;
6509
6510 default: elem_bt = T_ILLEGAL; break; // not supported
6511 }
6512 }
6513
6514 int inline_limit = 0;
6515 bool do_partial_inline = false;
6516
6517 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6518 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6519 do_partial_inline = inline_limit >= 16;
6520 }
6521
6522 if (do_partial_inline) {
6523 assert(elem_bt != T_ILLEGAL, "sanity");
6524
6525 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6526 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6527 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6528
6529 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6530 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6531 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6532
6533 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6534
6535 if (!stopped()) {
6536 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6537
6538 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6539 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6540 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6541 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6542
6543 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6544 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6545 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6546 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6547
6548 exit_block->init_req(inline_path, control());
6549 memory_phi->init_req(inline_path, map()->memory());
6550 result_phi->init_req(inline_path, result);
6551
6552 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6553 clear_upper_avx();
6554 }
6555 }
6556 }
6557
6558 if (call_stub_path != nullptr) {
6559 set_control(call_stub_path);
6560
6561 Node* call = make_runtime_call(RC_LEAF,
6562 OptoRuntime::vectorizedMismatch_Type(),
6563 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6564 obja_adr, objb_adr, length, scale);
6565
6566 exit_block->init_req(stub_path, control());
6567 memory_phi->init_req(stub_path, map()->memory());
6568 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6569 }
6570
6571 exit_block = _gvn.transform(exit_block);
6572 memory_phi = _gvn.transform(memory_phi);
6573 result_phi = _gvn.transform(result_phi);
6574
6575 set_control(exit_block);
6576 set_all_memory(memory_phi);
6577 set_result(result_phi);
6578
6579 return true;
6580 }
6581
6582 //------------------------------inline_vectorizedHashcode----------------------------
6583 bool LibraryCallKit::inline_vectorizedHashCode() {
6584 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6585
6586 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6587 Node* array = argument(0);
6588 Node* offset = argument(1);
6589 Node* length = argument(2);
6590 Node* initialValue = argument(3);
6591 Node* basic_type = argument(4);
6592
6593 if (basic_type == top()) {
6594 return false; // failed input validation
6595 }
6596
6597 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6598 if (!basic_type_t->is_con()) {
6599 return false; // Only intrinsify if mode argument is constant
6600 }
6601
6602 array = must_be_not_null(array, true);
6603
6604 BasicType bt = (BasicType)basic_type_t->get_con();
6605
6606 // Resolve address of first element
6607 Node* array_start = array_element_address(array, offset, bt);
6608
6609 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6610 array_start, length, initialValue, basic_type)));
6611 clear_upper_avx();
6612
6613 return true;
6614 }
6615
6616 /**
6617 * Calculate CRC32 for byte.
6618 * int java.util.zip.CRC32.update(int crc, int b)
6619 */
6620 bool LibraryCallKit::inline_updateCRC32() {
6621 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6622 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6623 // no receiver since it is static method
6624 Node* crc = argument(0); // type: int
6625 Node* b = argument(1); // type: int
6626
6627 /*
6628 * int c = ~ crc;
6629 * b = timesXtoThe32[(b ^ c) & 0xFF];
6630 * b = b ^ (c >>> 8);
6631 * crc = ~b;
6632 */
6633
6634 Node* M1 = intcon(-1);
6635 crc = _gvn.transform(new XorINode(crc, M1));
6636 Node* result = _gvn.transform(new XorINode(crc, b));
6637 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6638
6639 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6640 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6641 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6642 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6643
6644 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6645 result = _gvn.transform(new XorINode(crc, result));
6646 result = _gvn.transform(new XorINode(result, M1));
6647 set_result(result);
6648 return true;
6649 }
6650
6651 /**
6652 * Calculate CRC32 for byte[] array.
6653 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6654 */
6655 bool LibraryCallKit::inline_updateBytesCRC32() {
6656 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6657 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6658 // no receiver since it is static method
6659 Node* crc = argument(0); // type: int
6660 Node* src = argument(1); // type: oop
6661 Node* offset = argument(2); // type: int
6662 Node* length = argument(3); // type: int
6663
6664 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6665 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6666 // failed array check
6667 return false;
6668 }
6669
6670 // Figure out the size and type of the elements we will be copying.
6671 BasicType src_elem = src_type->elem()->array_element_basic_type();
6672 if (src_elem != T_BYTE) {
6673 return false;
6674 }
6675
6676 // 'src_start' points to src array + scaled offset
6677 src = must_be_not_null(src, true);
6678 Node* src_start = array_element_address(src, offset, src_elem);
6679
6680 // We assume that range check is done by caller.
6681 // TODO: generate range check (offset+length < src.length) in debug VM.
6682
6683 // Call the stub.
6684 address stubAddr = StubRoutines::updateBytesCRC32();
6685 const char *stubName = "updateBytesCRC32";
6686
6687 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6688 stubAddr, stubName, TypePtr::BOTTOM,
6689 crc, src_start, length);
6690 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6691 set_result(result);
6692 return true;
6693 }
6694
6695 /**
6696 * Calculate CRC32 for ByteBuffer.
6697 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6698 */
6699 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6700 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6701 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6702 // no receiver since it is static method
6703 Node* crc = argument(0); // type: int
6704 Node* src = argument(1); // type: long
6705 Node* offset = argument(3); // type: int
6706 Node* length = argument(4); // type: int
6707
6708 src = ConvL2X(src); // adjust Java long to machine word
6709 Node* base = _gvn.transform(new CastX2PNode(src));
6710 offset = ConvI2X(offset);
6711
6712 // 'src_start' points to src array + scaled offset
6713 Node* src_start = basic_plus_adr(top(), base, offset);
6714
6715 // Call the stub.
6716 address stubAddr = StubRoutines::updateBytesCRC32();
6717 const char *stubName = "updateBytesCRC32";
6718
6719 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6720 stubAddr, stubName, TypePtr::BOTTOM,
6721 crc, src_start, length);
6722 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6723 set_result(result);
6724 return true;
6725 }
6726
6727 //------------------------------get_table_from_crc32c_class-----------------------
6728 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6729 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6730 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6731
6732 return table;
6733 }
6734
6735 //------------------------------inline_updateBytesCRC32C-----------------------
6736 //
6737 // Calculate CRC32C for byte[] array.
6738 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6739 //
6740 bool LibraryCallKit::inline_updateBytesCRC32C() {
6741 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6742 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6743 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6744 // no receiver since it is a static method
6745 Node* crc = argument(0); // type: int
6746 Node* src = argument(1); // type: oop
6747 Node* offset = argument(2); // type: int
6748 Node* end = argument(3); // type: int
6749
6750 Node* length = _gvn.transform(new SubINode(end, offset));
6751
6752 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6753 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6754 // failed array check
6755 return false;
6756 }
6757
6758 // Figure out the size and type of the elements we will be copying.
6759 BasicType src_elem = src_type->elem()->array_element_basic_type();
6760 if (src_elem != T_BYTE) {
6761 return false;
6762 }
6763
6764 // 'src_start' points to src array + scaled offset
6765 src = must_be_not_null(src, true);
6766 Node* src_start = array_element_address(src, offset, src_elem);
6767
6768 // static final int[] byteTable in class CRC32C
6769 Node* table = get_table_from_crc32c_class(callee()->holder());
6770 table = must_be_not_null(table, true);
6771 Node* table_start = array_element_address(table, intcon(0), T_INT);
6772
6773 // We assume that range check is done by caller.
6774 // TODO: generate range check (offset+length < src.length) in debug VM.
6775
6776 // Call the stub.
6777 address stubAddr = StubRoutines::updateBytesCRC32C();
6778 const char *stubName = "updateBytesCRC32C";
6779
6780 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6781 stubAddr, stubName, TypePtr::BOTTOM,
6782 crc, src_start, length, table_start);
6783 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6784 set_result(result);
6785 return true;
6786 }
6787
6788 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6789 //
6790 // Calculate CRC32C for DirectByteBuffer.
6791 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6792 //
6793 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6794 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6795 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6796 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6797 // no receiver since it is a static method
6798 Node* crc = argument(0); // type: int
6799 Node* src = argument(1); // type: long
6800 Node* offset = argument(3); // type: int
6801 Node* end = argument(4); // type: int
6802
6803 Node* length = _gvn.transform(new SubINode(end, offset));
6804
6805 src = ConvL2X(src); // adjust Java long to machine word
6806 Node* base = _gvn.transform(new CastX2PNode(src));
6807 offset = ConvI2X(offset);
6808
6809 // 'src_start' points to src array + scaled offset
6810 Node* src_start = basic_plus_adr(top(), base, offset);
6811
6812 // static final int[] byteTable in class CRC32C
6813 Node* table = get_table_from_crc32c_class(callee()->holder());
6814 table = must_be_not_null(table, true);
6815 Node* table_start = array_element_address(table, intcon(0), T_INT);
6816
6817 // Call the stub.
6818 address stubAddr = StubRoutines::updateBytesCRC32C();
6819 const char *stubName = "updateBytesCRC32C";
6820
6821 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6822 stubAddr, stubName, TypePtr::BOTTOM,
6823 crc, src_start, length, table_start);
6824 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6825 set_result(result);
6826 return true;
6827 }
6828
6829 //------------------------------inline_updateBytesAdler32----------------------
6830 //
6831 // Calculate Adler32 checksum for byte[] array.
6832 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6833 //
6834 bool LibraryCallKit::inline_updateBytesAdler32() {
6835 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6836 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6837 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6838 // no receiver since it is static method
6839 Node* crc = argument(0); // type: int
6840 Node* src = argument(1); // type: oop
6841 Node* offset = argument(2); // type: int
6842 Node* length = argument(3); // type: int
6843
6844 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6845 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6846 // failed array check
6847 return false;
6848 }
6849
6850 // Figure out the size and type of the elements we will be copying.
6851 BasicType src_elem = src_type->elem()->array_element_basic_type();
6852 if (src_elem != T_BYTE) {
6853 return false;
6854 }
6855
6856 // 'src_start' points to src array + scaled offset
6857 Node* src_start = array_element_address(src, offset, src_elem);
6858
6859 // We assume that range check is done by caller.
6860 // TODO: generate range check (offset+length < src.length) in debug VM.
6861
6862 // Call the stub.
6863 address stubAddr = StubRoutines::updateBytesAdler32();
6864 const char *stubName = "updateBytesAdler32";
6865
6866 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6867 stubAddr, stubName, TypePtr::BOTTOM,
6868 crc, src_start, length);
6869 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6870 set_result(result);
6871 return true;
6872 }
6873
6874 //------------------------------inline_updateByteBufferAdler32---------------
6875 //
6876 // Calculate Adler32 checksum for DirectByteBuffer.
6877 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6878 //
6879 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6880 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6881 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6882 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6883 // no receiver since it is static method
6884 Node* crc = argument(0); // type: int
6885 Node* src = argument(1); // type: long
6886 Node* offset = argument(3); // type: int
6887 Node* length = argument(4); // type: int
6888
6889 src = ConvL2X(src); // adjust Java long to machine word
6890 Node* base = _gvn.transform(new CastX2PNode(src));
6891 offset = ConvI2X(offset);
6892
6893 // 'src_start' points to src array + scaled offset
6894 Node* src_start = basic_plus_adr(top(), base, offset);
6895
6896 // Call the stub.
6897 address stubAddr = StubRoutines::updateBytesAdler32();
6898 const char *stubName = "updateBytesAdler32";
6899
6900 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6901 stubAddr, stubName, TypePtr::BOTTOM,
6902 crc, src_start, length);
6903
6904 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6905 set_result(result);
6906 return true;
6907 }
6908
6909 //----------------------------inline_reference_get----------------------------
6910 // public T java.lang.ref.Reference.get();
6911 bool LibraryCallKit::inline_reference_get() {
6912 const int referent_offset = java_lang_ref_Reference::referent_offset();
6913
6914 // Get the argument:
6915 Node* reference_obj = null_check_receiver();
6916 if (stopped()) return true;
6917
6918 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6919 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6920 decorators, /*is_static*/ false, nullptr);
6921 if (result == nullptr) return false;
6922
6923 // Add memory barrier to prevent commoning reads from this field
6924 // across safepoint since GC can change its value.
6925 insert_mem_bar(Op_MemBarCPUOrder);
6926
6927 set_result(result);
6928 return true;
6929 }
6930
6931 //----------------------------inline_reference_refersTo0----------------------------
6932 // bool java.lang.ref.Reference.refersTo0();
6933 // bool java.lang.ref.PhantomReference.refersTo0();
6934 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6935 // Get arguments:
6936 Node* reference_obj = null_check_receiver();
6937 Node* other_obj = argument(1);
6938 if (stopped()) return true;
6939
6940 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6941 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6942 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6943 decorators, /*is_static*/ false, nullptr);
6944 if (referent == nullptr) return false;
6945
6946 // Add memory barrier to prevent commoning reads from this field
6947 // across safepoint since GC can change its value.
6948 insert_mem_bar(Op_MemBarCPUOrder);
6949
6950 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6951 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6952 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6953
6954 RegionNode* region = new RegionNode(3);
6955 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6956
6957 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6958 region->init_req(1, if_true);
6959 phi->init_req(1, intcon(1));
6960
6961 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6962 region->init_req(2, if_false);
6963 phi->init_req(2, intcon(0));
6964
6965 set_control(_gvn.transform(region));
6966 record_for_igvn(region);
6967 set_result(_gvn.transform(phi));
6968 return true;
6969 }
6970
6971 //----------------------------inline_reference_clear0----------------------------
6972 // void java.lang.ref.Reference.clear0();
6973 // void java.lang.ref.PhantomReference.clear0();
6974 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6975 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6976
6977 // Get arguments
6978 Node* reference_obj = null_check_receiver();
6979 if (stopped()) return true;
6980
6981 // Common access parameters
6982 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6983 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6984 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6985 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6986 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6987
6988 Node* referent = access_load_at(reference_obj,
6989 referent_field_addr,
6990 referent_field_addr_type,
6991 val_type,
6992 T_OBJECT,
6993 decorators);
6994
6995 IdealKit ideal(this);
6996 #define __ ideal.
6997 __ if_then(referent, BoolTest::ne, null());
6998 sync_kit(ideal);
6999 access_store_at(reference_obj,
7000 referent_field_addr,
7001 referent_field_addr_type,
7002 null(),
7003 val_type,
7004 T_OBJECT,
7005 decorators);
7006 __ sync_kit(this);
7007 __ end_if();
7008 final_sync(ideal);
7009 #undef __
7010
7011 return true;
7012 }
7013
7014 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7015 DecoratorSet decorators, bool is_static,
7016 ciInstanceKlass* fromKls) {
7017 if (fromKls == nullptr) {
7018 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7019 assert(tinst != nullptr, "obj is null");
7020 assert(tinst->is_loaded(), "obj is not loaded");
7021 fromKls = tinst->instance_klass();
7022 } else {
7023 assert(is_static, "only for static field access");
7024 }
7025 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7026 ciSymbol::make(fieldTypeString),
7027 is_static);
7028
7029 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7030 if (field == nullptr) return (Node *) nullptr;
7031
7032 if (is_static) {
7033 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7034 fromObj = makecon(tip);
7035 }
7036
7037 // Next code copied from Parse::do_get_xxx():
7038
7039 // Compute address and memory type.
7040 int offset = field->offset_in_bytes();
7041 bool is_vol = field->is_volatile();
7042 ciType* field_klass = field->type();
7043 assert(field_klass->is_loaded(), "should be loaded");
7044 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7045 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7046 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7047 "slice of address and input slice don't match");
7048 BasicType bt = field->layout_type();
7049
7050 // Build the resultant type of the load
7051 const Type *type;
7052 if (bt == T_OBJECT) {
7053 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7054 } else {
7055 type = Type::get_const_basic_type(bt);
7056 }
7057
7058 if (is_vol) {
7059 decorators |= MO_SEQ_CST;
7060 }
7061
7062 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7063 }
7064
7065 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7066 bool is_exact /* true */, bool is_static /* false */,
7067 ciInstanceKlass * fromKls /* nullptr */) {
7068 if (fromKls == nullptr) {
7069 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7070 assert(tinst != nullptr, "obj is null");
7071 assert(tinst->is_loaded(), "obj is not loaded");
7072 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7073 fromKls = tinst->instance_klass();
7074 }
7075 else {
7076 assert(is_static, "only for static field access");
7077 }
7078 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7079 ciSymbol::make(fieldTypeString),
7080 is_static);
7081
7082 assert(field != nullptr, "undefined field");
7083 assert(!field->is_volatile(), "not defined for volatile fields");
7084
7085 if (is_static) {
7086 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7087 fromObj = makecon(tip);
7088 }
7089
7090 // Next code copied from Parse::do_get_xxx():
7091
7092 // Compute address and memory type.
7093 int offset = field->offset_in_bytes();
7094 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7095
7096 return adr;
7097 }
7098
7099 //------------------------------inline_aescrypt_Block-----------------------
7100 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7101 address stubAddr = nullptr;
7102 const char *stubName;
7103 assert(UseAES, "need AES instruction support");
7104
7105 switch(id) {
7106 case vmIntrinsics::_aescrypt_encryptBlock:
7107 stubAddr = StubRoutines::aescrypt_encryptBlock();
7108 stubName = "aescrypt_encryptBlock";
7109 break;
7110 case vmIntrinsics::_aescrypt_decryptBlock:
7111 stubAddr = StubRoutines::aescrypt_decryptBlock();
7112 stubName = "aescrypt_decryptBlock";
7113 break;
7114 default:
7115 break;
7116 }
7117 if (stubAddr == nullptr) return false;
7118
7119 Node* aescrypt_object = argument(0);
7120 Node* src = argument(1);
7121 Node* src_offset = argument(2);
7122 Node* dest = argument(3);
7123 Node* dest_offset = argument(4);
7124
7125 src = must_be_not_null(src, true);
7126 dest = must_be_not_null(dest, true);
7127
7128 // (1) src and dest are arrays.
7129 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7130 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7131 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7132 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7133
7134 // for the quick and dirty code we will skip all the checks.
7135 // we are just trying to get the call to be generated.
7136 Node* src_start = src;
7137 Node* dest_start = dest;
7138 if (src_offset != nullptr || dest_offset != nullptr) {
7139 assert(src_offset != nullptr && dest_offset != nullptr, "");
7140 src_start = array_element_address(src, src_offset, T_BYTE);
7141 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7142 }
7143
7144 // now need to get the start of its expanded key array
7145 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7146 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7147 if (k_start == nullptr) return false;
7148
7149 // Call the stub.
7150 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7151 stubAddr, stubName, TypePtr::BOTTOM,
7152 src_start, dest_start, k_start);
7153
7154 return true;
7155 }
7156
7157 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7158 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7159 address stubAddr = nullptr;
7160 const char *stubName = nullptr;
7161
7162 assert(UseAES, "need AES instruction support");
7163
7164 switch(id) {
7165 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7166 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7167 stubName = "cipherBlockChaining_encryptAESCrypt";
7168 break;
7169 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7170 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7171 stubName = "cipherBlockChaining_decryptAESCrypt";
7172 break;
7173 default:
7174 break;
7175 }
7176 if (stubAddr == nullptr) return false;
7177
7178 Node* cipherBlockChaining_object = argument(0);
7179 Node* src = argument(1);
7180 Node* src_offset = argument(2);
7181 Node* len = argument(3);
7182 Node* dest = argument(4);
7183 Node* dest_offset = argument(5);
7184
7185 src = must_be_not_null(src, false);
7186 dest = must_be_not_null(dest, false);
7187
7188 // (1) src and dest are arrays.
7189 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7190 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7191 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7192 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7193
7194 // checks are the responsibility of the caller
7195 Node* src_start = src;
7196 Node* dest_start = dest;
7197 if (src_offset != nullptr || dest_offset != nullptr) {
7198 assert(src_offset != nullptr && dest_offset != nullptr, "");
7199 src_start = array_element_address(src, src_offset, T_BYTE);
7200 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7201 }
7202
7203 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7204 // (because of the predicated logic executed earlier).
7205 // so we cast it here safely.
7206 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7207
7208 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7209 if (embeddedCipherObj == nullptr) return false;
7210
7211 // cast it to what we know it will be at runtime
7212 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7213 assert(tinst != nullptr, "CBC obj is null");
7214 assert(tinst->is_loaded(), "CBC obj is not loaded");
7215 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7216 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7217
7218 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7219 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7220 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7221 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7222 aescrypt_object = _gvn.transform(aescrypt_object);
7223
7224 // we need to get the start of the aescrypt_object's expanded key array
7225 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7226 if (k_start == nullptr) return false;
7227
7228 // similarly, get the start address of the r vector
7229 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7230 if (objRvec == nullptr) return false;
7231 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7232
7233 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7234 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7235 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7236 stubAddr, stubName, TypePtr::BOTTOM,
7237 src_start, dest_start, k_start, r_start, len);
7238
7239 // return cipher length (int)
7240 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7241 set_result(retvalue);
7242 return true;
7243 }
7244
7245 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7246 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7247 address stubAddr = nullptr;
7248 const char *stubName = nullptr;
7249
7250 assert(UseAES, "need AES instruction support");
7251
7252 switch (id) {
7253 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7254 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7255 stubName = "electronicCodeBook_encryptAESCrypt";
7256 break;
7257 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7258 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7259 stubName = "electronicCodeBook_decryptAESCrypt";
7260 break;
7261 default:
7262 break;
7263 }
7264
7265 if (stubAddr == nullptr) return false;
7266
7267 Node* electronicCodeBook_object = argument(0);
7268 Node* src = argument(1);
7269 Node* src_offset = argument(2);
7270 Node* len = argument(3);
7271 Node* dest = argument(4);
7272 Node* dest_offset = argument(5);
7273
7274 // (1) src and dest are arrays.
7275 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7276 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7277 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7278 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7279
7280 // checks are the responsibility of the caller
7281 Node* src_start = src;
7282 Node* dest_start = dest;
7283 if (src_offset != nullptr || dest_offset != nullptr) {
7284 assert(src_offset != nullptr && dest_offset != nullptr, "");
7285 src_start = array_element_address(src, src_offset, T_BYTE);
7286 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7287 }
7288
7289 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7290 // (because of the predicated logic executed earlier).
7291 // so we cast it here safely.
7292 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7293
7294 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7295 if (embeddedCipherObj == nullptr) return false;
7296
7297 // cast it to what we know it will be at runtime
7298 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7299 assert(tinst != nullptr, "ECB obj is null");
7300 assert(tinst->is_loaded(), "ECB obj is not loaded");
7301 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7302 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7303
7304 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7305 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7306 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7307 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7308 aescrypt_object = _gvn.transform(aescrypt_object);
7309
7310 // we need to get the start of the aescrypt_object's expanded key array
7311 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7312 if (k_start == nullptr) return false;
7313
7314 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7315 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7316 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7317 stubAddr, stubName, TypePtr::BOTTOM,
7318 src_start, dest_start, k_start, len);
7319
7320 // return cipher length (int)
7321 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7322 set_result(retvalue);
7323 return true;
7324 }
7325
7326 //------------------------------inline_counterMode_AESCrypt-----------------------
7327 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7328 assert(UseAES, "need AES instruction support");
7329 if (!UseAESCTRIntrinsics) return false;
7330
7331 address stubAddr = nullptr;
7332 const char *stubName = nullptr;
7333 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7334 stubAddr = StubRoutines::counterMode_AESCrypt();
7335 stubName = "counterMode_AESCrypt";
7336 }
7337 if (stubAddr == nullptr) return false;
7338
7339 Node* counterMode_object = argument(0);
7340 Node* src = argument(1);
7341 Node* src_offset = argument(2);
7342 Node* len = argument(3);
7343 Node* dest = argument(4);
7344 Node* dest_offset = argument(5);
7345
7346 // (1) src and dest are arrays.
7347 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7348 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7349 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7350 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7351
7352 // checks are the responsibility of the caller
7353 Node* src_start = src;
7354 Node* dest_start = dest;
7355 if (src_offset != nullptr || dest_offset != nullptr) {
7356 assert(src_offset != nullptr && dest_offset != nullptr, "");
7357 src_start = array_element_address(src, src_offset, T_BYTE);
7358 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7359 }
7360
7361 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7362 // (because of the predicated logic executed earlier).
7363 // so we cast it here safely.
7364 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7365 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7366 if (embeddedCipherObj == nullptr) return false;
7367 // cast it to what we know it will be at runtime
7368 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7369 assert(tinst != nullptr, "CTR obj is null");
7370 assert(tinst->is_loaded(), "CTR obj is not loaded");
7371 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7372 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7373 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7374 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7375 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7376 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7377 aescrypt_object = _gvn.transform(aescrypt_object);
7378 // we need to get the start of the aescrypt_object's expanded key array
7379 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7380 if (k_start == nullptr) return false;
7381 // similarly, get the start address of the r vector
7382 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7383 if (obj_counter == nullptr) return false;
7384 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7385
7386 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7387 if (saved_encCounter == nullptr) return false;
7388 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7389 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7390
7391 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7392 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7393 OptoRuntime::counterMode_aescrypt_Type(),
7394 stubAddr, stubName, TypePtr::BOTTOM,
7395 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7396
7397 // return cipher length (int)
7398 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7399 set_result(retvalue);
7400 return true;
7401 }
7402
7403 //------------------------------get_key_start_from_aescrypt_object-----------------------
7404 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7405 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7406 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7407 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7408 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7409 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7410 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7411 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7412 if (objSessionK == nullptr) {
7413 return (Node *) nullptr;
7414 }
7415 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7416 #else
7417 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7418 #endif // PPC64
7419 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7420 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7421
7422 // now have the array, need to get the start address of the K array
7423 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7424 return k_start;
7425 }
7426
7427 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7428 // Return node representing slow path of predicate check.
7429 // the pseudo code we want to emulate with this predicate is:
7430 // for encryption:
7431 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7432 // for decryption:
7433 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7434 // note cipher==plain is more conservative than the original java code but that's OK
7435 //
7436 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7437 // The receiver was checked for null already.
7438 Node* objCBC = argument(0);
7439
7440 Node* src = argument(1);
7441 Node* dest = argument(4);
7442
7443 // Load embeddedCipher field of CipherBlockChaining object.
7444 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7445
7446 // get AESCrypt klass for instanceOf check
7447 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7448 // will have same classloader as CipherBlockChaining object
7449 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7450 assert(tinst != nullptr, "CBCobj is null");
7451 assert(tinst->is_loaded(), "CBCobj is not loaded");
7452
7453 // we want to do an instanceof comparison against the AESCrypt class
7454 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7455 if (!klass_AESCrypt->is_loaded()) {
7456 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7457 Node* ctrl = control();
7458 set_control(top()); // no regular fast path
7459 return ctrl;
7460 }
7461
7462 src = must_be_not_null(src, true);
7463 dest = must_be_not_null(dest, true);
7464
7465 // Resolve oops to stable for CmpP below.
7466 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7467
7468 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7469 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7470 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7471
7472 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7473
7474 // for encryption, we are done
7475 if (!decrypting)
7476 return instof_false; // even if it is null
7477
7478 // for decryption, we need to add a further check to avoid
7479 // taking the intrinsic path when cipher and plain are the same
7480 // see the original java code for why.
7481 RegionNode* region = new RegionNode(3);
7482 region->init_req(1, instof_false);
7483
7484 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7485 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7486 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7487 region->init_req(2, src_dest_conjoint);
7488
7489 record_for_igvn(region);
7490 return _gvn.transform(region);
7491 }
7492
7493 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7494 // Return node representing slow path of predicate check.
7495 // the pseudo code we want to emulate with this predicate is:
7496 // for encryption:
7497 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7498 // for decryption:
7499 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7500 // note cipher==plain is more conservative than the original java code but that's OK
7501 //
7502 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7503 // The receiver was checked for null already.
7504 Node* objECB = argument(0);
7505
7506 // Load embeddedCipher field of ElectronicCodeBook object.
7507 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7508
7509 // get AESCrypt klass for instanceOf check
7510 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7511 // will have same classloader as ElectronicCodeBook object
7512 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7513 assert(tinst != nullptr, "ECBobj is null");
7514 assert(tinst->is_loaded(), "ECBobj is not loaded");
7515
7516 // we want to do an instanceof comparison against the AESCrypt class
7517 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7518 if (!klass_AESCrypt->is_loaded()) {
7519 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7520 Node* ctrl = control();
7521 set_control(top()); // no regular fast path
7522 return ctrl;
7523 }
7524 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7525
7526 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7527 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7528 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7529
7530 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7531
7532 // for encryption, we are done
7533 if (!decrypting)
7534 return instof_false; // even if it is null
7535
7536 // for decryption, we need to add a further check to avoid
7537 // taking the intrinsic path when cipher and plain are the same
7538 // see the original java code for why.
7539 RegionNode* region = new RegionNode(3);
7540 region->init_req(1, instof_false);
7541 Node* src = argument(1);
7542 Node* dest = argument(4);
7543 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7544 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7545 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7546 region->init_req(2, src_dest_conjoint);
7547
7548 record_for_igvn(region);
7549 return _gvn.transform(region);
7550 }
7551
7552 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7553 // Return node representing slow path of predicate check.
7554 // the pseudo code we want to emulate with this predicate is:
7555 // for encryption:
7556 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7557 // for decryption:
7558 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7559 // note cipher==plain is more conservative than the original java code but that's OK
7560 //
7561
7562 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7563 // The receiver was checked for null already.
7564 Node* objCTR = argument(0);
7565
7566 // Load embeddedCipher field of CipherBlockChaining object.
7567 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7568
7569 // get AESCrypt klass for instanceOf check
7570 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7571 // will have same classloader as CipherBlockChaining object
7572 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7573 assert(tinst != nullptr, "CTRobj is null");
7574 assert(tinst->is_loaded(), "CTRobj is not loaded");
7575
7576 // we want to do an instanceof comparison against the AESCrypt class
7577 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7578 if (!klass_AESCrypt->is_loaded()) {
7579 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7580 Node* ctrl = control();
7581 set_control(top()); // no regular fast path
7582 return ctrl;
7583 }
7584
7585 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7586 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7587 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7588 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7589 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7590
7591 return instof_false; // even if it is null
7592 }
7593
7594 //------------------------------inline_ghash_processBlocks
7595 bool LibraryCallKit::inline_ghash_processBlocks() {
7596 address stubAddr;
7597 const char *stubName;
7598 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7599
7600 stubAddr = StubRoutines::ghash_processBlocks();
7601 stubName = "ghash_processBlocks";
7602
7603 Node* data = argument(0);
7604 Node* offset = argument(1);
7605 Node* len = argument(2);
7606 Node* state = argument(3);
7607 Node* subkeyH = argument(4);
7608
7609 state = must_be_not_null(state, true);
7610 subkeyH = must_be_not_null(subkeyH, true);
7611 data = must_be_not_null(data, true);
7612
7613 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7614 assert(state_start, "state is null");
7615 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7616 assert(subkeyH_start, "subkeyH is null");
7617 Node* data_start = array_element_address(data, offset, T_BYTE);
7618 assert(data_start, "data is null");
7619
7620 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7621 OptoRuntime::ghash_processBlocks_Type(),
7622 stubAddr, stubName, TypePtr::BOTTOM,
7623 state_start, subkeyH_start, data_start, len);
7624 return true;
7625 }
7626
7627 //------------------------------inline_chacha20Block
7628 bool LibraryCallKit::inline_chacha20Block() {
7629 address stubAddr;
7630 const char *stubName;
7631 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7632
7633 stubAddr = StubRoutines::chacha20Block();
7634 stubName = "chacha20Block";
7635
7636 Node* state = argument(0);
7637 Node* result = argument(1);
7638
7639 state = must_be_not_null(state, true);
7640 result = must_be_not_null(result, true);
7641
7642 Node* state_start = array_element_address(state, intcon(0), T_INT);
7643 assert(state_start, "state is null");
7644 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7645 assert(result_start, "result is null");
7646
7647 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7648 OptoRuntime::chacha20Block_Type(),
7649 stubAddr, stubName, TypePtr::BOTTOM,
7650 state_start, result_start);
7651 // return key stream length (int)
7652 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7653 set_result(retvalue);
7654 return true;
7655 }
7656
7657 //------------------------------inline_kyberNtt
7658 bool LibraryCallKit::inline_kyberNtt() {
7659 address stubAddr;
7660 const char *stubName;
7661 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7662 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7663
7664 stubAddr = StubRoutines::kyberNtt();
7665 stubName = "kyberNtt";
7666 if (!stubAddr) return false;
7667
7668 Node* coeffs = argument(0);
7669 Node* ntt_zetas = argument(1);
7670
7671 coeffs = must_be_not_null(coeffs, true);
7672 ntt_zetas = must_be_not_null(ntt_zetas, true);
7673
7674 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7675 assert(coeffs_start, "coeffs is null");
7676 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7677 assert(ntt_zetas_start, "ntt_zetas is null");
7678 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7679 OptoRuntime::kyberNtt_Type(),
7680 stubAddr, stubName, TypePtr::BOTTOM,
7681 coeffs_start, ntt_zetas_start);
7682 // return an int
7683 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7684 set_result(retvalue);
7685 return true;
7686 }
7687
7688 //------------------------------inline_kyberInverseNtt
7689 bool LibraryCallKit::inline_kyberInverseNtt() {
7690 address stubAddr;
7691 const char *stubName;
7692 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7693 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7694
7695 stubAddr = StubRoutines::kyberInverseNtt();
7696 stubName = "kyberInverseNtt";
7697 if (!stubAddr) return false;
7698
7699 Node* coeffs = argument(0);
7700 Node* zetas = argument(1);
7701
7702 coeffs = must_be_not_null(coeffs, true);
7703 zetas = must_be_not_null(zetas, true);
7704
7705 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7706 assert(coeffs_start, "coeffs is null");
7707 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7708 assert(zetas_start, "inverseNtt_zetas is null");
7709 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7710 OptoRuntime::kyberInverseNtt_Type(),
7711 stubAddr, stubName, TypePtr::BOTTOM,
7712 coeffs_start, zetas_start);
7713
7714 // return an int
7715 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7716 set_result(retvalue);
7717 return true;
7718 }
7719
7720 //------------------------------inline_kyberNttMult
7721 bool LibraryCallKit::inline_kyberNttMult() {
7722 address stubAddr;
7723 const char *stubName;
7724 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7725 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7726
7727 stubAddr = StubRoutines::kyberNttMult();
7728 stubName = "kyberNttMult";
7729 if (!stubAddr) return false;
7730
7731 Node* result = argument(0);
7732 Node* ntta = argument(1);
7733 Node* nttb = argument(2);
7734 Node* zetas = argument(3);
7735
7736 result = must_be_not_null(result, true);
7737 ntta = must_be_not_null(ntta, true);
7738 nttb = must_be_not_null(nttb, true);
7739 zetas = must_be_not_null(zetas, true);
7740 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7741 assert(result_start, "result is null");
7742 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7743 assert(ntta_start, "ntta is null");
7744 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7745 assert(nttb_start, "nttb is null");
7746 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7747 assert(zetas_start, "nttMult_zetas is null");
7748 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7749 OptoRuntime::kyberNttMult_Type(),
7750 stubAddr, stubName, TypePtr::BOTTOM,
7751 result_start, ntta_start, nttb_start,
7752 zetas_start);
7753
7754 // return an int
7755 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7756 set_result(retvalue);
7757
7758 return true;
7759 }
7760
7761 //------------------------------inline_kyberAddPoly_2
7762 bool LibraryCallKit::inline_kyberAddPoly_2() {
7763 address stubAddr;
7764 const char *stubName;
7765 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7766 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7767
7768 stubAddr = StubRoutines::kyberAddPoly_2();
7769 stubName = "kyberAddPoly_2";
7770 if (!stubAddr) return false;
7771
7772 Node* result = argument(0);
7773 Node* a = argument(1);
7774 Node* b = argument(2);
7775
7776 result = must_be_not_null(result, true);
7777 a = must_be_not_null(a, true);
7778 b = must_be_not_null(b, true);
7779
7780 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7781 assert(result_start, "result is null");
7782 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7783 assert(a_start, "a is null");
7784 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7785 assert(b_start, "b is null");
7786 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7787 OptoRuntime::kyberAddPoly_2_Type(),
7788 stubAddr, stubName, TypePtr::BOTTOM,
7789 result_start, a_start, b_start);
7790 // return an int
7791 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7792 set_result(retvalue);
7793 return true;
7794 }
7795
7796 //------------------------------inline_kyberAddPoly_3
7797 bool LibraryCallKit::inline_kyberAddPoly_3() {
7798 address stubAddr;
7799 const char *stubName;
7800 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7801 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7802
7803 stubAddr = StubRoutines::kyberAddPoly_3();
7804 stubName = "kyberAddPoly_3";
7805 if (!stubAddr) return false;
7806
7807 Node* result = argument(0);
7808 Node* a = argument(1);
7809 Node* b = argument(2);
7810 Node* c = argument(3);
7811
7812 result = must_be_not_null(result, true);
7813 a = must_be_not_null(a, true);
7814 b = must_be_not_null(b, true);
7815 c = must_be_not_null(c, true);
7816
7817 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7818 assert(result_start, "result is null");
7819 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7820 assert(a_start, "a is null");
7821 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7822 assert(b_start, "b is null");
7823 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7824 assert(c_start, "c is null");
7825 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7826 OptoRuntime::kyberAddPoly_3_Type(),
7827 stubAddr, stubName, TypePtr::BOTTOM,
7828 result_start, a_start, b_start, c_start);
7829 // return an int
7830 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7831 set_result(retvalue);
7832 return true;
7833 }
7834
7835 //------------------------------inline_kyber12To16
7836 bool LibraryCallKit::inline_kyber12To16() {
7837 address stubAddr;
7838 const char *stubName;
7839 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7840 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7841
7842 stubAddr = StubRoutines::kyber12To16();
7843 stubName = "kyber12To16";
7844 if (!stubAddr) return false;
7845
7846 Node* condensed = argument(0);
7847 Node* condensedOffs = argument(1);
7848 Node* parsed = argument(2);
7849 Node* parsedLength = argument(3);
7850
7851 condensed = must_be_not_null(condensed, true);
7852 parsed = must_be_not_null(parsed, true);
7853
7854 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7855 assert(condensed_start, "condensed is null");
7856 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7857 assert(parsed_start, "parsed is null");
7858 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7859 OptoRuntime::kyber12To16_Type(),
7860 stubAddr, stubName, TypePtr::BOTTOM,
7861 condensed_start, condensedOffs, parsed_start, parsedLength);
7862 // return an int
7863 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7864 set_result(retvalue);
7865 return true;
7866
7867 }
7868
7869 //------------------------------inline_kyberBarrettReduce
7870 bool LibraryCallKit::inline_kyberBarrettReduce() {
7871 address stubAddr;
7872 const char *stubName;
7873 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7874 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7875
7876 stubAddr = StubRoutines::kyberBarrettReduce();
7877 stubName = "kyberBarrettReduce";
7878 if (!stubAddr) return false;
7879
7880 Node* coeffs = argument(0);
7881
7882 coeffs = must_be_not_null(coeffs, true);
7883
7884 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7885 assert(coeffs_start, "coeffs is null");
7886 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7887 OptoRuntime::kyberBarrettReduce_Type(),
7888 stubAddr, stubName, TypePtr::BOTTOM,
7889 coeffs_start);
7890 // return an int
7891 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7892 set_result(retvalue);
7893 return true;
7894 }
7895
7896 //------------------------------inline_dilithiumAlmostNtt
7897 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7898 address stubAddr;
7899 const char *stubName;
7900 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7901 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7902
7903 stubAddr = StubRoutines::dilithiumAlmostNtt();
7904 stubName = "dilithiumAlmostNtt";
7905 if (!stubAddr) return false;
7906
7907 Node* coeffs = argument(0);
7908 Node* ntt_zetas = argument(1);
7909
7910 coeffs = must_be_not_null(coeffs, true);
7911 ntt_zetas = must_be_not_null(ntt_zetas, true);
7912
7913 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7914 assert(coeffs_start, "coeffs is null");
7915 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7916 assert(ntt_zetas_start, "ntt_zetas is null");
7917 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7918 OptoRuntime::dilithiumAlmostNtt_Type(),
7919 stubAddr, stubName, TypePtr::BOTTOM,
7920 coeffs_start, ntt_zetas_start);
7921 // return an int
7922 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7923 set_result(retvalue);
7924 return true;
7925 }
7926
7927 //------------------------------inline_dilithiumAlmostInverseNtt
7928 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7929 address stubAddr;
7930 const char *stubName;
7931 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7932 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7933
7934 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7935 stubName = "dilithiumAlmostInverseNtt";
7936 if (!stubAddr) return false;
7937
7938 Node* coeffs = argument(0);
7939 Node* zetas = argument(1);
7940
7941 coeffs = must_be_not_null(coeffs, true);
7942 zetas = must_be_not_null(zetas, true);
7943
7944 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7945 assert(coeffs_start, "coeffs is null");
7946 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7947 assert(zetas_start, "inverseNtt_zetas is null");
7948 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7949 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7950 stubAddr, stubName, TypePtr::BOTTOM,
7951 coeffs_start, zetas_start);
7952 // return an int
7953 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7954 set_result(retvalue);
7955 return true;
7956 }
7957
7958 //------------------------------inline_dilithiumNttMult
7959 bool LibraryCallKit::inline_dilithiumNttMult() {
7960 address stubAddr;
7961 const char *stubName;
7962 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7963 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7964
7965 stubAddr = StubRoutines::dilithiumNttMult();
7966 stubName = "dilithiumNttMult";
7967 if (!stubAddr) return false;
7968
7969 Node* result = argument(0);
7970 Node* ntta = argument(1);
7971 Node* nttb = argument(2);
7972 Node* zetas = argument(3);
7973
7974 result = must_be_not_null(result, true);
7975 ntta = must_be_not_null(ntta, true);
7976 nttb = must_be_not_null(nttb, true);
7977 zetas = must_be_not_null(zetas, true);
7978
7979 Node* result_start = array_element_address(result, intcon(0), T_INT);
7980 assert(result_start, "result is null");
7981 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
7982 assert(ntta_start, "ntta is null");
7983 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
7984 assert(nttb_start, "nttb is null");
7985 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7986 OptoRuntime::dilithiumNttMult_Type(),
7987 stubAddr, stubName, TypePtr::BOTTOM,
7988 result_start, ntta_start, nttb_start);
7989
7990 // return an int
7991 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
7992 set_result(retvalue);
7993
7994 return true;
7995 }
7996
7997 //------------------------------inline_dilithiumMontMulByConstant
7998 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
7999 address stubAddr;
8000 const char *stubName;
8001 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8002 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8003
8004 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8005 stubName = "dilithiumMontMulByConstant";
8006 if (!stubAddr) return false;
8007
8008 Node* coeffs = argument(0);
8009 Node* constant = argument(1);
8010
8011 coeffs = must_be_not_null(coeffs, true);
8012
8013 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8014 assert(coeffs_start, "coeffs is null");
8015 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8016 OptoRuntime::dilithiumMontMulByConstant_Type(),
8017 stubAddr, stubName, TypePtr::BOTTOM,
8018 coeffs_start, constant);
8019
8020 // return an int
8021 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8022 set_result(retvalue);
8023 return true;
8024 }
8025
8026
8027 //------------------------------inline_dilithiumDecomposePoly
8028 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8029 address stubAddr;
8030 const char *stubName;
8031 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8032 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8033
8034 stubAddr = StubRoutines::dilithiumDecomposePoly();
8035 stubName = "dilithiumDecomposePoly";
8036 if (!stubAddr) return false;
8037
8038 Node* input = argument(0);
8039 Node* lowPart = argument(1);
8040 Node* highPart = argument(2);
8041 Node* twoGamma2 = argument(3);
8042 Node* multiplier = argument(4);
8043
8044 input = must_be_not_null(input, true);
8045 lowPart = must_be_not_null(lowPart, true);
8046 highPart = must_be_not_null(highPart, true);
8047
8048 Node* input_start = array_element_address(input, intcon(0), T_INT);
8049 assert(input_start, "input is null");
8050 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8051 assert(lowPart_start, "lowPart is null");
8052 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8053 assert(highPart_start, "highPart is null");
8054
8055 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8056 OptoRuntime::dilithiumDecomposePoly_Type(),
8057 stubAddr, stubName, TypePtr::BOTTOM,
8058 input_start, lowPart_start, highPart_start,
8059 twoGamma2, multiplier);
8060
8061 // return an int
8062 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8063 set_result(retvalue);
8064 return true;
8065 }
8066
8067 bool LibraryCallKit::inline_base64_encodeBlock() {
8068 address stubAddr;
8069 const char *stubName;
8070 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8071 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8072 stubAddr = StubRoutines::base64_encodeBlock();
8073 stubName = "encodeBlock";
8074
8075 if (!stubAddr) return false;
8076 Node* base64obj = argument(0);
8077 Node* src = argument(1);
8078 Node* offset = argument(2);
8079 Node* len = argument(3);
8080 Node* dest = argument(4);
8081 Node* dp = argument(5);
8082 Node* isURL = argument(6);
8083
8084 src = must_be_not_null(src, true);
8085 dest = must_be_not_null(dest, true);
8086
8087 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8088 assert(src_start, "source array is null");
8089 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8090 assert(dest_start, "destination array is null");
8091
8092 Node* base64 = make_runtime_call(RC_LEAF,
8093 OptoRuntime::base64_encodeBlock_Type(),
8094 stubAddr, stubName, TypePtr::BOTTOM,
8095 src_start, offset, len, dest_start, dp, isURL);
8096 return true;
8097 }
8098
8099 bool LibraryCallKit::inline_base64_decodeBlock() {
8100 address stubAddr;
8101 const char *stubName;
8102 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8103 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8104 stubAddr = StubRoutines::base64_decodeBlock();
8105 stubName = "decodeBlock";
8106
8107 if (!stubAddr) return false;
8108 Node* base64obj = argument(0);
8109 Node* src = argument(1);
8110 Node* src_offset = argument(2);
8111 Node* len = argument(3);
8112 Node* dest = argument(4);
8113 Node* dest_offset = argument(5);
8114 Node* isURL = argument(6);
8115 Node* isMIME = argument(7);
8116
8117 src = must_be_not_null(src, true);
8118 dest = must_be_not_null(dest, true);
8119
8120 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8121 assert(src_start, "source array is null");
8122 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8123 assert(dest_start, "destination array is null");
8124
8125 Node* call = make_runtime_call(RC_LEAF,
8126 OptoRuntime::base64_decodeBlock_Type(),
8127 stubAddr, stubName, TypePtr::BOTTOM,
8128 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8129 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8130 set_result(result);
8131 return true;
8132 }
8133
8134 bool LibraryCallKit::inline_poly1305_processBlocks() {
8135 address stubAddr;
8136 const char *stubName;
8137 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8138 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8139 stubAddr = StubRoutines::poly1305_processBlocks();
8140 stubName = "poly1305_processBlocks";
8141
8142 if (!stubAddr) return false;
8143 null_check_receiver(); // null-check receiver
8144 if (stopped()) return true;
8145
8146 Node* input = argument(1);
8147 Node* input_offset = argument(2);
8148 Node* len = argument(3);
8149 Node* alimbs = argument(4);
8150 Node* rlimbs = argument(5);
8151
8152 input = must_be_not_null(input, true);
8153 alimbs = must_be_not_null(alimbs, true);
8154 rlimbs = must_be_not_null(rlimbs, true);
8155
8156 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8157 assert(input_start, "input array is null");
8158 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8159 assert(acc_start, "acc array is null");
8160 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8161 assert(r_start, "r array is null");
8162
8163 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8164 OptoRuntime::poly1305_processBlocks_Type(),
8165 stubAddr, stubName, TypePtr::BOTTOM,
8166 input_start, len, acc_start, r_start);
8167 return true;
8168 }
8169
8170 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8171 address stubAddr;
8172 const char *stubName;
8173 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8174 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8175 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8176 stubName = "intpoly_montgomeryMult_P256";
8177
8178 if (!stubAddr) return false;
8179 null_check_receiver(); // null-check receiver
8180 if (stopped()) return true;
8181
8182 Node* a = argument(1);
8183 Node* b = argument(2);
8184 Node* r = argument(3);
8185
8186 a = must_be_not_null(a, true);
8187 b = must_be_not_null(b, true);
8188 r = must_be_not_null(r, true);
8189
8190 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8191 assert(a_start, "a array is null");
8192 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8193 assert(b_start, "b array is null");
8194 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8195 assert(r_start, "r array is null");
8196
8197 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8198 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8199 stubAddr, stubName, TypePtr::BOTTOM,
8200 a_start, b_start, r_start);
8201 return true;
8202 }
8203
8204 bool LibraryCallKit::inline_intpoly_assign() {
8205 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8206 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8207 const char *stubName = "intpoly_assign";
8208 address stubAddr = StubRoutines::intpoly_assign();
8209 if (!stubAddr) return false;
8210
8211 Node* set = argument(0);
8212 Node* a = argument(1);
8213 Node* b = argument(2);
8214 Node* arr_length = load_array_length(a);
8215
8216 a = must_be_not_null(a, true);
8217 b = must_be_not_null(b, true);
8218
8219 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8220 assert(a_start, "a array is null");
8221 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8222 assert(b_start, "b array is null");
8223
8224 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8225 OptoRuntime::intpoly_assign_Type(),
8226 stubAddr, stubName, TypePtr::BOTTOM,
8227 set, a_start, b_start, arr_length);
8228 return true;
8229 }
8230
8231 //------------------------------inline_digestBase_implCompress-----------------------
8232 //
8233 // Calculate MD5 for single-block byte[] array.
8234 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8235 //
8236 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8237 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8238 //
8239 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8240 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8241 //
8242 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8243 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8244 //
8245 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8246 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8247 //
8248 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8249 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8250
8251 Node* digestBase_obj = argument(0);
8252 Node* src = argument(1); // type oop
8253 Node* ofs = argument(2); // type int
8254
8255 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8256 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8257 // failed array check
8258 return false;
8259 }
8260 // Figure out the size and type of the elements we will be copying.
8261 BasicType src_elem = src_type->elem()->array_element_basic_type();
8262 if (src_elem != T_BYTE) {
8263 return false;
8264 }
8265 // 'src_start' points to src array + offset
8266 src = must_be_not_null(src, true);
8267 Node* src_start = array_element_address(src, ofs, src_elem);
8268 Node* state = nullptr;
8269 Node* block_size = nullptr;
8270 address stubAddr;
8271 const char *stubName;
8272
8273 switch(id) {
8274 case vmIntrinsics::_md5_implCompress:
8275 assert(UseMD5Intrinsics, "need MD5 instruction support");
8276 state = get_state_from_digest_object(digestBase_obj, T_INT);
8277 stubAddr = StubRoutines::md5_implCompress();
8278 stubName = "md5_implCompress";
8279 break;
8280 case vmIntrinsics::_sha_implCompress:
8281 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8282 state = get_state_from_digest_object(digestBase_obj, T_INT);
8283 stubAddr = StubRoutines::sha1_implCompress();
8284 stubName = "sha1_implCompress";
8285 break;
8286 case vmIntrinsics::_sha2_implCompress:
8287 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8288 state = get_state_from_digest_object(digestBase_obj, T_INT);
8289 stubAddr = StubRoutines::sha256_implCompress();
8290 stubName = "sha256_implCompress";
8291 break;
8292 case vmIntrinsics::_sha5_implCompress:
8293 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8294 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8295 stubAddr = StubRoutines::sha512_implCompress();
8296 stubName = "sha512_implCompress";
8297 break;
8298 case vmIntrinsics::_sha3_implCompress:
8299 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8300 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8301 stubAddr = StubRoutines::sha3_implCompress();
8302 stubName = "sha3_implCompress";
8303 block_size = get_block_size_from_digest_object(digestBase_obj);
8304 if (block_size == nullptr) return false;
8305 break;
8306 default:
8307 fatal_unexpected_iid(id);
8308 return false;
8309 }
8310 if (state == nullptr) return false;
8311
8312 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8313 if (stubAddr == nullptr) return false;
8314
8315 // Call the stub.
8316 Node* call;
8317 if (block_size == nullptr) {
8318 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8319 stubAddr, stubName, TypePtr::BOTTOM,
8320 src_start, state);
8321 } else {
8322 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8323 stubAddr, stubName, TypePtr::BOTTOM,
8324 src_start, state, block_size);
8325 }
8326
8327 return true;
8328 }
8329
8330 //------------------------------inline_double_keccak
8331 bool LibraryCallKit::inline_double_keccak() {
8332 address stubAddr;
8333 const char *stubName;
8334 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8335 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8336
8337 stubAddr = StubRoutines::double_keccak();
8338 stubName = "double_keccak";
8339 if (!stubAddr) return false;
8340
8341 Node* status0 = argument(0);
8342 Node* status1 = argument(1);
8343
8344 status0 = must_be_not_null(status0, true);
8345 status1 = must_be_not_null(status1, true);
8346
8347 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8348 assert(status0_start, "status0 is null");
8349 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8350 assert(status1_start, "status1 is null");
8351 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8352 OptoRuntime::double_keccak_Type(),
8353 stubAddr, stubName, TypePtr::BOTTOM,
8354 status0_start, status1_start);
8355 // return an int
8356 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8357 set_result(retvalue);
8358 return true;
8359 }
8360
8361
8362 //------------------------------inline_digestBase_implCompressMB-----------------------
8363 //
8364 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8365 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8366 //
8367 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8368 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8369 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8370 assert((uint)predicate < 5, "sanity");
8371 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8372
8373 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8374 Node* src = argument(1); // byte[] array
8375 Node* ofs = argument(2); // type int
8376 Node* limit = argument(3); // type int
8377
8378 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8379 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8380 // failed array check
8381 return false;
8382 }
8383 // Figure out the size and type of the elements we will be copying.
8384 BasicType src_elem = src_type->elem()->array_element_basic_type();
8385 if (src_elem != T_BYTE) {
8386 return false;
8387 }
8388 // 'src_start' points to src array + offset
8389 src = must_be_not_null(src, false);
8390 Node* src_start = array_element_address(src, ofs, src_elem);
8391
8392 const char* klass_digestBase_name = nullptr;
8393 const char* stub_name = nullptr;
8394 address stub_addr = nullptr;
8395 BasicType elem_type = T_INT;
8396
8397 switch (predicate) {
8398 case 0:
8399 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8400 klass_digestBase_name = "sun/security/provider/MD5";
8401 stub_name = "md5_implCompressMB";
8402 stub_addr = StubRoutines::md5_implCompressMB();
8403 }
8404 break;
8405 case 1:
8406 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8407 klass_digestBase_name = "sun/security/provider/SHA";
8408 stub_name = "sha1_implCompressMB";
8409 stub_addr = StubRoutines::sha1_implCompressMB();
8410 }
8411 break;
8412 case 2:
8413 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8414 klass_digestBase_name = "sun/security/provider/SHA2";
8415 stub_name = "sha256_implCompressMB";
8416 stub_addr = StubRoutines::sha256_implCompressMB();
8417 }
8418 break;
8419 case 3:
8420 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8421 klass_digestBase_name = "sun/security/provider/SHA5";
8422 stub_name = "sha512_implCompressMB";
8423 stub_addr = StubRoutines::sha512_implCompressMB();
8424 elem_type = T_LONG;
8425 }
8426 break;
8427 case 4:
8428 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8429 klass_digestBase_name = "sun/security/provider/SHA3";
8430 stub_name = "sha3_implCompressMB";
8431 stub_addr = StubRoutines::sha3_implCompressMB();
8432 elem_type = T_LONG;
8433 }
8434 break;
8435 default:
8436 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8437 }
8438 if (klass_digestBase_name != nullptr) {
8439 assert(stub_addr != nullptr, "Stub is generated");
8440 if (stub_addr == nullptr) return false;
8441
8442 // get DigestBase klass to lookup for SHA klass
8443 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8444 assert(tinst != nullptr, "digestBase_obj is not instance???");
8445 assert(tinst->is_loaded(), "DigestBase is not loaded");
8446
8447 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8448 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8449 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8450 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8451 }
8452 return false;
8453 }
8454
8455 //------------------------------inline_digestBase_implCompressMB-----------------------
8456 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8457 BasicType elem_type, address stubAddr, const char *stubName,
8458 Node* src_start, Node* ofs, Node* limit) {
8459 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8460 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8461 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8462 digest_obj = _gvn.transform(digest_obj);
8463
8464 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8465 if (state == nullptr) return false;
8466
8467 Node* block_size = nullptr;
8468 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8469 block_size = get_block_size_from_digest_object(digest_obj);
8470 if (block_size == nullptr) return false;
8471 }
8472
8473 // Call the stub.
8474 Node* call;
8475 if (block_size == nullptr) {
8476 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8477 OptoRuntime::digestBase_implCompressMB_Type(false),
8478 stubAddr, stubName, TypePtr::BOTTOM,
8479 src_start, state, ofs, limit);
8480 } else {
8481 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8482 OptoRuntime::digestBase_implCompressMB_Type(true),
8483 stubAddr, stubName, TypePtr::BOTTOM,
8484 src_start, state, block_size, ofs, limit);
8485 }
8486
8487 // return ofs (int)
8488 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8489 set_result(result);
8490
8491 return true;
8492 }
8493
8494 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8495 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8496 assert(UseAES, "need AES instruction support");
8497 address stubAddr = nullptr;
8498 const char *stubName = nullptr;
8499 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8500 stubName = "galoisCounterMode_AESCrypt";
8501
8502 if (stubAddr == nullptr) return false;
8503
8504 Node* in = argument(0);
8505 Node* inOfs = argument(1);
8506 Node* len = argument(2);
8507 Node* ct = argument(3);
8508 Node* ctOfs = argument(4);
8509 Node* out = argument(5);
8510 Node* outOfs = argument(6);
8511 Node* gctr_object = argument(7);
8512 Node* ghash_object = argument(8);
8513
8514 // (1) in, ct and out are arrays.
8515 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8516 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8517 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8518 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8519 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8520 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8521
8522 // checks are the responsibility of the caller
8523 Node* in_start = in;
8524 Node* ct_start = ct;
8525 Node* out_start = out;
8526 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8527 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8528 in_start = array_element_address(in, inOfs, T_BYTE);
8529 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8530 out_start = array_element_address(out, outOfs, T_BYTE);
8531 }
8532
8533 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8534 // (because of the predicated logic executed earlier).
8535 // so we cast it here safely.
8536 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8537 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8538 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8539 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8540 Node* state = load_field_from_object(ghash_object, "state", "[J");
8541
8542 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8543 return false;
8544 }
8545 // cast it to what we know it will be at runtime
8546 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8547 assert(tinst != nullptr, "GCTR obj is null");
8548 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8549 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8550 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8551 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8552 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8553 const TypeOopPtr* xtype = aklass->as_instance_type();
8554 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8555 aescrypt_object = _gvn.transform(aescrypt_object);
8556 // we need to get the start of the aescrypt_object's expanded key array
8557 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8558 if (k_start == nullptr) return false;
8559 // similarly, get the start address of the r vector
8560 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8561 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8562 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8563
8564
8565 // Call the stub, passing params
8566 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8567 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8568 stubAddr, stubName, TypePtr::BOTTOM,
8569 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8570
8571 // return cipher length (int)
8572 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8573 set_result(retvalue);
8574
8575 return true;
8576 }
8577
8578 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8579 // Return node representing slow path of predicate check.
8580 // the pseudo code we want to emulate with this predicate is:
8581 // for encryption:
8582 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8583 // for decryption:
8584 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8585 // note cipher==plain is more conservative than the original java code but that's OK
8586 //
8587
8588 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8589 // The receiver was checked for null already.
8590 Node* objGCTR = argument(7);
8591 // Load embeddedCipher field of GCTR object.
8592 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8593 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8594
8595 // get AESCrypt klass for instanceOf check
8596 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8597 // will have same classloader as CipherBlockChaining object
8598 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8599 assert(tinst != nullptr, "GCTR obj is null");
8600 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8601
8602 // we want to do an instanceof comparison against the AESCrypt class
8603 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8604 if (!klass_AESCrypt->is_loaded()) {
8605 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8606 Node* ctrl = control();
8607 set_control(top()); // no regular fast path
8608 return ctrl;
8609 }
8610
8611 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8612 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8613 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8614 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8615 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8616
8617 return instof_false; // even if it is null
8618 }
8619
8620 //------------------------------get_state_from_digest_object-----------------------
8621 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8622 const char* state_type;
8623 switch (elem_type) {
8624 case T_BYTE: state_type = "[B"; break;
8625 case T_INT: state_type = "[I"; break;
8626 case T_LONG: state_type = "[J"; break;
8627 default: ShouldNotReachHere();
8628 }
8629 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8630 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8631 if (digest_state == nullptr) return (Node *) nullptr;
8632
8633 // now have the array, need to get the start address of the state array
8634 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8635 return state;
8636 }
8637
8638 //------------------------------get_block_size_from_sha3_object----------------------------------
8639 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8640 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8641 assert (block_size != nullptr, "sanity");
8642 return block_size;
8643 }
8644
8645 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8646 // Return node representing slow path of predicate check.
8647 // the pseudo code we want to emulate with this predicate is:
8648 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8649 //
8650 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8651 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8652 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8653 assert((uint)predicate < 5, "sanity");
8654
8655 // The receiver was checked for null already.
8656 Node* digestBaseObj = argument(0);
8657
8658 // get DigestBase klass for instanceOf check
8659 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8660 assert(tinst != nullptr, "digestBaseObj is null");
8661 assert(tinst->is_loaded(), "DigestBase is not loaded");
8662
8663 const char* klass_name = nullptr;
8664 switch (predicate) {
8665 case 0:
8666 if (UseMD5Intrinsics) {
8667 // we want to do an instanceof comparison against the MD5 class
8668 klass_name = "sun/security/provider/MD5";
8669 }
8670 break;
8671 case 1:
8672 if (UseSHA1Intrinsics) {
8673 // we want to do an instanceof comparison against the SHA class
8674 klass_name = "sun/security/provider/SHA";
8675 }
8676 break;
8677 case 2:
8678 if (UseSHA256Intrinsics) {
8679 // we want to do an instanceof comparison against the SHA2 class
8680 klass_name = "sun/security/provider/SHA2";
8681 }
8682 break;
8683 case 3:
8684 if (UseSHA512Intrinsics) {
8685 // we want to do an instanceof comparison against the SHA5 class
8686 klass_name = "sun/security/provider/SHA5";
8687 }
8688 break;
8689 case 4:
8690 if (UseSHA3Intrinsics) {
8691 // we want to do an instanceof comparison against the SHA3 class
8692 klass_name = "sun/security/provider/SHA3";
8693 }
8694 break;
8695 default:
8696 fatal("unknown SHA intrinsic predicate: %d", predicate);
8697 }
8698
8699 ciKlass* klass = nullptr;
8700 if (klass_name != nullptr) {
8701 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8702 }
8703 if ((klass == nullptr) || !klass->is_loaded()) {
8704 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8705 Node* ctrl = control();
8706 set_control(top()); // no intrinsic path
8707 return ctrl;
8708 }
8709 ciInstanceKlass* instklass = klass->as_instance_klass();
8710
8711 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8712 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8713 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8714 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8715
8716 return instof_false; // even if it is null
8717 }
8718
8719 //-------------inline_fma-----------------------------------
8720 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8721 Node *a = nullptr;
8722 Node *b = nullptr;
8723 Node *c = nullptr;
8724 Node* result = nullptr;
8725 switch (id) {
8726 case vmIntrinsics::_fmaD:
8727 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8728 // no receiver since it is static method
8729 a = argument(0);
8730 b = argument(2);
8731 c = argument(4);
8732 result = _gvn.transform(new FmaDNode(a, b, c));
8733 break;
8734 case vmIntrinsics::_fmaF:
8735 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8736 a = argument(0);
8737 b = argument(1);
8738 c = argument(2);
8739 result = _gvn.transform(new FmaFNode(a, b, c));
8740 break;
8741 default:
8742 fatal_unexpected_iid(id); break;
8743 }
8744 set_result(result);
8745 return true;
8746 }
8747
8748 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8749 // argument(0) is receiver
8750 Node* codePoint = argument(1);
8751 Node* n = nullptr;
8752
8753 switch (id) {
8754 case vmIntrinsics::_isDigit :
8755 n = new DigitNode(control(), codePoint);
8756 break;
8757 case vmIntrinsics::_isLowerCase :
8758 n = new LowerCaseNode(control(), codePoint);
8759 break;
8760 case vmIntrinsics::_isUpperCase :
8761 n = new UpperCaseNode(control(), codePoint);
8762 break;
8763 case vmIntrinsics::_isWhitespace :
8764 n = new WhitespaceNode(control(), codePoint);
8765 break;
8766 default:
8767 fatal_unexpected_iid(id);
8768 }
8769
8770 set_result(_gvn.transform(n));
8771 return true;
8772 }
8773
8774 bool LibraryCallKit::inline_profileBoolean() {
8775 Node* counts = argument(1);
8776 const TypeAryPtr* ary = nullptr;
8777 ciArray* aobj = nullptr;
8778 if (counts->is_Con()
8779 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8780 && (aobj = ary->const_oop()->as_array()) != nullptr
8781 && (aobj->length() == 2)) {
8782 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8783 jint false_cnt = aobj->element_value(0).as_int();
8784 jint true_cnt = aobj->element_value(1).as_int();
8785
8786 if (C->log() != nullptr) {
8787 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8788 false_cnt, true_cnt);
8789 }
8790
8791 if (false_cnt + true_cnt == 0) {
8792 // According to profile, never executed.
8793 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8794 Deoptimization::Action_reinterpret);
8795 return true;
8796 }
8797
8798 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8799 // is a number of each value occurrences.
8800 Node* result = argument(0);
8801 if (false_cnt == 0 || true_cnt == 0) {
8802 // According to profile, one value has been never seen.
8803 int expected_val = (false_cnt == 0) ? 1 : 0;
8804
8805 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8806 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8807
8808 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8809 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8810 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8811
8812 { // Slow path: uncommon trap for never seen value and then reexecute
8813 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8814 // the value has been seen at least once.
8815 PreserveJVMState pjvms(this);
8816 PreserveReexecuteState preexecs(this);
8817 jvms()->set_should_reexecute(true);
8818
8819 set_control(slow_path);
8820 set_i_o(i_o());
8821
8822 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8823 Deoptimization::Action_reinterpret);
8824 }
8825 // The guard for never seen value enables sharpening of the result and
8826 // returning a constant. It allows to eliminate branches on the same value
8827 // later on.
8828 set_control(fast_path);
8829 result = intcon(expected_val);
8830 }
8831 // Stop profiling.
8832 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8833 // By replacing method body with profile data (represented as ProfileBooleanNode
8834 // on IR level) we effectively disable profiling.
8835 // It enables full speed execution once optimized code is generated.
8836 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8837 C->record_for_igvn(profile);
8838 set_result(profile);
8839 return true;
8840 } else {
8841 // Continue profiling.
8842 // Profile data isn't available at the moment. So, execute method's bytecode version.
8843 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8844 // is compiled and counters aren't available since corresponding MethodHandle
8845 // isn't a compile-time constant.
8846 return false;
8847 }
8848 }
8849
8850 bool LibraryCallKit::inline_isCompileConstant() {
8851 Node* n = argument(0);
8852 set_result(n->is_Con() ? intcon(1) : intcon(0));
8853 return true;
8854 }
8855
8856 //------------------------------- inline_getObjectSize --------------------------------------
8857 //
8858 // Calculate the runtime size of the object/array.
8859 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8860 //
8861 bool LibraryCallKit::inline_getObjectSize() {
8862 Node* obj = argument(3);
8863 Node* klass_node = load_object_klass(obj);
8864
8865 jint layout_con = Klass::_lh_neutral_value;
8866 Node* layout_val = get_layout_helper(klass_node, layout_con);
8867 int layout_is_con = (layout_val == nullptr);
8868
8869 if (layout_is_con) {
8870 // Layout helper is constant, can figure out things at compile time.
8871
8872 if (Klass::layout_helper_is_instance(layout_con)) {
8873 // Instance case: layout_con contains the size itself.
8874 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8875 set_result(size);
8876 } else {
8877 // Array case: size is round(header + element_size*arraylength).
8878 // Since arraylength is different for every array instance, we have to
8879 // compute the whole thing at runtime.
8880
8881 Node* arr_length = load_array_length(obj);
8882
8883 int round_mask = MinObjAlignmentInBytes - 1;
8884 int hsize = Klass::layout_helper_header_size(layout_con);
8885 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8886
8887 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8888 round_mask = 0; // strength-reduce it if it goes away completely
8889 }
8890 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8891 Node* header_size = intcon(hsize + round_mask);
8892
8893 Node* lengthx = ConvI2X(arr_length);
8894 Node* headerx = ConvI2X(header_size);
8895
8896 Node* abody = lengthx;
8897 if (eshift != 0) {
8898 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8899 }
8900 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8901 if (round_mask != 0) {
8902 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8903 }
8904 size = ConvX2L(size);
8905 set_result(size);
8906 }
8907 } else {
8908 // Layout helper is not constant, need to test for array-ness at runtime.
8909
8910 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8911 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8912 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8913 record_for_igvn(result_reg);
8914
8915 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8916 if (array_ctl != nullptr) {
8917 // Array case: size is round(header + element_size*arraylength).
8918 // Since arraylength is different for every array instance, we have to
8919 // compute the whole thing at runtime.
8920
8921 PreserveJVMState pjvms(this);
8922 set_control(array_ctl);
8923 Node* arr_length = load_array_length(obj);
8924
8925 int round_mask = MinObjAlignmentInBytes - 1;
8926 Node* mask = intcon(round_mask);
8927
8928 Node* hss = intcon(Klass::_lh_header_size_shift);
8929 Node* hsm = intcon(Klass::_lh_header_size_mask);
8930 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8931 header_size = _gvn.transform(new AndINode(header_size, hsm));
8932 header_size = _gvn.transform(new AddINode(header_size, mask));
8933
8934 // There is no need to mask or shift this value.
8935 // The semantics of LShiftINode include an implicit mask to 0x1F.
8936 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8937 Node* elem_shift = layout_val;
8938
8939 Node* lengthx = ConvI2X(arr_length);
8940 Node* headerx = ConvI2X(header_size);
8941
8942 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8943 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8944 if (round_mask != 0) {
8945 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8946 }
8947 size = ConvX2L(size);
8948
8949 result_reg->init_req(_array_path, control());
8950 result_val->init_req(_array_path, size);
8951 }
8952
8953 if (!stopped()) {
8954 // Instance case: the layout helper gives us instance size almost directly,
8955 // but we need to mask out the _lh_instance_slow_path_bit.
8956 Node* size = ConvI2X(layout_val);
8957 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8958 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8959 size = _gvn.transform(new AndXNode(size, mask));
8960 size = ConvX2L(size);
8961
8962 result_reg->init_req(_instance_path, control());
8963 result_val->init_req(_instance_path, size);
8964 }
8965
8966 set_result(result_reg, result_val);
8967 }
8968
8969 return true;
8970 }
8971
8972 //------------------------------- inline_blackhole --------------------------------------
8973 //
8974 // Make sure all arguments to this node are alive.
8975 // This matches methods that were requested to be blackholed through compile commands.
8976 //
8977 bool LibraryCallKit::inline_blackhole() {
8978 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8979 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8980 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8981
8982 // Blackhole node pinches only the control, not memory. This allows
8983 // the blackhole to be pinned in the loop that computes blackholed
8984 // values, but have no other side effects, like breaking the optimizations
8985 // across the blackhole.
8986
8987 Node* bh = _gvn.transform(new BlackholeNode(control()));
8988 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8989
8990 // Bind call arguments as blackhole arguments to keep them alive
8991 uint nargs = callee()->arg_size();
8992 for (uint i = 0; i < nargs; i++) {
8993 bh->add_req(argument(i));
8994 }
8995
8996 return true;
8997 }
8998
8999 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9000 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9001 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9002 return nullptr; // box klass is not Float16
9003 }
9004
9005 // Null check; get notnull casted pointer
9006 Node* null_ctl = top();
9007 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9008 // If not_null_box is dead, only null-path is taken
9009 if (stopped()) {
9010 set_control(null_ctl);
9011 return nullptr;
9012 }
9013 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9014 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9015 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9016 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9017 }
9018
9019 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9020 PreserveReexecuteState preexecs(this);
9021 jvms()->set_should_reexecute(true);
9022
9023 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9024 Node* klass_node = makecon(klass_type);
9025 Node* box = new_instance(klass_node);
9026
9027 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9028 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9029
9030 Node* field_store = _gvn.transform(access_store_at(box,
9031 value_field,
9032 value_adr_type,
9033 value,
9034 TypeInt::SHORT,
9035 T_SHORT,
9036 IN_HEAP));
9037 set_memory(field_store, value_adr_type);
9038 return box;
9039 }
9040
9041 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9042 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9043 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9044 return false;
9045 }
9046
9047 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9048 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9049 return false;
9050 }
9051
9052 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9053 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9054 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9055 ciSymbols::short_signature(),
9056 false);
9057 assert(field != nullptr, "");
9058
9059 // Transformed nodes
9060 Node* fld1 = nullptr;
9061 Node* fld2 = nullptr;
9062 Node* fld3 = nullptr;
9063 switch(num_args) {
9064 case 3:
9065 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9066 if (fld3 == nullptr) {
9067 return false;
9068 }
9069 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9070 // fall-through
9071 case 2:
9072 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9073 if (fld2 == nullptr) {
9074 return false;
9075 }
9076 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9077 // fall-through
9078 case 1:
9079 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9080 if (fld1 == nullptr) {
9081 return false;
9082 }
9083 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9084 break;
9085 default: fatal("Unsupported number of arguments %d", num_args);
9086 }
9087
9088 Node* result = nullptr;
9089 switch (id) {
9090 // Unary operations
9091 case vmIntrinsics::_sqrt_float16:
9092 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9093 break;
9094 // Ternary operations
9095 case vmIntrinsics::_fma_float16:
9096 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9097 break;
9098 default:
9099 fatal_unexpected_iid(id);
9100 break;
9101 }
9102 result = _gvn.transform(new ReinterpretHF2SNode(result));
9103 set_result(box_fp16_value(float16_box_type, field, result));
9104 return true;
9105 }
9106