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*)CompilerThread::current()->compiler();
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::_dcbrt:
249 case vmIntrinsics::_dabs:
250 case vmIntrinsics::_fabs:
251 case vmIntrinsics::_iabs:
252 case vmIntrinsics::_labs:
253 case vmIntrinsics::_datan2:
254 case vmIntrinsics::_dsqrt:
255 case vmIntrinsics::_dsqrt_strict:
256 case vmIntrinsics::_dexp:
257 case vmIntrinsics::_dlog:
258 case vmIntrinsics::_dlog10:
259 case vmIntrinsics::_dpow:
260 case vmIntrinsics::_dcopySign:
261 case vmIntrinsics::_fcopySign:
262 case vmIntrinsics::_dsignum:
263 case vmIntrinsics::_roundF:
264 case vmIntrinsics::_roundD:
265 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
266
267 case vmIntrinsics::_notify:
268 case vmIntrinsics::_notifyAll:
269 return inline_notify(intrinsic_id());
270
271 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
272 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
273 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
274 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
275 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
276 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
277 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
278 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
279 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
280 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
281 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
282 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
283 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
284 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
285
286 case vmIntrinsics::_arraycopy: return inline_arraycopy();
287
288 case vmIntrinsics::_arraySort: return inline_array_sort();
289 case vmIntrinsics::_arrayPartition: return inline_array_partition();
290
291 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
292 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
293 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
294 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
295
296 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
297 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
298 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
299 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
303 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
304
305 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
306
307 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
308
309 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
310 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
311 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
312 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
313
314 case vmIntrinsics::_compressStringC:
315 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
316 case vmIntrinsics::_inflateStringC:
317 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
318
319 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
320 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
321 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
322 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
323 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
324 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
325 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
326 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
327 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
328
329 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
330 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
331 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
332 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
333 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
334 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
335 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
336 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
337 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
338
339 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
340 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
341 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
342 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
343 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
344 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
345 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
346 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
347 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
348
349 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
350 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
351 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
352 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
353 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
354 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
355 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
356 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
357 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
358
359 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
360 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
361 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
362 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
363
364 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
365 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
366 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
367 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
368
369 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
370 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
371 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
372 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
373 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
374 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
375 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
376 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
377 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
378
379 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
380 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
381 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
382 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
383 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
384 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
385 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
386 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
387 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
388
389 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
390 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
391 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
392 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
393 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
394 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
395 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
396 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
397 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
398
399 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
400 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
401 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
402 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
403 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
404 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
405 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
406 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
407 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
408
409 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
414
415 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
416 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
417 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
418 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
419 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
420 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
421 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
422 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
423 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
435
436 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
437 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
438 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
439 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
451
452 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
455 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
456
457 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
462
463 case vmIntrinsics::_loadFence:
464 case vmIntrinsics::_storeFence:
465 case vmIntrinsics::_storeStoreFence:
466 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
467
468 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
469
470 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
471 case vmIntrinsics::_currentThread: return inline_native_currentThread();
472 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
473
474 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
475 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
476
477 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
478 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
479
480 #if INCLUDE_JVMTI
481 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
482 "notifyJvmtiStart", true, false);
483 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
484 "notifyJvmtiEnd", false, true);
485 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
486 "notifyJvmtiMount", false, false);
487 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
488 "notifyJvmtiUnmount", false, false);
489 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
490 #endif
491
492 #ifdef JFR_HAVE_INTRINSICS
493 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
494 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
495 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
496 #endif
497 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
498 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
499 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
500 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
501 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
502 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
503 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
504 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
505 case vmIntrinsics::_getLength: return inline_native_getLength();
506 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
507 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
508 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
509 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
510 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
511 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
512 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
513
514 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
515 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
516
517 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
518
519 case vmIntrinsics::_isInstance:
520 case vmIntrinsics::_isHidden:
521 case vmIntrinsics::_getSuperclass:
522 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
523
524 case vmIntrinsics::_floatToRawIntBits:
525 case vmIntrinsics::_floatToIntBits:
526 case vmIntrinsics::_intBitsToFloat:
527 case vmIntrinsics::_doubleToRawLongBits:
528 case vmIntrinsics::_doubleToLongBits:
529 case vmIntrinsics::_longBitsToDouble:
530 case vmIntrinsics::_floatToFloat16:
531 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
532 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
533 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
534 case vmIntrinsics::_floatIsFinite:
535 case vmIntrinsics::_floatIsInfinite:
536 case vmIntrinsics::_doubleIsFinite:
537 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
538
539 case vmIntrinsics::_numberOfLeadingZeros_i:
540 case vmIntrinsics::_numberOfLeadingZeros_l:
541 case vmIntrinsics::_numberOfTrailingZeros_i:
542 case vmIntrinsics::_numberOfTrailingZeros_l:
543 case vmIntrinsics::_bitCount_i:
544 case vmIntrinsics::_bitCount_l:
545 case vmIntrinsics::_reverse_i:
546 case vmIntrinsics::_reverse_l:
547 case vmIntrinsics::_reverseBytes_i:
548 case vmIntrinsics::_reverseBytes_l:
549 case vmIntrinsics::_reverseBytes_s:
550 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
551
552 case vmIntrinsics::_compress_i:
553 case vmIntrinsics::_compress_l:
554 case vmIntrinsics::_expand_i:
555 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
556
557 case vmIntrinsics::_compareUnsigned_i:
558 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
559
560 case vmIntrinsics::_divideUnsigned_i:
561 case vmIntrinsics::_divideUnsigned_l:
562 case vmIntrinsics::_remainderUnsigned_i:
563 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
564
565 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
566
567 case vmIntrinsics::_Reference_get: return inline_reference_get();
568 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
569 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
570 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
571 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
572
573 case vmIntrinsics::_Class_cast: return inline_Class_cast();
574
575 case vmIntrinsics::_aescrypt_encryptBlock:
576 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
577
578 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
579 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
580 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
581
582 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
583 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
584 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
585
586 case vmIntrinsics::_counterMode_AESCrypt:
587 return inline_counterMode_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_galoisCounterMode_AESCrypt:
590 return inline_galoisCounterMode_AESCrypt();
591
592 case vmIntrinsics::_md5_implCompress:
593 case vmIntrinsics::_sha_implCompress:
594 case vmIntrinsics::_sha2_implCompress:
595 case vmIntrinsics::_sha5_implCompress:
596 case vmIntrinsics::_sha3_implCompress:
597 return inline_digestBase_implCompress(intrinsic_id());
598 case vmIntrinsics::_double_keccak:
599 return inline_double_keccak();
600
601 case vmIntrinsics::_digestBase_implCompressMB:
602 return inline_digestBase_implCompressMB(predicate);
603
604 case vmIntrinsics::_multiplyToLen:
605 return inline_multiplyToLen();
606
607 case vmIntrinsics::_squareToLen:
608 return inline_squareToLen();
609
610 case vmIntrinsics::_mulAdd:
611 return inline_mulAdd();
612
613 case vmIntrinsics::_montgomeryMultiply:
614 return inline_montgomeryMultiply();
615 case vmIntrinsics::_montgomerySquare:
616 return inline_montgomerySquare();
617
618 case vmIntrinsics::_bigIntegerRightShiftWorker:
619 return inline_bigIntegerShift(true);
620 case vmIntrinsics::_bigIntegerLeftShiftWorker:
621 return inline_bigIntegerShift(false);
622
623 case vmIntrinsics::_vectorizedMismatch:
624 return inline_vectorizedMismatch();
625
626 case vmIntrinsics::_ghash_processBlocks:
627 return inline_ghash_processBlocks();
628 case vmIntrinsics::_chacha20Block:
629 return inline_chacha20Block();
630 case vmIntrinsics::_kyberNtt:
631 return inline_kyberNtt();
632 case vmIntrinsics::_kyberInverseNtt:
633 return inline_kyberInverseNtt();
634 case vmIntrinsics::_kyberNttMult:
635 return inline_kyberNttMult();
636 case vmIntrinsics::_kyberAddPoly_2:
637 return inline_kyberAddPoly_2();
638 case vmIntrinsics::_kyberAddPoly_3:
639 return inline_kyberAddPoly_3();
640 case vmIntrinsics::_kyber12To16:
641 return inline_kyber12To16();
642 case vmIntrinsics::_kyberBarrettReduce:
643 return inline_kyberBarrettReduce();
644 case vmIntrinsics::_dilithiumAlmostNtt:
645 return inline_dilithiumAlmostNtt();
646 case vmIntrinsics::_dilithiumAlmostInverseNtt:
647 return inline_dilithiumAlmostInverseNtt();
648 case vmIntrinsics::_dilithiumNttMult:
649 return inline_dilithiumNttMult();
650 case vmIntrinsics::_dilithiumMontMulByConstant:
651 return inline_dilithiumMontMulByConstant();
652 case vmIntrinsics::_dilithiumDecomposePoly:
653 return inline_dilithiumDecomposePoly();
654 case vmIntrinsics::_base64_encodeBlock:
655 return inline_base64_encodeBlock();
656 case vmIntrinsics::_base64_decodeBlock:
657 return inline_base64_decodeBlock();
658 case vmIntrinsics::_poly1305_processBlocks:
659 return inline_poly1305_processBlocks();
660 case vmIntrinsics::_intpoly_montgomeryMult_P256:
661 return inline_intpoly_montgomeryMult_P256();
662 case vmIntrinsics::_intpoly_assign:
663 return inline_intpoly_assign();
664 case vmIntrinsics::_encodeISOArray:
665 case vmIntrinsics::_encodeByteISOArray:
666 return inline_encodeISOArray(false);
667 case vmIntrinsics::_encodeAsciiArray:
668 return inline_encodeISOArray(true);
669
670 case vmIntrinsics::_updateCRC32:
671 return inline_updateCRC32();
672 case vmIntrinsics::_updateBytesCRC32:
673 return inline_updateBytesCRC32();
674 case vmIntrinsics::_updateByteBufferCRC32:
675 return inline_updateByteBufferCRC32();
676
677 case vmIntrinsics::_updateBytesCRC32C:
678 return inline_updateBytesCRC32C();
679 case vmIntrinsics::_updateDirectByteBufferCRC32C:
680 return inline_updateDirectByteBufferCRC32C();
681
682 case vmIntrinsics::_updateBytesAdler32:
683 return inline_updateBytesAdler32();
684 case vmIntrinsics::_updateByteBufferAdler32:
685 return inline_updateByteBufferAdler32();
686
687 case vmIntrinsics::_profileBoolean:
688 return inline_profileBoolean();
689 case vmIntrinsics::_isCompileConstant:
690 return inline_isCompileConstant();
691
692 case vmIntrinsics::_countPositives:
693 return inline_countPositives();
694
695 case vmIntrinsics::_fmaD:
696 case vmIntrinsics::_fmaF:
697 return inline_fma(intrinsic_id());
698
699 case vmIntrinsics::_isDigit:
700 case vmIntrinsics::_isLowerCase:
701 case vmIntrinsics::_isUpperCase:
702 case vmIntrinsics::_isWhitespace:
703 return inline_character_compare(intrinsic_id());
704
705 case vmIntrinsics::_min:
706 case vmIntrinsics::_max:
707 case vmIntrinsics::_min_strict:
708 case vmIntrinsics::_max_strict:
709 case vmIntrinsics::_minL:
710 case vmIntrinsics::_maxL:
711 case vmIntrinsics::_minF:
712 case vmIntrinsics::_maxF:
713 case vmIntrinsics::_minD:
714 case vmIntrinsics::_maxD:
715 case vmIntrinsics::_minF_strict:
716 case vmIntrinsics::_maxF_strict:
717 case vmIntrinsics::_minD_strict:
718 case vmIntrinsics::_maxD_strict:
719 return inline_min_max(intrinsic_id());
720
721 case vmIntrinsics::_VectorUnaryOp:
722 return inline_vector_nary_operation(1);
723 case vmIntrinsics::_VectorBinaryOp:
724 return inline_vector_nary_operation(2);
725 case vmIntrinsics::_VectorUnaryLibOp:
726 return inline_vector_call(1);
727 case vmIntrinsics::_VectorBinaryLibOp:
728 return inline_vector_call(2);
729 case vmIntrinsics::_VectorTernaryOp:
730 return inline_vector_nary_operation(3);
731 case vmIntrinsics::_VectorFromBitsCoerced:
732 return inline_vector_frombits_coerced();
733 case vmIntrinsics::_VectorMaskOp:
734 return inline_vector_mask_operation();
735 case vmIntrinsics::_VectorLoadOp:
736 return inline_vector_mem_operation(/*is_store=*/false);
737 case vmIntrinsics::_VectorLoadMaskedOp:
738 return inline_vector_mem_masked_operation(/*is_store*/false);
739 case vmIntrinsics::_VectorStoreOp:
740 return inline_vector_mem_operation(/*is_store=*/true);
741 case vmIntrinsics::_VectorStoreMaskedOp:
742 return inline_vector_mem_masked_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorGatherOp:
744 return inline_vector_gather_scatter(/*is_scatter*/ false);
745 case vmIntrinsics::_VectorScatterOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ true);
747 case vmIntrinsics::_VectorReductionCoerced:
748 return inline_vector_reduction();
749 case vmIntrinsics::_VectorTest:
750 return inline_vector_test();
751 case vmIntrinsics::_VectorBlend:
752 return inline_vector_blend();
753 case vmIntrinsics::_VectorRearrange:
754 return inline_vector_rearrange();
755 case vmIntrinsics::_VectorSelectFrom:
756 return inline_vector_select_from();
757 case vmIntrinsics::_VectorCompare:
758 return inline_vector_compare();
759 case vmIntrinsics::_VectorBroadcastInt:
760 return inline_vector_broadcast_int();
761 case vmIntrinsics::_VectorConvert:
762 return inline_vector_convert();
763 case vmIntrinsics::_VectorInsert:
764 return inline_vector_insert();
765 case vmIntrinsics::_VectorExtract:
766 return inline_vector_extract();
767 case vmIntrinsics::_VectorCompressExpand:
768 return inline_vector_compress_expand();
769 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
770 return inline_vector_select_from_two_vectors();
771 case vmIntrinsics::_IndexVector:
772 return inline_index_vector();
773 case vmIntrinsics::_IndexPartiallyInUpperRange:
774 return inline_index_partially_in_upper_range();
775
776 case vmIntrinsics::_getObjectSize:
777 return inline_getObjectSize();
778
779 case vmIntrinsics::_blackhole:
780 return inline_blackhole();
781
782 default:
783 // If you get here, it may be that someone has added a new intrinsic
784 // to the list in vmIntrinsics.hpp without implementing it here.
785 #ifndef PRODUCT
786 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
787 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
788 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
789 }
790 #endif
791 return false;
792 }
793 }
794
795 Node* LibraryCallKit::try_to_predicate(int predicate) {
796 if (!jvms()->has_method()) {
797 // Root JVMState has a null method.
798 assert(map()->memory()->Opcode() == Op_Parm, "");
799 // Insert the memory aliasing node
800 set_all_memory(reset_memory());
801 }
802 assert(merged_memory(), "");
803
804 switch (intrinsic_id()) {
805 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
806 return inline_cipherBlockChaining_AESCrypt_predicate(false);
807 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(true);
809 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
810 return inline_electronicCodeBook_AESCrypt_predicate(false);
811 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(true);
813 case vmIntrinsics::_counterMode_AESCrypt:
814 return inline_counterMode_AESCrypt_predicate();
815 case vmIntrinsics::_digestBase_implCompressMB:
816 return inline_digestBase_implCompressMB_predicate(predicate);
817 case vmIntrinsics::_galoisCounterMode_AESCrypt:
818 return inline_galoisCounterMode_AESCrypt_predicate();
819
820 default:
821 // If you get here, it may be that someone has added a new intrinsic
822 // to the list in vmIntrinsics.hpp without implementing it here.
823 #ifndef PRODUCT
824 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
825 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
826 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
827 }
828 #endif
829 Node* slow_ctl = control();
830 set_control(top()); // No fast path intrinsic
831 return slow_ctl;
832 }
833 }
834
835 //------------------------------set_result-------------------------------
836 // Helper function for finishing intrinsics.
837 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
838 record_for_igvn(region);
839 set_control(_gvn.transform(region));
840 set_result( _gvn.transform(value));
841 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
842 }
843
844 //------------------------------generate_guard---------------------------
845 // Helper function for generating guarded fast-slow graph structures.
846 // The given 'test', if true, guards a slow path. If the test fails
847 // then a fast path can be taken. (We generally hope it fails.)
848 // In all cases, GraphKit::control() is updated to the fast path.
849 // The returned value represents the control for the slow path.
850 // The return value is never 'top'; it is either a valid control
851 // or null if it is obvious that the slow path can never be taken.
852 // Also, if region and the slow control are not null, the slow edge
853 // is appended to the region.
854 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
855 if (stopped()) {
856 // Already short circuited.
857 return nullptr;
858 }
859
860 // Build an if node and its projections.
861 // If test is true we take the slow path, which we assume is uncommon.
862 if (_gvn.type(test) == TypeInt::ZERO) {
863 // The slow branch is never taken. No need to build this guard.
864 return nullptr;
865 }
866
867 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
868
869 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
870 if (if_slow == top()) {
871 // The slow branch is never taken. No need to build this guard.
872 return nullptr;
873 }
874
875 if (region != nullptr)
876 region->add_req(if_slow);
877
878 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
879 set_control(if_fast);
880
881 return if_slow;
882 }
883
884 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
885 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
886 }
887 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
888 return generate_guard(test, region, PROB_FAIR);
889 }
890
891 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
892 Node* *pos_index) {
893 if (stopped())
894 return nullptr; // already stopped
895 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
896 return nullptr; // index is already adequately typed
897 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
898 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
899 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
900 if (is_neg != nullptr && pos_index != nullptr) {
901 // Emulate effect of Parse::adjust_map_after_if.
902 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
903 (*pos_index) = _gvn.transform(ccast);
904 }
905 return is_neg;
906 }
907
908 // Make sure that 'position' is a valid limit index, in [0..length].
909 // There are two equivalent plans for checking this:
910 // A. (offset + copyLength) unsigned<= arrayLength
911 // B. offset <= (arrayLength - copyLength)
912 // We require that all of the values above, except for the sum and
913 // difference, are already known to be non-negative.
914 // Plan A is robust in the face of overflow, if offset and copyLength
915 // are both hugely positive.
916 //
917 // Plan B is less direct and intuitive, but it does not overflow at
918 // all, since the difference of two non-negatives is always
919 // representable. Whenever Java methods must perform the equivalent
920 // check they generally use Plan B instead of Plan A.
921 // For the moment we use Plan A.
922 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
923 Node* subseq_length,
924 Node* array_length,
925 RegionNode* region) {
926 if (stopped())
927 return nullptr; // already stopped
928 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
929 if (zero_offset && subseq_length->eqv_uncast(array_length))
930 return nullptr; // common case of whole-array copy
931 Node* last = subseq_length;
932 if (!zero_offset) // last += offset
933 last = _gvn.transform(new AddINode(last, offset));
934 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
935 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
936 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
937 return is_over;
938 }
939
940 // Emit range checks for the given String.value byte array
941 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
942 if (stopped()) {
943 return; // already stopped
944 }
945 RegionNode* bailout = new RegionNode(1);
946 record_for_igvn(bailout);
947 if (char_count) {
948 // Convert char count to byte count
949 count = _gvn.transform(new LShiftINode(count, intcon(1)));
950 }
951
952 // Offset and count must not be negative
953 generate_negative_guard(offset, bailout);
954 generate_negative_guard(count, bailout);
955 // Offset + count must not exceed length of array
956 generate_limit_guard(offset, count, load_array_length(array), bailout);
957
958 if (bailout->req() > 1) {
959 PreserveJVMState pjvms(this);
960 set_control(_gvn.transform(bailout));
961 uncommon_trap(Deoptimization::Reason_intrinsic,
962 Deoptimization::Action_maybe_recompile);
963 }
964 }
965
966 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
967 bool is_immutable) {
968 ciKlass* thread_klass = env()->Thread_klass();
969 const Type* thread_type
970 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
971
972 Node* thread = _gvn.transform(new ThreadLocalNode());
973 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
974 tls_output = thread;
975
976 Node* thread_obj_handle
977 = (is_immutable
978 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
979 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
980 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
981 thread_obj_handle = _gvn.transform(thread_obj_handle);
982
983 DecoratorSet decorators = IN_NATIVE;
984 if (is_immutable) {
985 decorators |= C2_IMMUTABLE_MEMORY;
986 }
987 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
988 }
989
990 //--------------------------generate_current_thread--------------------
991 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
992 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
993 /*is_immutable*/false);
994 }
995
996 //--------------------------generate_virtual_thread--------------------
997 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
998 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
999 !C->method()->changes_current_thread());
1000 }
1001
1002 //------------------------------make_string_method_node------------------------
1003 // Helper method for String intrinsic functions. This version is called with
1004 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1005 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1006 // containing the lengths of str1 and str2.
1007 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1008 Node* result = nullptr;
1009 switch (opcode) {
1010 case Op_StrIndexOf:
1011 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1012 str1_start, cnt1, str2_start, cnt2, ae);
1013 break;
1014 case Op_StrComp:
1015 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1016 str1_start, cnt1, str2_start, cnt2, ae);
1017 break;
1018 case Op_StrEquals:
1019 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1020 // Use the constant length if there is one because optimized match rule may exist.
1021 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1022 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1023 break;
1024 default:
1025 ShouldNotReachHere();
1026 return nullptr;
1027 }
1028
1029 // All these intrinsics have checks.
1030 C->set_has_split_ifs(true); // Has chance for split-if optimization
1031 clear_upper_avx();
1032
1033 return _gvn.transform(result);
1034 }
1035
1036 //------------------------------inline_string_compareTo------------------------
1037 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1038 Node* arg1 = argument(0);
1039 Node* arg2 = argument(1);
1040
1041 arg1 = must_be_not_null(arg1, true);
1042 arg2 = must_be_not_null(arg2, true);
1043
1044 // Get start addr and length of first argument
1045 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1046 Node* arg1_cnt = load_array_length(arg1);
1047
1048 // Get start addr and length of second argument
1049 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1050 Node* arg2_cnt = load_array_length(arg2);
1051
1052 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1053 set_result(result);
1054 return true;
1055 }
1056
1057 //------------------------------inline_string_equals------------------------
1058 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1059 Node* arg1 = argument(0);
1060 Node* arg2 = argument(1);
1061
1062 // paths (plus control) merge
1063 RegionNode* region = new RegionNode(3);
1064 Node* phi = new PhiNode(region, TypeInt::BOOL);
1065
1066 if (!stopped()) {
1067
1068 arg1 = must_be_not_null(arg1, true);
1069 arg2 = must_be_not_null(arg2, true);
1070
1071 // Get start addr and length of first argument
1072 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1073 Node* arg1_cnt = load_array_length(arg1);
1074
1075 // Get start addr and length of second argument
1076 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1077 Node* arg2_cnt = load_array_length(arg2);
1078
1079 // Check for arg1_cnt != arg2_cnt
1080 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1081 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1082 Node* if_ne = generate_slow_guard(bol, nullptr);
1083 if (if_ne != nullptr) {
1084 phi->init_req(2, intcon(0));
1085 region->init_req(2, if_ne);
1086 }
1087
1088 // Check for count == 0 is done by assembler code for StrEquals.
1089
1090 if (!stopped()) {
1091 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1092 phi->init_req(1, equals);
1093 region->init_req(1, control());
1094 }
1095 }
1096
1097 // post merge
1098 set_control(_gvn.transform(region));
1099 record_for_igvn(region);
1100
1101 set_result(_gvn.transform(phi));
1102 return true;
1103 }
1104
1105 //------------------------------inline_array_equals----------------------------
1106 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1107 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1108 Node* arg1 = argument(0);
1109 Node* arg2 = argument(1);
1110
1111 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1112 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1113 clear_upper_avx();
1114
1115 return true;
1116 }
1117
1118
1119 //------------------------------inline_countPositives------------------------------
1120 bool LibraryCallKit::inline_countPositives() {
1121 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1122 return false;
1123 }
1124
1125 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1126 // no receiver since it is static method
1127 Node* ba = argument(0);
1128 Node* offset = argument(1);
1129 Node* len = argument(2);
1130
1131 ba = must_be_not_null(ba, true);
1132
1133 // Range checks
1134 generate_string_range_check(ba, offset, len, false);
1135 if (stopped()) {
1136 return true;
1137 }
1138 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1139 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1140 set_result(_gvn.transform(result));
1141 clear_upper_avx();
1142 return true;
1143 }
1144
1145 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1146 Node* index = argument(0);
1147 Node* length = bt == T_INT ? argument(1) : argument(2);
1148 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1149 return false;
1150 }
1151
1152 // check that length is positive
1153 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1154 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1155
1156 {
1157 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1158 uncommon_trap(Deoptimization::Reason_intrinsic,
1159 Deoptimization::Action_make_not_entrant);
1160 }
1161
1162 if (stopped()) {
1163 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1164 return true;
1165 }
1166
1167 // length is now known positive, add a cast node to make this explicit
1168 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1169 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1170 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1171 ConstraintCastNode::RegularDependency, bt);
1172 casted_length = _gvn.transform(casted_length);
1173 replace_in_map(length, casted_length);
1174 length = casted_length;
1175
1176 // Use an unsigned comparison for the range check itself
1177 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1178 BoolTest::mask btest = BoolTest::lt;
1179 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1180 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1181 _gvn.set_type(rc, rc->Value(&_gvn));
1182 if (!rc_bool->is_Con()) {
1183 record_for_igvn(rc);
1184 }
1185 set_control(_gvn.transform(new IfTrueNode(rc)));
1186 {
1187 PreserveJVMState pjvms(this);
1188 set_control(_gvn.transform(new IfFalseNode(rc)));
1189 uncommon_trap(Deoptimization::Reason_range_check,
1190 Deoptimization::Action_make_not_entrant);
1191 }
1192
1193 if (stopped()) {
1194 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1195 return true;
1196 }
1197
1198 // index is now known to be >= 0 and < length, cast it
1199 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1200 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1201 ConstraintCastNode::RegularDependency, bt);
1202 result = _gvn.transform(result);
1203 set_result(result);
1204 replace_in_map(index, result);
1205 return true;
1206 }
1207
1208 //------------------------------inline_string_indexOf------------------------
1209 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1210 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1211 return false;
1212 }
1213 Node* src = argument(0);
1214 Node* tgt = argument(1);
1215
1216 // Make the merge point
1217 RegionNode* result_rgn = new RegionNode(4);
1218 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1219
1220 src = must_be_not_null(src, true);
1221 tgt = must_be_not_null(tgt, true);
1222
1223 // Get start addr and length of source string
1224 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1225 Node* src_count = load_array_length(src);
1226
1227 // Get start addr and length of substring
1228 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1229 Node* tgt_count = load_array_length(tgt);
1230
1231 Node* result = nullptr;
1232 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1233
1234 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1235 // Divide src size by 2 if String is UTF16 encoded
1236 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1237 }
1238 if (ae == StrIntrinsicNode::UU) {
1239 // Divide substring size by 2 if String is UTF16 encoded
1240 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1241 }
1242
1243 if (call_opt_stub) {
1244 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1245 StubRoutines::_string_indexof_array[ae],
1246 "stringIndexOf", TypePtr::BOTTOM, src_start,
1247 src_count, tgt_start, tgt_count);
1248 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1249 } else {
1250 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1251 result_rgn, result_phi, ae);
1252 }
1253 if (result != nullptr) {
1254 result_phi->init_req(3, result);
1255 result_rgn->init_req(3, control());
1256 }
1257 set_control(_gvn.transform(result_rgn));
1258 record_for_igvn(result_rgn);
1259 set_result(_gvn.transform(result_phi));
1260
1261 return true;
1262 }
1263
1264 //-----------------------------inline_string_indexOfI-----------------------
1265 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1266 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1267 return false;
1268 }
1269 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1270 return false;
1271 }
1272
1273 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1274 Node* src = argument(0); // byte[]
1275 Node* src_count = argument(1); // char count
1276 Node* tgt = argument(2); // byte[]
1277 Node* tgt_count = argument(3); // char count
1278 Node* from_index = argument(4); // char index
1279
1280 src = must_be_not_null(src, true);
1281 tgt = must_be_not_null(tgt, true);
1282
1283 // Multiply byte array index by 2 if String is UTF16 encoded
1284 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1285 src_count = _gvn.transform(new SubINode(src_count, from_index));
1286 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1287 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1288
1289 // Range checks
1290 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1291 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1292 if (stopped()) {
1293 return true;
1294 }
1295
1296 RegionNode* region = new RegionNode(5);
1297 Node* phi = new PhiNode(region, TypeInt::INT);
1298 Node* result = nullptr;
1299
1300 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1301
1302 if (call_opt_stub) {
1303 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1304 StubRoutines::_string_indexof_array[ae],
1305 "stringIndexOf", TypePtr::BOTTOM, src_start,
1306 src_count, tgt_start, tgt_count);
1307 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1308 } else {
1309 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1310 region, phi, ae);
1311 }
1312 if (result != nullptr) {
1313 // The result is index relative to from_index if substring was found, -1 otherwise.
1314 // Generate code which will fold into cmove.
1315 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1316 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1317
1318 Node* if_lt = generate_slow_guard(bol, nullptr);
1319 if (if_lt != nullptr) {
1320 // result == -1
1321 phi->init_req(3, result);
1322 region->init_req(3, if_lt);
1323 }
1324 if (!stopped()) {
1325 result = _gvn.transform(new AddINode(result, from_index));
1326 phi->init_req(4, result);
1327 region->init_req(4, control());
1328 }
1329 }
1330
1331 set_control(_gvn.transform(region));
1332 record_for_igvn(region);
1333 set_result(_gvn.transform(phi));
1334 clear_upper_avx();
1335
1336 return true;
1337 }
1338
1339 // Create StrIndexOfNode with fast path checks
1340 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1341 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1342 // Check for substr count > string count
1343 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1344 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1345 Node* if_gt = generate_slow_guard(bol, nullptr);
1346 if (if_gt != nullptr) {
1347 phi->init_req(1, intcon(-1));
1348 region->init_req(1, if_gt);
1349 }
1350 if (!stopped()) {
1351 // Check for substr count == 0
1352 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1353 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1354 Node* if_zero = generate_slow_guard(bol, nullptr);
1355 if (if_zero != nullptr) {
1356 phi->init_req(2, intcon(0));
1357 region->init_req(2, if_zero);
1358 }
1359 }
1360 if (!stopped()) {
1361 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1362 }
1363 return nullptr;
1364 }
1365
1366 //-----------------------------inline_string_indexOfChar-----------------------
1367 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1368 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1369 return false;
1370 }
1371 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1372 return false;
1373 }
1374 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1375 Node* src = argument(0); // byte[]
1376 Node* int_ch = argument(1);
1377 Node* from_index = argument(2);
1378 Node* max = argument(3);
1379
1380 src = must_be_not_null(src, true);
1381
1382 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1383 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1384 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1385
1386 // Range checks
1387 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1388
1389 // Check for int_ch >= 0
1390 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1391 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1392 {
1393 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1394 uncommon_trap(Deoptimization::Reason_intrinsic,
1395 Deoptimization::Action_maybe_recompile);
1396 }
1397 if (stopped()) {
1398 return true;
1399 }
1400
1401 RegionNode* region = new RegionNode(3);
1402 Node* phi = new PhiNode(region, TypeInt::INT);
1403
1404 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1405 C->set_has_split_ifs(true); // Has chance for split-if optimization
1406 _gvn.transform(result);
1407
1408 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1409 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1410
1411 Node* if_lt = generate_slow_guard(bol, nullptr);
1412 if (if_lt != nullptr) {
1413 // result == -1
1414 phi->init_req(2, result);
1415 region->init_req(2, if_lt);
1416 }
1417 if (!stopped()) {
1418 result = _gvn.transform(new AddINode(result, from_index));
1419 phi->init_req(1, result);
1420 region->init_req(1, control());
1421 }
1422 set_control(_gvn.transform(region));
1423 record_for_igvn(region);
1424 set_result(_gvn.transform(phi));
1425 clear_upper_avx();
1426
1427 return true;
1428 }
1429 //---------------------------inline_string_copy---------------------
1430 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1431 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1432 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1433 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1434 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1435 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1436 bool LibraryCallKit::inline_string_copy(bool compress) {
1437 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1438 return false;
1439 }
1440 int nargs = 5; // 2 oops, 3 ints
1441 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1442
1443 Node* src = argument(0);
1444 Node* src_offset = argument(1);
1445 Node* dst = argument(2);
1446 Node* dst_offset = argument(3);
1447 Node* length = argument(4);
1448
1449 // Check for allocation before we add nodes that would confuse
1450 // tightly_coupled_allocation()
1451 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1452
1453 // Figure out the size and type of the elements we will be copying.
1454 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1455 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1456 if (src_type == nullptr || dst_type == nullptr) {
1457 return false;
1458 }
1459 BasicType src_elem = src_type->elem()->array_element_basic_type();
1460 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1461 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1462 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1463 "Unsupported array types for inline_string_copy");
1464
1465 src = must_be_not_null(src, true);
1466 dst = must_be_not_null(dst, true);
1467
1468 // Convert char[] offsets to byte[] offsets
1469 bool convert_src = (compress && src_elem == T_BYTE);
1470 bool convert_dst = (!compress && dst_elem == T_BYTE);
1471 if (convert_src) {
1472 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1473 } else if (convert_dst) {
1474 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1475 }
1476
1477 // Range checks
1478 generate_string_range_check(src, src_offset, length, convert_src);
1479 generate_string_range_check(dst, dst_offset, length, convert_dst);
1480 if (stopped()) {
1481 return true;
1482 }
1483
1484 Node* src_start = array_element_address(src, src_offset, src_elem);
1485 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1486 // 'src_start' points to src array + scaled offset
1487 // 'dst_start' points to dst array + scaled offset
1488 Node* count = nullptr;
1489 if (compress) {
1490 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1491 } else {
1492 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1493 }
1494
1495 if (alloc != nullptr) {
1496 if (alloc->maybe_set_complete(&_gvn)) {
1497 // "You break it, you buy it."
1498 InitializeNode* init = alloc->initialization();
1499 assert(init->is_complete(), "we just did this");
1500 init->set_complete_with_arraycopy();
1501 assert(dst->is_CheckCastPP(), "sanity");
1502 assert(dst->in(0)->in(0) == init, "dest pinned");
1503 }
1504 // Do not let stores that initialize this object be reordered with
1505 // a subsequent store that would make this object accessible by
1506 // other threads.
1507 // Record what AllocateNode this StoreStore protects so that
1508 // escape analysis can go from the MemBarStoreStoreNode to the
1509 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1510 // based on the escape status of the AllocateNode.
1511 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1512 }
1513 if (compress) {
1514 set_result(_gvn.transform(count));
1515 }
1516 clear_upper_avx();
1517
1518 return true;
1519 }
1520
1521 #ifdef _LP64
1522 #define XTOP ,top() /*additional argument*/
1523 #else //_LP64
1524 #define XTOP /*no additional argument*/
1525 #endif //_LP64
1526
1527 //------------------------inline_string_toBytesU--------------------------
1528 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1529 bool LibraryCallKit::inline_string_toBytesU() {
1530 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1531 return false;
1532 }
1533 // Get the arguments.
1534 Node* value = argument(0);
1535 Node* offset = argument(1);
1536 Node* length = argument(2);
1537
1538 Node* newcopy = nullptr;
1539
1540 // Set the original stack and the reexecute bit for the interpreter to reexecute
1541 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1542 { PreserveReexecuteState preexecs(this);
1543 jvms()->set_should_reexecute(true);
1544
1545 // Check if a null path was taken unconditionally.
1546 value = null_check(value);
1547
1548 RegionNode* bailout = new RegionNode(1);
1549 record_for_igvn(bailout);
1550
1551 // Range checks
1552 generate_negative_guard(offset, bailout);
1553 generate_negative_guard(length, bailout);
1554 generate_limit_guard(offset, length, load_array_length(value), bailout);
1555 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1556 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1557
1558 if (bailout->req() > 1) {
1559 PreserveJVMState pjvms(this);
1560 set_control(_gvn.transform(bailout));
1561 uncommon_trap(Deoptimization::Reason_intrinsic,
1562 Deoptimization::Action_maybe_recompile);
1563 }
1564 if (stopped()) {
1565 return true;
1566 }
1567
1568 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1569 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1570 newcopy = new_array(klass_node, size, 0); // no arguments to push
1571 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1572 guarantee(alloc != nullptr, "created above");
1573
1574 // Calculate starting addresses.
1575 Node* src_start = array_element_address(value, offset, T_CHAR);
1576 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1577
1578 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1579 const TypeInt* toffset = gvn().type(offset)->is_int();
1580 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1581
1582 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1583 const char* copyfunc_name = "arraycopy";
1584 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1585 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1586 OptoRuntime::fast_arraycopy_Type(),
1587 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1588 src_start, dst_start, ConvI2X(length) XTOP);
1589 // Do not let reads from the cloned object float above the arraycopy.
1590 if (alloc->maybe_set_complete(&_gvn)) {
1591 // "You break it, you buy it."
1592 InitializeNode* init = alloc->initialization();
1593 assert(init->is_complete(), "we just did this");
1594 init->set_complete_with_arraycopy();
1595 assert(newcopy->is_CheckCastPP(), "sanity");
1596 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1597 }
1598 // Do not let stores that initialize this object be reordered with
1599 // a subsequent store that would make this object accessible by
1600 // other threads.
1601 // Record what AllocateNode this StoreStore protects so that
1602 // escape analysis can go from the MemBarStoreStoreNode to the
1603 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1604 // based on the escape status of the AllocateNode.
1605 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1606 } // original reexecute is set back here
1607
1608 C->set_has_split_ifs(true); // Has chance for split-if optimization
1609 if (!stopped()) {
1610 set_result(newcopy);
1611 }
1612 clear_upper_avx();
1613
1614 return true;
1615 }
1616
1617 //------------------------inline_string_getCharsU--------------------------
1618 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1619 bool LibraryCallKit::inline_string_getCharsU() {
1620 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1621 return false;
1622 }
1623
1624 // Get the arguments.
1625 Node* src = argument(0);
1626 Node* src_begin = argument(1);
1627 Node* src_end = argument(2); // exclusive offset (i < src_end)
1628 Node* dst = argument(3);
1629 Node* dst_begin = argument(4);
1630
1631 // Check for allocation before we add nodes that would confuse
1632 // tightly_coupled_allocation()
1633 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1634
1635 // Check if a null path was taken unconditionally.
1636 src = null_check(src);
1637 dst = null_check(dst);
1638 if (stopped()) {
1639 return true;
1640 }
1641
1642 // Get length and convert char[] offset to byte[] offset
1643 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1644 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1645
1646 // Range checks
1647 generate_string_range_check(src, src_begin, length, true);
1648 generate_string_range_check(dst, dst_begin, length, false);
1649 if (stopped()) {
1650 return true;
1651 }
1652
1653 if (!stopped()) {
1654 // Calculate starting addresses.
1655 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1656 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1657
1658 // Check if array addresses are aligned to HeapWordSize
1659 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1660 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1661 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1662 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1663
1664 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1665 const char* copyfunc_name = "arraycopy";
1666 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1667 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1668 OptoRuntime::fast_arraycopy_Type(),
1669 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1670 src_start, dst_start, ConvI2X(length) XTOP);
1671 // Do not let reads from the cloned object float above the arraycopy.
1672 if (alloc != nullptr) {
1673 if (alloc->maybe_set_complete(&_gvn)) {
1674 // "You break it, you buy it."
1675 InitializeNode* init = alloc->initialization();
1676 assert(init->is_complete(), "we just did this");
1677 init->set_complete_with_arraycopy();
1678 assert(dst->is_CheckCastPP(), "sanity");
1679 assert(dst->in(0)->in(0) == init, "dest pinned");
1680 }
1681 // Do not let stores that initialize this object be reordered with
1682 // a subsequent store that would make this object accessible by
1683 // other threads.
1684 // Record what AllocateNode this StoreStore protects so that
1685 // escape analysis can go from the MemBarStoreStoreNode to the
1686 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1687 // based on the escape status of the AllocateNode.
1688 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1689 } else {
1690 insert_mem_bar(Op_MemBarCPUOrder);
1691 }
1692 }
1693
1694 C->set_has_split_ifs(true); // Has chance for split-if optimization
1695 return true;
1696 }
1697
1698 //----------------------inline_string_char_access----------------------------
1699 // Store/Load char to/from byte[] array.
1700 // static void StringUTF16.putChar(byte[] val, int index, int c)
1701 // static char StringUTF16.getChar(byte[] val, int index)
1702 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1703 Node* value = argument(0);
1704 Node* index = argument(1);
1705 Node* ch = is_store ? argument(2) : nullptr;
1706
1707 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1708 // correctly requires matched array shapes.
1709 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1710 "sanity: byte[] and char[] bases agree");
1711 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1712 "sanity: byte[] and char[] scales agree");
1713
1714 // Bail when getChar over constants is requested: constant folding would
1715 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1716 // Java method would constant fold nicely instead.
1717 if (!is_store && value->is_Con() && index->is_Con()) {
1718 return false;
1719 }
1720
1721 // Save state and restore on bailout
1722 uint old_sp = sp();
1723 SafePointNode* old_map = clone_map();
1724
1725 value = must_be_not_null(value, true);
1726
1727 Node* adr = array_element_address(value, index, T_CHAR);
1728 if (adr->is_top()) {
1729 set_map(old_map);
1730 set_sp(old_sp);
1731 return false;
1732 }
1733 destruct_map_clone(old_map);
1734 if (is_store) {
1735 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1736 } else {
1737 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);
1738 set_result(ch);
1739 }
1740 return true;
1741 }
1742
1743
1744 //------------------------------inline_math-----------------------------------
1745 // public static double Math.abs(double)
1746 // public static double Math.sqrt(double)
1747 // public static double Math.log(double)
1748 // public static double Math.log10(double)
1749 // public static double Math.round(double)
1750 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1751 Node* arg = argument(0);
1752 Node* n = nullptr;
1753 switch (id) {
1754 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1755 case vmIntrinsics::_dsqrt:
1756 case vmIntrinsics::_dsqrt_strict:
1757 n = new SqrtDNode(C, control(), arg); break;
1758 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1759 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1760 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1761 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1762 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1763 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1764 default: fatal_unexpected_iid(id); break;
1765 }
1766 set_result(_gvn.transform(n));
1767 return true;
1768 }
1769
1770 //------------------------------inline_math-----------------------------------
1771 // public static float Math.abs(float)
1772 // public static int Math.abs(int)
1773 // public static long Math.abs(long)
1774 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1775 Node* arg = argument(0);
1776 Node* n = nullptr;
1777 switch (id) {
1778 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1779 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1780 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1781 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1782 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1783 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1784 default: fatal_unexpected_iid(id); break;
1785 }
1786 set_result(_gvn.transform(n));
1787 return true;
1788 }
1789
1790 //------------------------------runtime_math-----------------------------
1791 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1792 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1793 "must be (DD)D or (D)D type");
1794
1795 // Inputs
1796 Node* a = argument(0);
1797 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1798
1799 const TypePtr* no_memory_effects = nullptr;
1800 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1801 no_memory_effects,
1802 a, top(), b, b ? top() : nullptr);
1803 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1804 #ifdef ASSERT
1805 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1806 assert(value_top == top(), "second value must be top");
1807 #endif
1808
1809 set_result(value);
1810 return true;
1811 }
1812
1813 //------------------------------inline_math_pow-----------------------------
1814 bool LibraryCallKit::inline_math_pow() {
1815 Node* exp = argument(2);
1816 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1817 if (d != nullptr) {
1818 if (d->getd() == 2.0) {
1819 // Special case: pow(x, 2.0) => x * x
1820 Node* base = argument(0);
1821 set_result(_gvn.transform(new MulDNode(base, base)));
1822 return true;
1823 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1824 // Special case: pow(x, 0.5) => sqrt(x)
1825 Node* base = argument(0);
1826 Node* zero = _gvn.zerocon(T_DOUBLE);
1827
1828 RegionNode* region = new RegionNode(3);
1829 Node* phi = new PhiNode(region, Type::DOUBLE);
1830
1831 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1832 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1833 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1834 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1835 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1836
1837 Node* if_pow = generate_slow_guard(test, nullptr);
1838 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1839 phi->init_req(1, value_sqrt);
1840 region->init_req(1, control());
1841
1842 if (if_pow != nullptr) {
1843 set_control(if_pow);
1844 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1845 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1846 const TypePtr* no_memory_effects = nullptr;
1847 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1848 no_memory_effects, base, top(), exp, top());
1849 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1850 #ifdef ASSERT
1851 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1852 assert(value_top == top(), "second value must be top");
1853 #endif
1854 phi->init_req(2, value_pow);
1855 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1856 }
1857
1858 C->set_has_split_ifs(true); // Has chance for split-if optimization
1859 set_control(_gvn.transform(region));
1860 record_for_igvn(region);
1861 set_result(_gvn.transform(phi));
1862
1863 return true;
1864 }
1865 }
1866
1867 return StubRoutines::dpow() != nullptr ?
1868 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1869 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1870 }
1871
1872 //------------------------------inline_math_native-----------------------------
1873 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1874 switch (id) {
1875 case vmIntrinsics::_dsin:
1876 return StubRoutines::dsin() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1879 case vmIntrinsics::_dcos:
1880 return StubRoutines::dcos() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1883 case vmIntrinsics::_dtan:
1884 return StubRoutines::dtan() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1887 case vmIntrinsics::_dtanh:
1888 return StubRoutines::dtanh() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1890 case vmIntrinsics::_dcbrt:
1891 return StubRoutines::dcbrt() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1893 case vmIntrinsics::_dexp:
1894 return StubRoutines::dexp() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1897 case vmIntrinsics::_dlog:
1898 return StubRoutines::dlog() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1900 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1901 case vmIntrinsics::_dlog10:
1902 return StubRoutines::dlog10() != nullptr ?
1903 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1904 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1905
1906 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1907 case vmIntrinsics::_ceil:
1908 case vmIntrinsics::_floor:
1909 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1910
1911 case vmIntrinsics::_dsqrt:
1912 case vmIntrinsics::_dsqrt_strict:
1913 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1914 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1916 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1917 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1918
1919 case vmIntrinsics::_dpow: return inline_math_pow();
1920 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1921 case vmIntrinsics::_fcopySign: return inline_math(id);
1922 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1923 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1924 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1925
1926 // These intrinsics are not yet correctly implemented
1927 case vmIntrinsics::_datan2:
1928 return false;
1929
1930 default:
1931 fatal_unexpected_iid(id);
1932 return false;
1933 }
1934 }
1935
1936 //----------------------------inline_notify-----------------------------------*
1937 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1938 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1939 address func;
1940 if (id == vmIntrinsics::_notify) {
1941 func = OptoRuntime::monitor_notify_Java();
1942 } else {
1943 func = OptoRuntime::monitor_notifyAll_Java();
1944 }
1945 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1946 make_slow_call_ex(call, env()->Throwable_klass(), false);
1947 return true;
1948 }
1949
1950
1951 //----------------------------inline_min_max-----------------------------------
1952 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1953 Node* a = nullptr;
1954 Node* b = nullptr;
1955 Node* n = nullptr;
1956 switch (id) {
1957 case vmIntrinsics::_min:
1958 case vmIntrinsics::_max:
1959 case vmIntrinsics::_minF:
1960 case vmIntrinsics::_maxF:
1961 case vmIntrinsics::_minF_strict:
1962 case vmIntrinsics::_maxF_strict:
1963 case vmIntrinsics::_min_strict:
1964 case vmIntrinsics::_max_strict:
1965 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1966 a = argument(0);
1967 b = argument(1);
1968 break;
1969 case vmIntrinsics::_minD:
1970 case vmIntrinsics::_maxD:
1971 case vmIntrinsics::_minD_strict:
1972 case vmIntrinsics::_maxD_strict:
1973 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1974 a = argument(0);
1975 b = argument(2);
1976 break;
1977 case vmIntrinsics::_minL:
1978 case vmIntrinsics::_maxL:
1979 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1980 a = argument(0);
1981 b = argument(2);
1982 break;
1983 default:
1984 fatal_unexpected_iid(id);
1985 break;
1986 }
1987
1988 switch (id) {
1989 case vmIntrinsics::_min:
1990 case vmIntrinsics::_min_strict:
1991 n = new MinINode(a, b);
1992 break;
1993 case vmIntrinsics::_max:
1994 case vmIntrinsics::_max_strict:
1995 n = new MaxINode(a, b);
1996 break;
1997 case vmIntrinsics::_minF:
1998 case vmIntrinsics::_minF_strict:
1999 n = new MinFNode(a, b);
2000 break;
2001 case vmIntrinsics::_maxF:
2002 case vmIntrinsics::_maxF_strict:
2003 n = new MaxFNode(a, b);
2004 break;
2005 case vmIntrinsics::_minD:
2006 case vmIntrinsics::_minD_strict:
2007 n = new MinDNode(a, b);
2008 break;
2009 case vmIntrinsics::_maxD:
2010 case vmIntrinsics::_maxD_strict:
2011 n = new MaxDNode(a, b);
2012 break;
2013 case vmIntrinsics::_minL:
2014 n = new MinLNode(_gvn.C, a, b);
2015 break;
2016 case vmIntrinsics::_maxL:
2017 n = new MaxLNode(_gvn.C, a, b);
2018 break;
2019 default:
2020 fatal_unexpected_iid(id);
2021 break;
2022 }
2023
2024 set_result(_gvn.transform(n));
2025 return true;
2026 }
2027
2028 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2029 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2030 env()->ArithmeticException_instance())) {
2031 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2032 // so let's bail out intrinsic rather than risking deopting again.
2033 return false;
2034 }
2035
2036 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2037 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2038 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2039 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2040
2041 {
2042 PreserveJVMState pjvms(this);
2043 PreserveReexecuteState preexecs(this);
2044 jvms()->set_should_reexecute(true);
2045
2046 set_control(slow_path);
2047 set_i_o(i_o());
2048
2049 builtin_throw(Deoptimization::Reason_intrinsic,
2050 env()->ArithmeticException_instance(),
2051 /*allow_too_many_traps*/ false);
2052 }
2053
2054 set_control(fast_path);
2055 set_result(math);
2056 return true;
2057 }
2058
2059 template <typename OverflowOp>
2060 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2061 typedef typename OverflowOp::MathOp MathOp;
2062
2063 MathOp* mathOp = new MathOp(arg1, arg2);
2064 Node* operation = _gvn.transform( mathOp );
2065 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2066 return inline_math_mathExact(operation, ofcheck);
2067 }
2068
2069 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2070 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2071 }
2072
2073 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2074 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2075 }
2076
2077 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2078 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2079 }
2080
2081 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2082 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2083 }
2084
2085 bool LibraryCallKit::inline_math_negateExactI() {
2086 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2087 }
2088
2089 bool LibraryCallKit::inline_math_negateExactL() {
2090 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2091 }
2092
2093 bool LibraryCallKit::inline_math_multiplyExactI() {
2094 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2095 }
2096
2097 bool LibraryCallKit::inline_math_multiplyExactL() {
2098 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2099 }
2100
2101 bool LibraryCallKit::inline_math_multiplyHigh() {
2102 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2103 return true;
2104 }
2105
2106 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2107 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2108 return true;
2109 }
2110
2111 inline int
2112 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2113 const TypePtr* base_type = TypePtr::NULL_PTR;
2114 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2115 if (base_type == nullptr) {
2116 // Unknown type.
2117 return Type::AnyPtr;
2118 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2119 // Since this is a null+long form, we have to switch to a rawptr.
2120 base = _gvn.transform(new CastX2PNode(offset));
2121 offset = MakeConX(0);
2122 return Type::RawPtr;
2123 } else if (base_type->base() == Type::RawPtr) {
2124 return Type::RawPtr;
2125 } else if (base_type->isa_oopptr()) {
2126 // Base is never null => always a heap address.
2127 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2128 return Type::OopPtr;
2129 }
2130 // Offset is small => always a heap address.
2131 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2132 if (offset_type != nullptr &&
2133 base_type->offset() == 0 && // (should always be?)
2134 offset_type->_lo >= 0 &&
2135 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2136 return Type::OopPtr;
2137 } else if (type == T_OBJECT) {
2138 // off heap access to an oop doesn't make any sense. Has to be on
2139 // heap.
2140 return Type::OopPtr;
2141 }
2142 // Otherwise, it might either be oop+off or null+addr.
2143 return Type::AnyPtr;
2144 } else {
2145 // No information:
2146 return Type::AnyPtr;
2147 }
2148 }
2149
2150 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2151 Node* uncasted_base = base;
2152 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2153 if (kind == Type::RawPtr) {
2154 return basic_plus_adr(top(), uncasted_base, offset);
2155 } else if (kind == Type::AnyPtr) {
2156 assert(base == uncasted_base, "unexpected base change");
2157 if (can_cast) {
2158 if (!_gvn.type(base)->speculative_maybe_null() &&
2159 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2160 // According to profiling, this access is always on
2161 // heap. Casting the base to not null and thus avoiding membars
2162 // around the access should allow better optimizations
2163 Node* null_ctl = top();
2164 base = null_check_oop(base, &null_ctl, true, true, true);
2165 assert(null_ctl->is_top(), "no null control here");
2166 return basic_plus_adr(base, offset);
2167 } else if (_gvn.type(base)->speculative_always_null() &&
2168 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2169 // According to profiling, this access is always off
2170 // heap.
2171 base = null_assert(base);
2172 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2173 offset = MakeConX(0);
2174 return basic_plus_adr(top(), raw_base, offset);
2175 }
2176 }
2177 // We don't know if it's an on heap or off heap access. Fall back
2178 // to raw memory access.
2179 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2180 return basic_plus_adr(top(), raw, offset);
2181 } else {
2182 assert(base == uncasted_base, "unexpected base change");
2183 // We know it's an on heap access so base can't be null
2184 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2185 base = must_be_not_null(base, true);
2186 }
2187 return basic_plus_adr(base, offset);
2188 }
2189 }
2190
2191 //--------------------------inline_number_methods-----------------------------
2192 // inline int Integer.numberOfLeadingZeros(int)
2193 // inline int Long.numberOfLeadingZeros(long)
2194 //
2195 // inline int Integer.numberOfTrailingZeros(int)
2196 // inline int Long.numberOfTrailingZeros(long)
2197 //
2198 // inline int Integer.bitCount(int)
2199 // inline int Long.bitCount(long)
2200 //
2201 // inline char Character.reverseBytes(char)
2202 // inline short Short.reverseBytes(short)
2203 // inline int Integer.reverseBytes(int)
2204 // inline long Long.reverseBytes(long)
2205 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2206 Node* arg = argument(0);
2207 Node* n = nullptr;
2208 switch (id) {
2209 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2210 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2211 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2212 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2213 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2214 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2215 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2216 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2217 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2218 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2219 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2220 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2221 default: fatal_unexpected_iid(id); break;
2222 }
2223 set_result(_gvn.transform(n));
2224 return true;
2225 }
2226
2227 //--------------------------inline_bitshuffle_methods-----------------------------
2228 // inline int Integer.compress(int, int)
2229 // inline int Integer.expand(int, int)
2230 // inline long Long.compress(long, long)
2231 // inline long Long.expand(long, long)
2232 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2233 Node* n = nullptr;
2234 switch (id) {
2235 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2236 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2237 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2238 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2239 default: fatal_unexpected_iid(id); break;
2240 }
2241 set_result(_gvn.transform(n));
2242 return true;
2243 }
2244
2245 //--------------------------inline_number_methods-----------------------------
2246 // inline int Integer.compareUnsigned(int, int)
2247 // inline int Long.compareUnsigned(long, long)
2248 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2249 Node* arg1 = argument(0);
2250 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2251 Node* n = nullptr;
2252 switch (id) {
2253 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2254 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2255 default: fatal_unexpected_iid(id); break;
2256 }
2257 set_result(_gvn.transform(n));
2258 return true;
2259 }
2260
2261 //--------------------------inline_unsigned_divmod_methods-----------------------------
2262 // inline int Integer.divideUnsigned(int, int)
2263 // inline int Integer.remainderUnsigned(int, int)
2264 // inline long Long.divideUnsigned(long, long)
2265 // inline long Long.remainderUnsigned(long, long)
2266 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2267 Node* n = nullptr;
2268 switch (id) {
2269 case vmIntrinsics::_divideUnsigned_i: {
2270 zero_check_int(argument(1));
2271 // Compile-time detect of null-exception
2272 if (stopped()) {
2273 return true; // keep the graph constructed so far
2274 }
2275 n = new UDivINode(control(), argument(0), argument(1));
2276 break;
2277 }
2278 case vmIntrinsics::_divideUnsigned_l: {
2279 zero_check_long(argument(2));
2280 // Compile-time detect of null-exception
2281 if (stopped()) {
2282 return true; // keep the graph constructed so far
2283 }
2284 n = new UDivLNode(control(), argument(0), argument(2));
2285 break;
2286 }
2287 case vmIntrinsics::_remainderUnsigned_i: {
2288 zero_check_int(argument(1));
2289 // Compile-time detect of null-exception
2290 if (stopped()) {
2291 return true; // keep the graph constructed so far
2292 }
2293 n = new UModINode(control(), argument(0), argument(1));
2294 break;
2295 }
2296 case vmIntrinsics::_remainderUnsigned_l: {
2297 zero_check_long(argument(2));
2298 // Compile-time detect of null-exception
2299 if (stopped()) {
2300 return true; // keep the graph constructed so far
2301 }
2302 n = new UModLNode(control(), argument(0), argument(2));
2303 break;
2304 }
2305 default: fatal_unexpected_iid(id); break;
2306 }
2307 set_result(_gvn.transform(n));
2308 return true;
2309 }
2310
2311 //----------------------------inline_unsafe_access----------------------------
2312
2313 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2314 // Attempt to infer a sharper value type from the offset and base type.
2315 ciKlass* sharpened_klass = nullptr;
2316
2317 // See if it is an instance field, with an object type.
2318 if (alias_type->field() != nullptr) {
2319 if (alias_type->field()->type()->is_klass()) {
2320 sharpened_klass = alias_type->field()->type()->as_klass();
2321 }
2322 }
2323
2324 const TypeOopPtr* result = nullptr;
2325 // See if it is a narrow oop array.
2326 if (adr_type->isa_aryptr()) {
2327 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2328 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2329 if (elem_type != nullptr && elem_type->is_loaded()) {
2330 // Sharpen the value type.
2331 result = elem_type;
2332 }
2333 }
2334 }
2335
2336 // The sharpened class might be unloaded if there is no class loader
2337 // contraint in place.
2338 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2339 // Sharpen the value type.
2340 result = TypeOopPtr::make_from_klass(sharpened_klass);
2341 }
2342 if (result != nullptr) {
2343 #ifndef PRODUCT
2344 if (C->print_intrinsics() || C->print_inlining()) {
2345 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2346 tty->print(" sharpened value: "); result->dump(); tty->cr();
2347 }
2348 #endif
2349 }
2350 return result;
2351 }
2352
2353 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2354 switch (kind) {
2355 case Relaxed:
2356 return MO_UNORDERED;
2357 case Opaque:
2358 return MO_RELAXED;
2359 case Acquire:
2360 return MO_ACQUIRE;
2361 case Release:
2362 return MO_RELEASE;
2363 case Volatile:
2364 return MO_SEQ_CST;
2365 default:
2366 ShouldNotReachHere();
2367 return 0;
2368 }
2369 }
2370
2371 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2372 if (callee()->is_static()) return false; // caller must have the capability!
2373 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2374 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2375 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2376 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2377
2378 if (is_reference_type(type)) {
2379 decorators |= ON_UNKNOWN_OOP_REF;
2380 }
2381
2382 if (unaligned) {
2383 decorators |= C2_UNALIGNED;
2384 }
2385
2386 #ifndef PRODUCT
2387 {
2388 ResourceMark rm;
2389 // Check the signatures.
2390 ciSignature* sig = callee()->signature();
2391 #ifdef ASSERT
2392 if (!is_store) {
2393 // Object getReference(Object base, int/long offset), etc.
2394 BasicType rtype = sig->return_type()->basic_type();
2395 assert(rtype == type, "getter must return the expected value");
2396 assert(sig->count() == 2, "oop getter has 2 arguments");
2397 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2398 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2399 } else {
2400 // void putReference(Object base, int/long offset, Object x), etc.
2401 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2402 assert(sig->count() == 3, "oop putter has 3 arguments");
2403 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2404 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2405 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2406 assert(vtype == type, "putter must accept the expected value");
2407 }
2408 #endif // ASSERT
2409 }
2410 #endif //PRODUCT
2411
2412 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2413
2414 Node* receiver = argument(0); // type: oop
2415
2416 // Build address expression.
2417 Node* heap_base_oop = top();
2418
2419 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2420 Node* base = argument(1); // type: oop
2421 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2422 Node* offset = argument(2); // type: long
2423 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2424 // to be plain byte offsets, which are also the same as those accepted
2425 // by oopDesc::field_addr.
2426 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2427 "fieldOffset must be byte-scaled");
2428 // 32-bit machines ignore the high half!
2429 offset = ConvL2X(offset);
2430
2431 // Save state and restore on bailout
2432 uint old_sp = sp();
2433 SafePointNode* old_map = clone_map();
2434
2435 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2436 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2437
2438 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2439 if (type != T_OBJECT) {
2440 decorators |= IN_NATIVE; // off-heap primitive access
2441 } else {
2442 set_map(old_map);
2443 set_sp(old_sp);
2444 return false; // off-heap oop accesses are not supported
2445 }
2446 } else {
2447 heap_base_oop = base; // on-heap or mixed access
2448 }
2449
2450 // Can base be null? Otherwise, always on-heap access.
2451 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2452
2453 if (!can_access_non_heap) {
2454 decorators |= IN_HEAP;
2455 }
2456
2457 Node* val = is_store ? argument(4) : nullptr;
2458
2459 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2460 if (adr_type == TypePtr::NULL_PTR) {
2461 set_map(old_map);
2462 set_sp(old_sp);
2463 return false; // off-heap access with zero address
2464 }
2465
2466 // Try to categorize the address.
2467 Compile::AliasType* alias_type = C->alias_type(adr_type);
2468 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2469
2470 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2471 alias_type->adr_type() == TypeAryPtr::RANGE) {
2472 set_map(old_map);
2473 set_sp(old_sp);
2474 return false; // not supported
2475 }
2476
2477 bool mismatched = false;
2478 BasicType bt = alias_type->basic_type();
2479 if (bt != T_ILLEGAL) {
2480 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2481 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2482 // Alias type doesn't differentiate between byte[] and boolean[]).
2483 // Use address type to get the element type.
2484 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2485 }
2486 if (is_reference_type(bt, true)) {
2487 // accessing an array field with getReference is not a mismatch
2488 bt = T_OBJECT;
2489 }
2490 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2491 // Don't intrinsify mismatched object accesses
2492 set_map(old_map);
2493 set_sp(old_sp);
2494 return false;
2495 }
2496 mismatched = (bt != type);
2497 } else if (alias_type->adr_type()->isa_oopptr()) {
2498 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2499 }
2500
2501 destruct_map_clone(old_map);
2502 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2503
2504 if (mismatched) {
2505 decorators |= C2_MISMATCHED;
2506 }
2507
2508 // First guess at the value type.
2509 const Type *value_type = Type::get_const_basic_type(type);
2510
2511 // Figure out the memory ordering.
2512 decorators |= mo_decorator_for_access_kind(kind);
2513
2514 if (!is_store && type == T_OBJECT) {
2515 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2516 if (tjp != nullptr) {
2517 value_type = tjp;
2518 }
2519 }
2520
2521 receiver = null_check(receiver);
2522 if (stopped()) {
2523 return true;
2524 }
2525 // Heap pointers get a null-check from the interpreter,
2526 // as a courtesy. However, this is not guaranteed by Unsafe,
2527 // and it is not possible to fully distinguish unintended nulls
2528 // from intended ones in this API.
2529
2530 if (!is_store) {
2531 Node* p = nullptr;
2532 // Try to constant fold a load from a constant field
2533 ciField* field = alias_type->field();
2534 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2535 // final or stable field
2536 p = make_constant_from_field(field, heap_base_oop);
2537 }
2538
2539 if (p == nullptr) { // Could not constant fold the load
2540 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2541 // Normalize the value returned by getBoolean in the following cases
2542 if (type == T_BOOLEAN &&
2543 (mismatched ||
2544 heap_base_oop == top() || // - heap_base_oop is null or
2545 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2546 // and the unsafe access is made to large offset
2547 // (i.e., larger than the maximum offset necessary for any
2548 // field access)
2549 ) {
2550 IdealKit ideal = IdealKit(this);
2551 #define __ ideal.
2552 IdealVariable normalized_result(ideal);
2553 __ declarations_done();
2554 __ set(normalized_result, p);
2555 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2556 __ set(normalized_result, ideal.ConI(1));
2557 ideal.end_if();
2558 final_sync(ideal);
2559 p = __ value(normalized_result);
2560 #undef __
2561 }
2562 }
2563 if (type == T_ADDRESS) {
2564 p = gvn().transform(new CastP2XNode(nullptr, p));
2565 p = ConvX2UL(p);
2566 }
2567 // The load node has the control of the preceding MemBarCPUOrder. All
2568 // following nodes will have the control of the MemBarCPUOrder inserted at
2569 // the end of this method. So, pushing the load onto the stack at a later
2570 // point is fine.
2571 set_result(p);
2572 } else {
2573 if (bt == T_ADDRESS) {
2574 // Repackage the long as a pointer.
2575 val = ConvL2X(val);
2576 val = gvn().transform(new CastX2PNode(val));
2577 }
2578 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2579 }
2580
2581 return true;
2582 }
2583
2584 //----------------------------inline_unsafe_load_store----------------------------
2585 // This method serves a couple of different customers (depending on LoadStoreKind):
2586 //
2587 // LS_cmp_swap:
2588 //
2589 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2590 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2591 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2592 //
2593 // LS_cmp_swap_weak:
2594 //
2595 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2596 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2597 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2599 //
2600 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2601 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2602 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2603 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2604 //
2605 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2606 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2607 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2608 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2609 //
2610 // LS_cmp_exchange:
2611 //
2612 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2615 //
2616 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2617 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2619 //
2620 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2621 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2623 //
2624 // LS_get_add:
2625 //
2626 // int getAndAddInt( Object o, long offset, int delta)
2627 // long getAndAddLong(Object o, long offset, long delta)
2628 //
2629 // LS_get_set:
2630 //
2631 // int getAndSet(Object o, long offset, int newValue)
2632 // long getAndSet(Object o, long offset, long newValue)
2633 // Object getAndSet(Object o, long offset, Object newValue)
2634 //
2635 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2636 // This basic scheme here is the same as inline_unsafe_access, but
2637 // differs in enough details that combining them would make the code
2638 // overly confusing. (This is a true fact! I originally combined
2639 // them, but even I was confused by it!) As much code/comments as
2640 // possible are retained from inline_unsafe_access though to make
2641 // the correspondences clearer. - dl
2642
2643 if (callee()->is_static()) return false; // caller must have the capability!
2644
2645 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2646 decorators |= mo_decorator_for_access_kind(access_kind);
2647
2648 #ifndef PRODUCT
2649 BasicType rtype;
2650 {
2651 ResourceMark rm;
2652 // Check the signatures.
2653 ciSignature* sig = callee()->signature();
2654 rtype = sig->return_type()->basic_type();
2655 switch(kind) {
2656 case LS_get_add:
2657 case LS_get_set: {
2658 // Check the signatures.
2659 #ifdef ASSERT
2660 assert(rtype == type, "get and set must return the expected type");
2661 assert(sig->count() == 3, "get and set has 3 arguments");
2662 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2663 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2664 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2665 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2666 #endif // ASSERT
2667 break;
2668 }
2669 case LS_cmp_swap:
2670 case LS_cmp_swap_weak: {
2671 // Check the signatures.
2672 #ifdef ASSERT
2673 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2674 assert(sig->count() == 4, "CAS has 4 arguments");
2675 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2676 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2677 #endif // ASSERT
2678 break;
2679 }
2680 case LS_cmp_exchange: {
2681 // Check the signatures.
2682 #ifdef ASSERT
2683 assert(rtype == type, "CAS must return the expected type");
2684 assert(sig->count() == 4, "CAS has 4 arguments");
2685 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2686 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2687 #endif // ASSERT
2688 break;
2689 }
2690 default:
2691 ShouldNotReachHere();
2692 }
2693 }
2694 #endif //PRODUCT
2695
2696 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2697
2698 // Get arguments:
2699 Node* receiver = nullptr;
2700 Node* base = nullptr;
2701 Node* offset = nullptr;
2702 Node* oldval = nullptr;
2703 Node* newval = nullptr;
2704 switch(kind) {
2705 case LS_cmp_swap:
2706 case LS_cmp_swap_weak:
2707 case LS_cmp_exchange: {
2708 const bool two_slot_type = type2size[type] == 2;
2709 receiver = argument(0); // type: oop
2710 base = argument(1); // type: oop
2711 offset = argument(2); // type: long
2712 oldval = argument(4); // type: oop, int, or long
2713 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2714 break;
2715 }
2716 case LS_get_add:
2717 case LS_get_set: {
2718 receiver = argument(0); // type: oop
2719 base = argument(1); // type: oop
2720 offset = argument(2); // type: long
2721 oldval = nullptr;
2722 newval = argument(4); // type: oop, int, or long
2723 break;
2724 }
2725 default:
2726 ShouldNotReachHere();
2727 }
2728
2729 // Build field offset expression.
2730 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2731 // to be plain byte offsets, which are also the same as those accepted
2732 // by oopDesc::field_addr.
2733 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2734 // 32-bit machines ignore the high half of long offsets
2735 offset = ConvL2X(offset);
2736 // Save state and restore on bailout
2737 uint old_sp = sp();
2738 SafePointNode* old_map = clone_map();
2739 Node* adr = make_unsafe_address(base, offset,type, false);
2740 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2741
2742 Compile::AliasType* alias_type = C->alias_type(adr_type);
2743 BasicType bt = alias_type->basic_type();
2744 if (bt != T_ILLEGAL &&
2745 (is_reference_type(bt) != (type == T_OBJECT))) {
2746 // Don't intrinsify mismatched object accesses.
2747 set_map(old_map);
2748 set_sp(old_sp);
2749 return false;
2750 }
2751
2752 destruct_map_clone(old_map);
2753
2754 // For CAS, unlike inline_unsafe_access, there seems no point in
2755 // trying to refine types. Just use the coarse types here.
2756 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2757 const Type *value_type = Type::get_const_basic_type(type);
2758
2759 switch (kind) {
2760 case LS_get_set:
2761 case LS_cmp_exchange: {
2762 if (type == T_OBJECT) {
2763 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2764 if (tjp != nullptr) {
2765 value_type = tjp;
2766 }
2767 }
2768 break;
2769 }
2770 case LS_cmp_swap:
2771 case LS_cmp_swap_weak:
2772 case LS_get_add:
2773 break;
2774 default:
2775 ShouldNotReachHere();
2776 }
2777
2778 // Null check receiver.
2779 receiver = null_check(receiver);
2780 if (stopped()) {
2781 return true;
2782 }
2783
2784 int alias_idx = C->get_alias_index(adr_type);
2785
2786 if (is_reference_type(type)) {
2787 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2788
2789 // Transformation of a value which could be null pointer (CastPP #null)
2790 // could be delayed during Parse (for example, in adjust_map_after_if()).
2791 // Execute transformation here to avoid barrier generation in such case.
2792 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2793 newval = _gvn.makecon(TypePtr::NULL_PTR);
2794
2795 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2796 // Refine the value to a null constant, when it is known to be null
2797 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2798 }
2799 }
2800
2801 Node* result = nullptr;
2802 switch (kind) {
2803 case LS_cmp_exchange: {
2804 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2805 oldval, newval, value_type, type, decorators);
2806 break;
2807 }
2808 case LS_cmp_swap_weak:
2809 decorators |= C2_WEAK_CMPXCHG;
2810 case LS_cmp_swap: {
2811 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2812 oldval, newval, value_type, type, decorators);
2813 break;
2814 }
2815 case LS_get_set: {
2816 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2817 newval, value_type, type, decorators);
2818 break;
2819 }
2820 case LS_get_add: {
2821 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2822 newval, value_type, type, decorators);
2823 break;
2824 }
2825 default:
2826 ShouldNotReachHere();
2827 }
2828
2829 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2830 set_result(result);
2831 return true;
2832 }
2833
2834 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2835 // Regardless of form, don't allow previous ld/st to move down,
2836 // then issue acquire, release, or volatile mem_bar.
2837 insert_mem_bar(Op_MemBarCPUOrder);
2838 switch(id) {
2839 case vmIntrinsics::_loadFence:
2840 insert_mem_bar(Op_LoadFence);
2841 return true;
2842 case vmIntrinsics::_storeFence:
2843 insert_mem_bar(Op_StoreFence);
2844 return true;
2845 case vmIntrinsics::_storeStoreFence:
2846 insert_mem_bar(Op_StoreStoreFence);
2847 return true;
2848 case vmIntrinsics::_fullFence:
2849 insert_mem_bar(Op_MemBarVolatile);
2850 return true;
2851 default:
2852 fatal_unexpected_iid(id);
2853 return false;
2854 }
2855 }
2856
2857 bool LibraryCallKit::inline_onspinwait() {
2858 insert_mem_bar(Op_OnSpinWait);
2859 return true;
2860 }
2861
2862 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2863 if (!kls->is_Con()) {
2864 return true;
2865 }
2866 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2867 if (klsptr == nullptr) {
2868 return true;
2869 }
2870 ciInstanceKlass* ik = klsptr->instance_klass();
2871 // don't need a guard for a klass that is already initialized
2872 return !ik->is_initialized();
2873 }
2874
2875 //----------------------------inline_unsafe_writeback0-------------------------
2876 // public native void Unsafe.writeback0(long address)
2877 bool LibraryCallKit::inline_unsafe_writeback0() {
2878 if (!Matcher::has_match_rule(Op_CacheWB)) {
2879 return false;
2880 }
2881 #ifndef PRODUCT
2882 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2883 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2884 ciSignature* sig = callee()->signature();
2885 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2886 #endif
2887 null_check_receiver(); // null-check, then ignore
2888 Node *addr = argument(1);
2889 addr = new CastX2PNode(addr);
2890 addr = _gvn.transform(addr);
2891 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2892 flush = _gvn.transform(flush);
2893 set_memory(flush, TypeRawPtr::BOTTOM);
2894 return true;
2895 }
2896
2897 //----------------------------inline_unsafe_writeback0-------------------------
2898 // public native void Unsafe.writeback0(long address)
2899 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2900 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2901 return false;
2902 }
2903 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2904 return false;
2905 }
2906 #ifndef PRODUCT
2907 assert(Matcher::has_match_rule(Op_CacheWB),
2908 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2909 : "found match rule for CacheWBPostSync but not CacheWB"));
2910
2911 #endif
2912 null_check_receiver(); // null-check, then ignore
2913 Node *sync;
2914 if (is_pre) {
2915 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2916 } else {
2917 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2918 }
2919 sync = _gvn.transform(sync);
2920 set_memory(sync, TypeRawPtr::BOTTOM);
2921 return true;
2922 }
2923
2924 //----------------------------inline_unsafe_allocate---------------------------
2925 // public native Object Unsafe.allocateInstance(Class<?> cls);
2926 bool LibraryCallKit::inline_unsafe_allocate() {
2927
2928 #if INCLUDE_JVMTI
2929 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2930 return false;
2931 }
2932 #endif //INCLUDE_JVMTI
2933
2934 if (callee()->is_static()) return false; // caller must have the capability!
2935
2936 null_check_receiver(); // null-check, then ignore
2937 Node* cls = null_check(argument(1));
2938 if (stopped()) return true;
2939
2940 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2941 kls = null_check(kls);
2942 if (stopped()) return true; // argument was like int.class
2943
2944 #if INCLUDE_JVMTI
2945 // Don't try to access new allocated obj in the intrinsic.
2946 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2947 // Deoptimize and allocate in interpreter instead.
2948 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2949 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2950 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2951 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2952 {
2953 BuildCutout unless(this, tst, PROB_MAX);
2954 uncommon_trap(Deoptimization::Reason_intrinsic,
2955 Deoptimization::Action_make_not_entrant);
2956 }
2957 if (stopped()) {
2958 return true;
2959 }
2960 #endif //INCLUDE_JVMTI
2961
2962 Node* test = nullptr;
2963 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2964 // Note: The argument might still be an illegal value like
2965 // Serializable.class or Object[].class. The runtime will handle it.
2966 // But we must make an explicit check for initialization.
2967 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2968 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2969 // can generate code to load it as unsigned byte.
2970 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2971 Node* bits = intcon(InstanceKlass::fully_initialized);
2972 test = _gvn.transform(new SubINode(inst, bits));
2973 // The 'test' is non-zero if we need to take a slow path.
2974 }
2975
2976 Node* obj = new_instance(kls, test);
2977 set_result(obj);
2978 return true;
2979 }
2980
2981 //------------------------inline_native_time_funcs--------------
2982 // inline code for System.currentTimeMillis() and System.nanoTime()
2983 // these have the same type and signature
2984 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2985 const TypeFunc* tf = OptoRuntime::void_long_Type();
2986 const TypePtr* no_memory_effects = nullptr;
2987 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2988 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2989 #ifdef ASSERT
2990 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2991 assert(value_top == top(), "second value must be top");
2992 #endif
2993 set_result(value);
2994 return true;
2995 }
2996
2997
2998 #if INCLUDE_JVMTI
2999
3000 // When notifications are disabled then just update the VTMS transition bit and return.
3001 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3002 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3003 if (!DoJVMTIVirtualThreadTransitions) {
3004 return true;
3005 }
3006 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3007 IdealKit ideal(this);
3008
3009 Node* ONE = ideal.ConI(1);
3010 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3011 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3012 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3013
3014 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3015 sync_kit(ideal);
3016 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3017 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3018 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3019 ideal.sync_kit(this);
3020 } ideal.else_(); {
3021 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3022 Node* thread = ideal.thread();
3023 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3024 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3025
3026 sync_kit(ideal);
3027 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3028 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3029
3030 ideal.sync_kit(this);
3031 } ideal.end_if();
3032 final_sync(ideal);
3033
3034 return true;
3035 }
3036
3037 // Always update the is_disable_suspend bit.
3038 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3039 if (!DoJVMTIVirtualThreadTransitions) {
3040 return true;
3041 }
3042 IdealKit ideal(this);
3043
3044 {
3045 // unconditionally update the is_disable_suspend bit in current JavaThread
3046 Node* thread = ideal.thread();
3047 Node* arg = _gvn.transform(argument(0)); // argument for notification
3048 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3049 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3050
3051 sync_kit(ideal);
3052 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3053 ideal.sync_kit(this);
3054 }
3055 final_sync(ideal);
3056
3057 return true;
3058 }
3059
3060 #endif // INCLUDE_JVMTI
3061
3062 #ifdef JFR_HAVE_INTRINSICS
3063
3064 /**
3065 * if oop->klass != null
3066 * // normal class
3067 * epoch = _epoch_state ? 2 : 1
3068 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3069 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3070 * }
3071 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3072 * else
3073 * // primitive class
3074 * if oop->array_klass != null
3075 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3076 * else
3077 * id = LAST_TYPE_ID + 1 // void class path
3078 * if (!signaled)
3079 * signaled = true
3080 */
3081 bool LibraryCallKit::inline_native_classID() {
3082 Node* cls = argument(0);
3083
3084 IdealKit ideal(this);
3085 #define __ ideal.
3086 IdealVariable result(ideal); __ declarations_done();
3087 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3088 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3089 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3090
3091
3092 __ if_then(kls, BoolTest::ne, null()); {
3093 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3094 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3095
3096 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3097 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3098 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3099 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3100 mask = _gvn.transform(new OrLNode(mask, epoch));
3101 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3102
3103 float unlikely = PROB_UNLIKELY(0.999);
3104 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3105 sync_kit(ideal);
3106 make_runtime_call(RC_LEAF,
3107 OptoRuntime::class_id_load_barrier_Type(),
3108 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3109 "class id load barrier",
3110 TypePtr::BOTTOM,
3111 kls);
3112 ideal.sync_kit(this);
3113 } __ end_if();
3114
3115 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3116 } __ else_(); {
3117 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3118 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3119 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3120 __ if_then(array_kls, BoolTest::ne, null()); {
3121 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3122 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3123 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3124 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3125 } __ else_(); {
3126 // void class case
3127 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3128 } __ end_if();
3129
3130 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3131 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3132 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3133 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3134 } __ end_if();
3135 } __ end_if();
3136
3137 final_sync(ideal);
3138 set_result(ideal.value(result));
3139 #undef __
3140 return true;
3141 }
3142
3143 //------------------------inline_native_jvm_commit------------------
3144 bool LibraryCallKit::inline_native_jvm_commit() {
3145 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3146
3147 // Save input memory and i_o state.
3148 Node* input_memory_state = reset_memory();
3149 set_all_memory(input_memory_state);
3150 Node* input_io_state = i_o();
3151
3152 // TLS.
3153 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3154 // Jfr java buffer.
3155 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3156 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3157 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3158
3159 // Load the current value of the notified field in the JfrThreadLocal.
3160 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3161 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3162
3163 // Test for notification.
3164 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3165 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3166 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3167
3168 // True branch, is notified.
3169 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3170 set_control(is_notified);
3171
3172 // Reset notified state.
3173 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3174 Node* notified_reset_memory = reset_memory();
3175
3176 // 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.
3177 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3178 // Convert the machine-word to a long.
3179 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3180
3181 // False branch, not notified.
3182 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3183 set_control(not_notified);
3184 set_all_memory(input_memory_state);
3185
3186 // Arg is the next position as a long.
3187 Node* arg = argument(0);
3188 // Convert long to machine-word.
3189 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3190
3191 // Store the next_position to the underlying jfr java buffer.
3192 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3193
3194 Node* commit_memory = reset_memory();
3195 set_all_memory(commit_memory);
3196
3197 // 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.
3198 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3199 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3200 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3201
3202 // And flags with lease constant.
3203 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3204
3205 // Branch on lease to conditionalize returning the leased java buffer.
3206 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3207 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3208 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3209
3210 // False branch, not a lease.
3211 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3212
3213 // True branch, is lease.
3214 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3215 set_control(is_lease);
3216
3217 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3218 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3219 OptoRuntime::void_void_Type(),
3220 SharedRuntime::jfr_return_lease(),
3221 "return_lease", TypePtr::BOTTOM);
3222 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3223
3224 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3225 record_for_igvn(lease_compare_rgn);
3226 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3227 record_for_igvn(lease_compare_mem);
3228 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3229 record_for_igvn(lease_compare_io);
3230 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3231 record_for_igvn(lease_result_value);
3232
3233 // Update control and phi nodes.
3234 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3235 lease_compare_rgn->init_req(_false_path, not_lease);
3236
3237 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3238 lease_compare_mem->init_req(_false_path, commit_memory);
3239
3240 lease_compare_io->init_req(_true_path, i_o());
3241 lease_compare_io->init_req(_false_path, input_io_state);
3242
3243 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3244 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3245
3246 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3247 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3248 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3249 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3250
3251 // Update control and phi nodes.
3252 result_rgn->init_req(_true_path, is_notified);
3253 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3254
3255 result_mem->init_req(_true_path, notified_reset_memory);
3256 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3257
3258 result_io->init_req(_true_path, input_io_state);
3259 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3260
3261 result_value->init_req(_true_path, current_pos);
3262 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3263
3264 // Set output state.
3265 set_control(_gvn.transform(result_rgn));
3266 set_all_memory(_gvn.transform(result_mem));
3267 set_i_o(_gvn.transform(result_io));
3268 set_result(result_rgn, result_value);
3269 return true;
3270 }
3271
3272 /*
3273 * The intrinsic is a model of this pseudo-code:
3274 *
3275 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3276 * jobject h_event_writer = tl->java_event_writer();
3277 * if (h_event_writer == nullptr) {
3278 * return nullptr;
3279 * }
3280 * oop threadObj = Thread::threadObj();
3281 * oop vthread = java_lang_Thread::vthread(threadObj);
3282 * traceid tid;
3283 * bool pinVirtualThread;
3284 * bool excluded;
3285 * if (vthread != threadObj) { // i.e. current thread is virtual
3286 * tid = java_lang_Thread::tid(vthread);
3287 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3288 * pinVirtualThread = VMContinuations;
3289 * excluded = vthread_epoch_raw & excluded_mask;
3290 * if (!excluded) {
3291 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3292 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3293 * if (vthread_epoch != current_epoch) {
3294 * write_checkpoint();
3295 * }
3296 * }
3297 * } else {
3298 * tid = java_lang_Thread::tid(threadObj);
3299 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3300 * pinVirtualThread = false;
3301 * excluded = thread_epoch_raw & excluded_mask;
3302 * }
3303 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3304 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3305 * if (tid_in_event_writer != tid) {
3306 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3307 * setField(event_writer, "excluded", excluded);
3308 * setField(event_writer, "threadID", tid);
3309 * }
3310 * return event_writer
3311 */
3312 bool LibraryCallKit::inline_native_getEventWriter() {
3313 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3314
3315 // Save input memory and i_o state.
3316 Node* input_memory_state = reset_memory();
3317 set_all_memory(input_memory_state);
3318 Node* input_io_state = i_o();
3319
3320 // The most significant bit of the u2 is used to denote thread exclusion
3321 Node* excluded_shift = _gvn.intcon(15);
3322 Node* excluded_mask = _gvn.intcon(1 << 15);
3323 // The epoch generation is the range [1-32767]
3324 Node* epoch_mask = _gvn.intcon(32767);
3325
3326 // TLS
3327 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3328
3329 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3330 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3331
3332 // Load the eventwriter jobject handle.
3333 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3334
3335 // Null check the jobject handle.
3336 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3337 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3338 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3339
3340 // False path, jobj is null.
3341 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3342
3343 // True path, jobj is not null.
3344 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3345
3346 set_control(jobj_is_not_null);
3347
3348 // Load the threadObj for the CarrierThread.
3349 Node* threadObj = generate_current_thread(tls_ptr);
3350
3351 // Load the vthread.
3352 Node* vthread = generate_virtual_thread(tls_ptr);
3353
3354 // If vthread != threadObj, this is a virtual thread.
3355 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3356 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3357 IfNode* iff_vthread_not_equal_threadObj =
3358 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3359
3360 // False branch, fallback to threadObj.
3361 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3362 set_control(vthread_equal_threadObj);
3363
3364 // Load the tid field from the vthread object.
3365 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3366
3367 // Load the raw epoch value from the threadObj.
3368 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3369 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3370 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3371 TypeInt::CHAR, T_CHAR,
3372 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3373
3374 // Mask off the excluded information from the epoch.
3375 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3376
3377 // True branch, this is a virtual thread.
3378 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3379 set_control(vthread_not_equal_threadObj);
3380
3381 // Load the tid field from the vthread object.
3382 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3383
3384 // Continuation support determines if a virtual thread should be pinned.
3385 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3386 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3387
3388 // Load the raw epoch value from the vthread.
3389 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3390 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3391 TypeInt::CHAR, T_CHAR,
3392 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3393
3394 // Mask off the excluded information from the epoch.
3395 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3396
3397 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3398 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3399 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3400 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3401
3402 // False branch, vthread is excluded, no need to write epoch info.
3403 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3404
3405 // True branch, vthread is included, update epoch info.
3406 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3407 set_control(included);
3408
3409 // Get epoch value.
3410 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3411
3412 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3413 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3414 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3415
3416 // Compare the epoch in the vthread to the current epoch generation.
3417 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3418 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3419 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3420
3421 // False path, epoch is equal, checkpoint information is valid.
3422 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3423
3424 // True path, epoch is not equal, write a checkpoint for the vthread.
3425 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3426
3427 set_control(epoch_is_not_equal);
3428
3429 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3430 // The call also updates the native thread local thread id and the vthread with the current epoch.
3431 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3432 OptoRuntime::jfr_write_checkpoint_Type(),
3433 SharedRuntime::jfr_write_checkpoint(),
3434 "write_checkpoint", TypePtr::BOTTOM);
3435 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3436
3437 // vthread epoch != current epoch
3438 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3439 record_for_igvn(epoch_compare_rgn);
3440 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3441 record_for_igvn(epoch_compare_mem);
3442 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3443 record_for_igvn(epoch_compare_io);
3444
3445 // Update control and phi nodes.
3446 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3447 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3448 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3449 epoch_compare_mem->init_req(_false_path, input_memory_state);
3450 epoch_compare_io->init_req(_true_path, i_o());
3451 epoch_compare_io->init_req(_false_path, input_io_state);
3452
3453 // excluded != true
3454 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3455 record_for_igvn(exclude_compare_rgn);
3456 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3457 record_for_igvn(exclude_compare_mem);
3458 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3459 record_for_igvn(exclude_compare_io);
3460
3461 // Update control and phi nodes.
3462 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3463 exclude_compare_rgn->init_req(_false_path, excluded);
3464 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3465 exclude_compare_mem->init_req(_false_path, input_memory_state);
3466 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3467 exclude_compare_io->init_req(_false_path, input_io_state);
3468
3469 // vthread != threadObj
3470 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3471 record_for_igvn(vthread_compare_rgn);
3472 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3473 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3474 record_for_igvn(vthread_compare_io);
3475 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3476 record_for_igvn(tid);
3477 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3478 record_for_igvn(exclusion);
3479 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3480 record_for_igvn(pinVirtualThread);
3481
3482 // Update control and phi nodes.
3483 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3484 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3485 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3486 vthread_compare_mem->init_req(_false_path, input_memory_state);
3487 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3488 vthread_compare_io->init_req(_false_path, input_io_state);
3489 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3490 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3491 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3492 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3493 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3494 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3495
3496 // Update branch state.
3497 set_control(_gvn.transform(vthread_compare_rgn));
3498 set_all_memory(_gvn.transform(vthread_compare_mem));
3499 set_i_o(_gvn.transform(vthread_compare_io));
3500
3501 // Load the event writer oop by dereferencing the jobject handle.
3502 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3503 assert(klass_EventWriter->is_loaded(), "invariant");
3504 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3505 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3506 const TypeOopPtr* const xtype = aklass->as_instance_type();
3507 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3508 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3509
3510 // Load the current thread id from the event writer object.
3511 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3512 // Get the field offset to, conditionally, store an updated tid value later.
3513 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3514 // Get the field offset to, conditionally, store an updated exclusion value later.
3515 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3516 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3517 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3518
3519 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3520 record_for_igvn(event_writer_tid_compare_rgn);
3521 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3522 record_for_igvn(event_writer_tid_compare_mem);
3523 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3524 record_for_igvn(event_writer_tid_compare_io);
3525
3526 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3527 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3528 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3529 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3530
3531 // False path, tids are the same.
3532 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3533
3534 // True path, tid is not equal, need to update the tid in the event writer.
3535 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3536 record_for_igvn(tid_is_not_equal);
3537
3538 // Store the pin state to the event writer.
3539 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3540
3541 // Store the exclusion state to the event writer.
3542 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3543 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3544
3545 // Store the tid to the event writer.
3546 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3547
3548 // Update control and phi nodes.
3549 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3550 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3551 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3552 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3553 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3554 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3555
3556 // Result of top level CFG, Memory, IO and Value.
3557 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3558 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3559 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3560 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3561
3562 // Result control.
3563 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3564 result_rgn->init_req(_false_path, jobj_is_null);
3565
3566 // Result memory.
3567 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3568 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3569
3570 // Result IO.
3571 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3572 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3573
3574 // Result value.
3575 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3576 result_value->init_req(_false_path, null()); // return null
3577
3578 // Set output state.
3579 set_control(_gvn.transform(result_rgn));
3580 set_all_memory(_gvn.transform(result_mem));
3581 set_i_o(_gvn.transform(result_io));
3582 set_result(result_rgn, result_value);
3583 return true;
3584 }
3585
3586 /*
3587 * The intrinsic is a model of this pseudo-code:
3588 *
3589 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3590 * if (carrierThread != thread) { // is virtual thread
3591 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3592 * bool excluded = vthread_epoch_raw & excluded_mask;
3593 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3594 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3595 * if (!excluded) {
3596 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3597 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3598 * }
3599 * Atomic::release_store(&tl->_vthread, true);
3600 * return;
3601 * }
3602 * Atomic::release_store(&tl->_vthread, false);
3603 */
3604 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3605 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3606
3607 Node* input_memory_state = reset_memory();
3608 set_all_memory(input_memory_state);
3609
3610 // The most significant bit of the u2 is used to denote thread exclusion
3611 Node* excluded_mask = _gvn.intcon(1 << 15);
3612 // The epoch generation is the range [1-32767]
3613 Node* epoch_mask = _gvn.intcon(32767);
3614
3615 Node* const carrierThread = generate_current_thread(jt);
3616 // If thread != carrierThread, this is a virtual thread.
3617 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3618 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3619 IfNode* iff_thread_not_equal_carrierThread =
3620 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3621
3622 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3623
3624 // False branch, is carrierThread.
3625 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3626 // Store release
3627 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3628
3629 set_all_memory(input_memory_state);
3630
3631 // True branch, is virtual thread.
3632 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3633 set_control(thread_not_equal_carrierThread);
3634
3635 // Load the raw epoch value from the vthread.
3636 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3637 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3638 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3639
3640 // Mask off the excluded information from the epoch.
3641 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3642
3643 // Load the tid field from the thread.
3644 Node* tid = load_field_from_object(thread, "tid", "J");
3645
3646 // Store the vthread tid to the jfr thread local.
3647 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3648 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3649
3650 // Branch is_excluded to conditionalize updating the epoch .
3651 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3652 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3653 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3654
3655 // True branch, vthread is excluded, no need to write epoch info.
3656 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3657 set_control(excluded);
3658 Node* vthread_is_excluded = _gvn.intcon(1);
3659
3660 // False branch, vthread is included, update epoch info.
3661 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3662 set_control(included);
3663 Node* vthread_is_included = _gvn.intcon(0);
3664
3665 // Get epoch value.
3666 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3667
3668 // Store the vthread epoch to the jfr thread local.
3669 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3670 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3671
3672 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3673 record_for_igvn(excluded_rgn);
3674 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3675 record_for_igvn(excluded_mem);
3676 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3677 record_for_igvn(exclusion);
3678
3679 // Merge the excluded control and memory.
3680 excluded_rgn->init_req(_true_path, excluded);
3681 excluded_rgn->init_req(_false_path, included);
3682 excluded_mem->init_req(_true_path, tid_memory);
3683 excluded_mem->init_req(_false_path, included_memory);
3684 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3685 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3686
3687 // Set intermediate state.
3688 set_control(_gvn.transform(excluded_rgn));
3689 set_all_memory(excluded_mem);
3690
3691 // Store the vthread exclusion state to the jfr thread local.
3692 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3693 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3694
3695 // Store release
3696 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3697
3698 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3699 record_for_igvn(thread_compare_rgn);
3700 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3701 record_for_igvn(thread_compare_mem);
3702 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3703 record_for_igvn(vthread);
3704
3705 // Merge the thread_compare control and memory.
3706 thread_compare_rgn->init_req(_true_path, control());
3707 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3708 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3709 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3710
3711 // Set output state.
3712 set_control(_gvn.transform(thread_compare_rgn));
3713 set_all_memory(_gvn.transform(thread_compare_mem));
3714 }
3715
3716 #endif // JFR_HAVE_INTRINSICS
3717
3718 //------------------------inline_native_currentCarrierThread------------------
3719 bool LibraryCallKit::inline_native_currentCarrierThread() {
3720 Node* junk = nullptr;
3721 set_result(generate_current_thread(junk));
3722 return true;
3723 }
3724
3725 //------------------------inline_native_currentThread------------------
3726 bool LibraryCallKit::inline_native_currentThread() {
3727 Node* junk = nullptr;
3728 set_result(generate_virtual_thread(junk));
3729 return true;
3730 }
3731
3732 //------------------------inline_native_setVthread------------------
3733 bool LibraryCallKit::inline_native_setCurrentThread() {
3734 assert(C->method()->changes_current_thread(),
3735 "method changes current Thread but is not annotated ChangesCurrentThread");
3736 Node* arr = argument(1);
3737 Node* thread = _gvn.transform(new ThreadLocalNode());
3738 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3739 Node* thread_obj_handle
3740 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3741 thread_obj_handle = _gvn.transform(thread_obj_handle);
3742 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3743 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3744
3745 // Change the _monitor_owner_id of the JavaThread
3746 Node* tid = load_field_from_object(arr, "tid", "J");
3747 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3748 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3749
3750 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3751 return true;
3752 }
3753
3754 const Type* LibraryCallKit::scopedValueCache_type() {
3755 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3756 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3757 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3758
3759 // Because we create the scopedValue cache lazily we have to make the
3760 // type of the result BotPTR.
3761 bool xk = etype->klass_is_exact();
3762 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3763 return objects_type;
3764 }
3765
3766 Node* LibraryCallKit::scopedValueCache_helper() {
3767 Node* thread = _gvn.transform(new ThreadLocalNode());
3768 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3769 // We cannot use immutable_memory() because we might flip onto a
3770 // different carrier thread, at which point we'll need to use that
3771 // carrier thread's cache.
3772 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3773 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3774 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3775 }
3776
3777 //------------------------inline_native_scopedValueCache------------------
3778 bool LibraryCallKit::inline_native_scopedValueCache() {
3779 Node* cache_obj_handle = scopedValueCache_helper();
3780 const Type* objects_type = scopedValueCache_type();
3781 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3782
3783 return true;
3784 }
3785
3786 //------------------------inline_native_setScopedValueCache------------------
3787 bool LibraryCallKit::inline_native_setScopedValueCache() {
3788 Node* arr = argument(0);
3789 Node* cache_obj_handle = scopedValueCache_helper();
3790 const Type* objects_type = scopedValueCache_type();
3791
3792 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3793 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3794
3795 return true;
3796 }
3797
3798 //------------------------inline_native_Continuation_pin and unpin-----------
3799
3800 // Shared implementation routine for both pin and unpin.
3801 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3802 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3803
3804 // Save input memory.
3805 Node* input_memory_state = reset_memory();
3806 set_all_memory(input_memory_state);
3807
3808 // TLS
3809 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3810 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3811 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3812
3813 // Null check the last continuation object.
3814 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3815 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3816 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3817
3818 // False path, last continuation is null.
3819 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3820
3821 // True path, last continuation is not null.
3822 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3823
3824 set_control(continuation_is_not_null);
3825
3826 // Load the pin count from the last continuation.
3827 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3828 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3829
3830 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3831 Node* pin_count_rhs;
3832 if (unpin) {
3833 pin_count_rhs = _gvn.intcon(0);
3834 } else {
3835 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3836 }
3837 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3838 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3839 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3840
3841 // True branch, pin count over/underflow.
3842 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3843 {
3844 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3845 // which will throw IllegalStateException for pin count over/underflow.
3846 // No memory changed so far - we can use memory create by reset_memory()
3847 // at the beginning of this intrinsic. No need to call reset_memory() again.
3848 PreserveJVMState pjvms(this);
3849 set_control(pin_count_over_underflow);
3850 uncommon_trap(Deoptimization::Reason_intrinsic,
3851 Deoptimization::Action_none);
3852 assert(stopped(), "invariant");
3853 }
3854
3855 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3856 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3857 set_control(valid_pin_count);
3858
3859 Node* next_pin_count;
3860 if (unpin) {
3861 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3862 } else {
3863 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3864 }
3865
3866 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3867
3868 // Result of top level CFG and Memory.
3869 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3870 record_for_igvn(result_rgn);
3871 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3872 record_for_igvn(result_mem);
3873
3874 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3875 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3876 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3877 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3878
3879 // Set output state.
3880 set_control(_gvn.transform(result_rgn));
3881 set_all_memory(_gvn.transform(result_mem));
3882
3883 return true;
3884 }
3885
3886 //---------------------------load_mirror_from_klass----------------------------
3887 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3888 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3889 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3890 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3891 // mirror = ((OopHandle)mirror)->resolve();
3892 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3893 }
3894
3895 //-----------------------load_klass_from_mirror_common-------------------------
3896 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3897 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3898 // and branch to the given path on the region.
3899 // If never_see_null, take an uncommon trap on null, so we can optimistically
3900 // compile for the non-null case.
3901 // If the region is null, force never_see_null = true.
3902 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3903 bool never_see_null,
3904 RegionNode* region,
3905 int null_path,
3906 int offset) {
3907 if (region == nullptr) never_see_null = true;
3908 Node* p = basic_plus_adr(mirror, offset);
3909 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3910 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3911 Node* null_ctl = top();
3912 kls = null_check_oop(kls, &null_ctl, never_see_null);
3913 if (region != nullptr) {
3914 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3915 region->init_req(null_path, null_ctl);
3916 } else {
3917 assert(null_ctl == top(), "no loose ends");
3918 }
3919 return kls;
3920 }
3921
3922 //--------------------(inline_native_Class_query helpers)---------------------
3923 // Use this for JVM_ACC_INTERFACE.
3924 // Fall through if (mods & mask) == bits, take the guard otherwise.
3925 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3926 ByteSize offset, const Type* type, BasicType bt) {
3927 // Branch around if the given klass has the given modifier bit set.
3928 // Like generate_guard, adds a new path onto the region.
3929 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3930 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3931 Node* mask = intcon(modifier_mask);
3932 Node* bits = intcon(modifier_bits);
3933 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3934 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3935 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3936 return generate_fair_guard(bol, region);
3937 }
3938 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3939 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3940 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3941 }
3942
3943 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3944 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3945 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3946 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3947 }
3948
3949 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3950 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3951 }
3952
3953 //-------------------------inline_native_Class_query-------------------
3954 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3955 const Type* return_type = TypeInt::BOOL;
3956 Node* prim_return_value = top(); // what happens if it's a primitive class?
3957 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3958 bool expect_prim = false; // most of these guys expect to work on refs
3959
3960 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3961
3962 Node* mirror = argument(0);
3963 Node* obj = top();
3964
3965 switch (id) {
3966 case vmIntrinsics::_isInstance:
3967 // nothing is an instance of a primitive type
3968 prim_return_value = intcon(0);
3969 obj = argument(1);
3970 break;
3971 case vmIntrinsics::_isHidden:
3972 prim_return_value = intcon(0);
3973 break;
3974 case vmIntrinsics::_getSuperclass:
3975 prim_return_value = null();
3976 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3977 break;
3978 case vmIntrinsics::_getClassAccessFlags:
3979 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3980 return_type = TypeInt::CHAR;
3981 break;
3982 default:
3983 fatal_unexpected_iid(id);
3984 break;
3985 }
3986
3987 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3988 if (mirror_con == nullptr) return false; // cannot happen?
3989
3990 #ifndef PRODUCT
3991 if (C->print_intrinsics() || C->print_inlining()) {
3992 ciType* k = mirror_con->java_mirror_type();
3993 if (k) {
3994 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3995 k->print_name();
3996 tty->cr();
3997 }
3998 }
3999 #endif
4000
4001 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4002 RegionNode* region = new RegionNode(PATH_LIMIT);
4003 record_for_igvn(region);
4004 PhiNode* phi = new PhiNode(region, return_type);
4005
4006 // The mirror will never be null of Reflection.getClassAccessFlags, however
4007 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4008 // if it is. See bug 4774291.
4009
4010 // For Reflection.getClassAccessFlags(), the null check occurs in
4011 // the wrong place; see inline_unsafe_access(), above, for a similar
4012 // situation.
4013 mirror = null_check(mirror);
4014 // If mirror or obj is dead, only null-path is taken.
4015 if (stopped()) return true;
4016
4017 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4018
4019 // Now load the mirror's klass metaobject, and null-check it.
4020 // Side-effects region with the control path if the klass is null.
4021 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4022 // If kls is null, we have a primitive mirror.
4023 phi->init_req(_prim_path, prim_return_value);
4024 if (stopped()) { set_result(region, phi); return true; }
4025 bool safe_for_replace = (region->in(_prim_path) == top());
4026
4027 Node* p; // handy temp
4028 Node* null_ctl;
4029
4030 // Now that we have the non-null klass, we can perform the real query.
4031 // For constant classes, the query will constant-fold in LoadNode::Value.
4032 Node* query_value = top();
4033 switch (id) {
4034 case vmIntrinsics::_isInstance:
4035 // nothing is an instance of a primitive type
4036 query_value = gen_instanceof(obj, kls, safe_for_replace);
4037 break;
4038
4039 case vmIntrinsics::_isHidden:
4040 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4041 if (generate_hidden_class_guard(kls, region) != nullptr)
4042 // A guard was added. If the guard is taken, it was an hidden class.
4043 phi->add_req(intcon(1));
4044 // If we fall through, it's a plain class.
4045 query_value = intcon(0);
4046 break;
4047
4048
4049 case vmIntrinsics::_getSuperclass:
4050 // The rules here are somewhat unfortunate, but we can still do better
4051 // with random logic than with a JNI call.
4052 // Interfaces store null or Object as _super, but must report null.
4053 // Arrays store an intermediate super as _super, but must report Object.
4054 // Other types can report the actual _super.
4055 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4056 if (generate_interface_guard(kls, region) != nullptr)
4057 // A guard was added. If the guard is taken, it was an interface.
4058 phi->add_req(null());
4059 if (generate_array_guard(kls, region) != nullptr)
4060 // A guard was added. If the guard is taken, it was an array.
4061 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4062 // If we fall through, it's a plain class. Get its _super.
4063 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4064 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4065 null_ctl = top();
4066 kls = null_check_oop(kls, &null_ctl);
4067 if (null_ctl != top()) {
4068 // If the guard is taken, Object.superClass is null (both klass and mirror).
4069 region->add_req(null_ctl);
4070 phi ->add_req(null());
4071 }
4072 if (!stopped()) {
4073 query_value = load_mirror_from_klass(kls);
4074 }
4075 break;
4076
4077 case vmIntrinsics::_getClassAccessFlags:
4078 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4079 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4080 break;
4081
4082 default:
4083 fatal_unexpected_iid(id);
4084 break;
4085 }
4086
4087 // Fall-through is the normal case of a query to a real class.
4088 phi->init_req(1, query_value);
4089 region->init_req(1, control());
4090
4091 C->set_has_split_ifs(true); // Has chance for split-if optimization
4092 set_result(region, phi);
4093 return true;
4094 }
4095
4096 //-------------------------inline_Class_cast-------------------
4097 bool LibraryCallKit::inline_Class_cast() {
4098 Node* mirror = argument(0); // Class
4099 Node* obj = argument(1);
4100 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4101 if (mirror_con == nullptr) {
4102 return false; // dead path (mirror->is_top()).
4103 }
4104 if (obj == nullptr || obj->is_top()) {
4105 return false; // dead path
4106 }
4107 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4108
4109 // First, see if Class.cast() can be folded statically.
4110 // java_mirror_type() returns non-null for compile-time Class constants.
4111 ciType* tm = mirror_con->java_mirror_type();
4112 if (tm != nullptr && tm->is_klass() &&
4113 tp != nullptr) {
4114 if (!tp->is_loaded()) {
4115 // Don't use intrinsic when class is not loaded.
4116 return false;
4117 } else {
4118 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4119 if (static_res == Compile::SSC_always_true) {
4120 // isInstance() is true - fold the code.
4121 set_result(obj);
4122 return true;
4123 } else if (static_res == Compile::SSC_always_false) {
4124 // Don't use intrinsic, have to throw ClassCastException.
4125 // If the reference is null, the non-intrinsic bytecode will
4126 // be optimized appropriately.
4127 return false;
4128 }
4129 }
4130 }
4131
4132 // Bailout intrinsic and do normal inlining if exception path is frequent.
4133 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4134 return false;
4135 }
4136
4137 // Generate dynamic checks.
4138 // Class.cast() is java implementation of _checkcast bytecode.
4139 // Do checkcast (Parse::do_checkcast()) optimizations here.
4140
4141 mirror = null_check(mirror);
4142 // If mirror is dead, only null-path is taken.
4143 if (stopped()) {
4144 return true;
4145 }
4146
4147 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4148 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4149 RegionNode* region = new RegionNode(PATH_LIMIT);
4150 record_for_igvn(region);
4151
4152 // Now load the mirror's klass metaobject, and null-check it.
4153 // If kls is null, we have a primitive mirror and
4154 // nothing is an instance of a primitive type.
4155 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4156
4157 Node* res = top();
4158 if (!stopped()) {
4159 Node* bad_type_ctrl = top();
4160 // Do checkcast optimizations.
4161 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4162 region->init_req(_bad_type_path, bad_type_ctrl);
4163 }
4164 if (region->in(_prim_path) != top() ||
4165 region->in(_bad_type_path) != top()) {
4166 // Let Interpreter throw ClassCastException.
4167 PreserveJVMState pjvms(this);
4168 set_control(_gvn.transform(region));
4169 uncommon_trap(Deoptimization::Reason_intrinsic,
4170 Deoptimization::Action_maybe_recompile);
4171 }
4172 if (!stopped()) {
4173 set_result(res);
4174 }
4175 return true;
4176 }
4177
4178
4179 //--------------------------inline_native_subtype_check------------------------
4180 // This intrinsic takes the JNI calls out of the heart of
4181 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4182 bool LibraryCallKit::inline_native_subtype_check() {
4183 // Pull both arguments off the stack.
4184 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4185 args[0] = argument(0);
4186 args[1] = argument(1);
4187 Node* klasses[2]; // corresponding Klasses: superk, subk
4188 klasses[0] = klasses[1] = top();
4189
4190 enum {
4191 // A full decision tree on {superc is prim, subc is prim}:
4192 _prim_0_path = 1, // {P,N} => false
4193 // {P,P} & superc!=subc => false
4194 _prim_same_path, // {P,P} & superc==subc => true
4195 _prim_1_path, // {N,P} => false
4196 _ref_subtype_path, // {N,N} & subtype check wins => true
4197 _both_ref_path, // {N,N} & subtype check loses => false
4198 PATH_LIMIT
4199 };
4200
4201 RegionNode* region = new RegionNode(PATH_LIMIT);
4202 Node* phi = new PhiNode(region, TypeInt::BOOL);
4203 record_for_igvn(region);
4204
4205 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4206 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4207 int class_klass_offset = java_lang_Class::klass_offset();
4208
4209 // First null-check both mirrors and load each mirror's klass metaobject.
4210 int which_arg;
4211 for (which_arg = 0; which_arg <= 1; which_arg++) {
4212 Node* arg = args[which_arg];
4213 arg = null_check(arg);
4214 if (stopped()) break;
4215 args[which_arg] = arg;
4216
4217 Node* p = basic_plus_adr(arg, class_klass_offset);
4218 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4219 klasses[which_arg] = _gvn.transform(kls);
4220 }
4221
4222 // Having loaded both klasses, test each for null.
4223 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4224 for (which_arg = 0; which_arg <= 1; which_arg++) {
4225 Node* kls = klasses[which_arg];
4226 Node* null_ctl = top();
4227 kls = null_check_oop(kls, &null_ctl, never_see_null);
4228 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4229 region->init_req(prim_path, null_ctl);
4230 if (stopped()) break;
4231 klasses[which_arg] = kls;
4232 }
4233
4234 if (!stopped()) {
4235 // now we have two reference types, in klasses[0..1]
4236 Node* subk = klasses[1]; // the argument to isAssignableFrom
4237 Node* superk = klasses[0]; // the receiver
4238 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4239 // now we have a successful reference subtype check
4240 region->set_req(_ref_subtype_path, control());
4241 }
4242
4243 // If both operands are primitive (both klasses null), then
4244 // we must return true when they are identical primitives.
4245 // It is convenient to test this after the first null klass check.
4246 set_control(region->in(_prim_0_path)); // go back to first null check
4247 if (!stopped()) {
4248 // Since superc is primitive, make a guard for the superc==subc case.
4249 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4250 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4251 generate_guard(bol_eq, region, PROB_FAIR);
4252 if (region->req() == PATH_LIMIT+1) {
4253 // A guard was added. If the added guard is taken, superc==subc.
4254 region->swap_edges(PATH_LIMIT, _prim_same_path);
4255 region->del_req(PATH_LIMIT);
4256 }
4257 region->set_req(_prim_0_path, control()); // Not equal after all.
4258 }
4259
4260 // these are the only paths that produce 'true':
4261 phi->set_req(_prim_same_path, intcon(1));
4262 phi->set_req(_ref_subtype_path, intcon(1));
4263
4264 // pull together the cases:
4265 assert(region->req() == PATH_LIMIT, "sane region");
4266 for (uint i = 1; i < region->req(); i++) {
4267 Node* ctl = region->in(i);
4268 if (ctl == nullptr || ctl == top()) {
4269 region->set_req(i, top());
4270 phi ->set_req(i, top());
4271 } else if (phi->in(i) == nullptr) {
4272 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4273 }
4274 }
4275
4276 set_control(_gvn.transform(region));
4277 set_result(_gvn.transform(phi));
4278 return true;
4279 }
4280
4281 //---------------------generate_array_guard_common------------------------
4282 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4283 bool obj_array, bool not_array, Node** obj) {
4284
4285 if (stopped()) {
4286 return nullptr;
4287 }
4288
4289 // If obj_array/non_array==false/false:
4290 // Branch around if the given klass is in fact an array (either obj or prim).
4291 // If obj_array/non_array==false/true:
4292 // Branch around if the given klass is not an array klass of any kind.
4293 // If obj_array/non_array==true/true:
4294 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4295 // If obj_array/non_array==true/false:
4296 // Branch around if the kls is an oop array (Object[] or subtype)
4297 //
4298 // Like generate_guard, adds a new path onto the region.
4299 jint layout_con = 0;
4300 Node* layout_val = get_layout_helper(kls, layout_con);
4301 if (layout_val == nullptr) {
4302 bool query = (obj_array
4303 ? Klass::layout_helper_is_objArray(layout_con)
4304 : Klass::layout_helper_is_array(layout_con));
4305 if (query == not_array) {
4306 return nullptr; // never a branch
4307 } else { // always a branch
4308 Node* always_branch = control();
4309 if (region != nullptr)
4310 region->add_req(always_branch);
4311 set_control(top());
4312 return always_branch;
4313 }
4314 }
4315 // Now test the correct condition.
4316 jint nval = (obj_array
4317 ? (jint)(Klass::_lh_array_tag_type_value
4318 << Klass::_lh_array_tag_shift)
4319 : Klass::_lh_neutral_value);
4320 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4321 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4322 // invert the test if we are looking for a non-array
4323 if (not_array) btest = BoolTest(btest).negate();
4324 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4325 Node* ctrl = generate_fair_guard(bol, region);
4326 Node* is_array_ctrl = not_array ? control() : ctrl;
4327 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4328 // Keep track of the fact that 'obj' is an array to prevent
4329 // array specific accesses from floating above the guard.
4330 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4331 }
4332 return ctrl;
4333 }
4334
4335
4336 //-----------------------inline_native_newArray--------------------------
4337 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4338 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4339 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4340 Node* mirror;
4341 Node* count_val;
4342 if (uninitialized) {
4343 null_check_receiver();
4344 mirror = argument(1);
4345 count_val = argument(2);
4346 } else {
4347 mirror = argument(0);
4348 count_val = argument(1);
4349 }
4350
4351 mirror = null_check(mirror);
4352 // If mirror or obj is dead, only null-path is taken.
4353 if (stopped()) return true;
4354
4355 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4356 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4357 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4358 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4359 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4360
4361 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4362 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4363 result_reg, _slow_path);
4364 Node* normal_ctl = control();
4365 Node* no_array_ctl = result_reg->in(_slow_path);
4366
4367 // Generate code for the slow case. We make a call to newArray().
4368 set_control(no_array_ctl);
4369 if (!stopped()) {
4370 // Either the input type is void.class, or else the
4371 // array klass has not yet been cached. Either the
4372 // ensuing call will throw an exception, or else it
4373 // will cache the array klass for next time.
4374 PreserveJVMState pjvms(this);
4375 CallJavaNode* slow_call = nullptr;
4376 if (uninitialized) {
4377 // Generate optimized virtual call (holder class 'Unsafe' is final)
4378 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4379 } else {
4380 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4381 }
4382 Node* slow_result = set_results_for_java_call(slow_call);
4383 // this->control() comes from set_results_for_java_call
4384 result_reg->set_req(_slow_path, control());
4385 result_val->set_req(_slow_path, slow_result);
4386 result_io ->set_req(_slow_path, i_o());
4387 result_mem->set_req(_slow_path, reset_memory());
4388 }
4389
4390 set_control(normal_ctl);
4391 if (!stopped()) {
4392 // Normal case: The array type has been cached in the java.lang.Class.
4393 // The following call works fine even if the array type is polymorphic.
4394 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4395 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4396 result_reg->init_req(_normal_path, control());
4397 result_val->init_req(_normal_path, obj);
4398 result_io ->init_req(_normal_path, i_o());
4399 result_mem->init_req(_normal_path, reset_memory());
4400
4401 if (uninitialized) {
4402 // Mark the allocation so that zeroing is skipped
4403 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4404 alloc->maybe_set_complete(&_gvn);
4405 }
4406 }
4407
4408 // Return the combined state.
4409 set_i_o( _gvn.transform(result_io) );
4410 set_all_memory( _gvn.transform(result_mem));
4411
4412 C->set_has_split_ifs(true); // Has chance for split-if optimization
4413 set_result(result_reg, result_val);
4414 return true;
4415 }
4416
4417 //----------------------inline_native_getLength--------------------------
4418 // public static native int java.lang.reflect.Array.getLength(Object array);
4419 bool LibraryCallKit::inline_native_getLength() {
4420 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4421
4422 Node* array = null_check(argument(0));
4423 // If array is dead, only null-path is taken.
4424 if (stopped()) return true;
4425
4426 // Deoptimize if it is a non-array.
4427 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4428
4429 if (non_array != nullptr) {
4430 PreserveJVMState pjvms(this);
4431 set_control(non_array);
4432 uncommon_trap(Deoptimization::Reason_intrinsic,
4433 Deoptimization::Action_maybe_recompile);
4434 }
4435
4436 // If control is dead, only non-array-path is taken.
4437 if (stopped()) return true;
4438
4439 // The works fine even if the array type is polymorphic.
4440 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4441 Node* result = load_array_length(array);
4442
4443 C->set_has_split_ifs(true); // Has chance for split-if optimization
4444 set_result(result);
4445 return true;
4446 }
4447
4448 //------------------------inline_array_copyOf----------------------------
4449 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4450 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4451 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4452 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4453
4454 // Get the arguments.
4455 Node* original = argument(0);
4456 Node* start = is_copyOfRange? argument(1): intcon(0);
4457 Node* end = is_copyOfRange? argument(2): argument(1);
4458 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4459
4460 Node* newcopy = nullptr;
4461
4462 // Set the original stack and the reexecute bit for the interpreter to reexecute
4463 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4464 { PreserveReexecuteState preexecs(this);
4465 jvms()->set_should_reexecute(true);
4466
4467 array_type_mirror = null_check(array_type_mirror);
4468 original = null_check(original);
4469
4470 // Check if a null path was taken unconditionally.
4471 if (stopped()) return true;
4472
4473 Node* orig_length = load_array_length(original);
4474
4475 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4476 klass_node = null_check(klass_node);
4477
4478 RegionNode* bailout = new RegionNode(1);
4479 record_for_igvn(bailout);
4480
4481 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4482 // Bail out if that is so.
4483 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4484 if (not_objArray != nullptr) {
4485 // Improve the klass node's type from the new optimistic assumption:
4486 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4487 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4488 Node* cast = new CastPPNode(control(), klass_node, akls);
4489 klass_node = _gvn.transform(cast);
4490 }
4491
4492 // Bail out if either start or end is negative.
4493 generate_negative_guard(start, bailout, &start);
4494 generate_negative_guard(end, bailout, &end);
4495
4496 Node* length = end;
4497 if (_gvn.type(start) != TypeInt::ZERO) {
4498 length = _gvn.transform(new SubINode(end, start));
4499 }
4500
4501 // Bail out if length is negative (i.e., if start > end).
4502 // Without this the new_array would throw
4503 // NegativeArraySizeException but IllegalArgumentException is what
4504 // should be thrown
4505 generate_negative_guard(length, bailout, &length);
4506
4507 // Bail out if start is larger than the original length
4508 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4509 generate_negative_guard(orig_tail, bailout, &orig_tail);
4510
4511 if (bailout->req() > 1) {
4512 PreserveJVMState pjvms(this);
4513 set_control(_gvn.transform(bailout));
4514 uncommon_trap(Deoptimization::Reason_intrinsic,
4515 Deoptimization::Action_maybe_recompile);
4516 }
4517
4518 if (!stopped()) {
4519 // How many elements will we copy from the original?
4520 // The answer is MinI(orig_tail, length).
4521 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4522
4523 // Generate a direct call to the right arraycopy function(s).
4524 // We know the copy is disjoint but we might not know if the
4525 // oop stores need checking.
4526 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4527 // This will fail a store-check if x contains any non-nulls.
4528
4529 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4530 // loads/stores but it is legal only if we're sure the
4531 // Arrays.copyOf would succeed. So we need all input arguments
4532 // to the copyOf to be validated, including that the copy to the
4533 // new array won't trigger an ArrayStoreException. That subtype
4534 // check can be optimized if we know something on the type of
4535 // the input array from type speculation.
4536 if (_gvn.type(klass_node)->singleton()) {
4537 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4538 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4539
4540 int test = C->static_subtype_check(superk, subk);
4541 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4542 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4543 if (t_original->speculative_type() != nullptr) {
4544 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4545 }
4546 }
4547 }
4548
4549 bool validated = false;
4550 // Reason_class_check rather than Reason_intrinsic because we
4551 // want to intrinsify even if this traps.
4552 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4553 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4554
4555 if (not_subtype_ctrl != top()) {
4556 PreserveJVMState pjvms(this);
4557 set_control(not_subtype_ctrl);
4558 uncommon_trap(Deoptimization::Reason_class_check,
4559 Deoptimization::Action_make_not_entrant);
4560 assert(stopped(), "Should be stopped");
4561 }
4562 validated = true;
4563 }
4564
4565 if (!stopped()) {
4566 newcopy = new_array(klass_node, length, 0); // no arguments to push
4567
4568 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4569 load_object_klass(original), klass_node);
4570 if (!is_copyOfRange) {
4571 ac->set_copyof(validated);
4572 } else {
4573 ac->set_copyofrange(validated);
4574 }
4575 Node* n = _gvn.transform(ac);
4576 if (n == ac) {
4577 ac->connect_outputs(this);
4578 } else {
4579 assert(validated, "shouldn't transform if all arguments not validated");
4580 set_all_memory(n);
4581 }
4582 }
4583 }
4584 } // original reexecute is set back here
4585
4586 C->set_has_split_ifs(true); // Has chance for split-if optimization
4587 if (!stopped()) {
4588 set_result(newcopy);
4589 }
4590 return true;
4591 }
4592
4593
4594 //----------------------generate_virtual_guard---------------------------
4595 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4596 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4597 RegionNode* slow_region) {
4598 ciMethod* method = callee();
4599 int vtable_index = method->vtable_index();
4600 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4601 "bad index %d", vtable_index);
4602 // Get the Method* out of the appropriate vtable entry.
4603 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4604 vtable_index*vtableEntry::size_in_bytes() +
4605 in_bytes(vtableEntry::method_offset());
4606 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4607 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4608
4609 // Compare the target method with the expected method (e.g., Object.hashCode).
4610 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4611
4612 Node* native_call = makecon(native_call_addr);
4613 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4614 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4615
4616 return generate_slow_guard(test_native, slow_region);
4617 }
4618
4619 //-----------------------generate_method_call----------------------------
4620 // Use generate_method_call to make a slow-call to the real
4621 // method if the fast path fails. An alternative would be to
4622 // use a stub like OptoRuntime::slow_arraycopy_Java.
4623 // This only works for expanding the current library call,
4624 // not another intrinsic. (E.g., don't use this for making an
4625 // arraycopy call inside of the copyOf intrinsic.)
4626 CallJavaNode*
4627 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4628 // When compiling the intrinsic method itself, do not use this technique.
4629 guarantee(callee() != C->method(), "cannot make slow-call to self");
4630
4631 ciMethod* method = callee();
4632 // ensure the JVMS we have will be correct for this call
4633 guarantee(method_id == method->intrinsic_id(), "must match");
4634
4635 const TypeFunc* tf = TypeFunc::make(method);
4636 if (res_not_null) {
4637 assert(tf->return_type() == T_OBJECT, "");
4638 const TypeTuple* range = tf->range();
4639 const Type** fields = TypeTuple::fields(range->cnt());
4640 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4641 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4642 tf = TypeFunc::make(tf->domain(), new_range);
4643 }
4644 CallJavaNode* slow_call;
4645 if (is_static) {
4646 assert(!is_virtual, "");
4647 slow_call = new CallStaticJavaNode(C, tf,
4648 SharedRuntime::get_resolve_static_call_stub(), method);
4649 } else if (is_virtual) {
4650 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4651 int vtable_index = Method::invalid_vtable_index;
4652 if (UseInlineCaches) {
4653 // Suppress the vtable call
4654 } else {
4655 // hashCode and clone are not a miranda methods,
4656 // so the vtable index is fixed.
4657 // No need to use the linkResolver to get it.
4658 vtable_index = method->vtable_index();
4659 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4660 "bad index %d", vtable_index);
4661 }
4662 slow_call = new CallDynamicJavaNode(tf,
4663 SharedRuntime::get_resolve_virtual_call_stub(),
4664 method, vtable_index);
4665 } else { // neither virtual nor static: opt_virtual
4666 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4667 slow_call = new CallStaticJavaNode(C, tf,
4668 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4669 slow_call->set_optimized_virtual(true);
4670 }
4671 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4672 // To be able to issue a direct call (optimized virtual or virtual)
4673 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4674 // about the method being invoked should be attached to the call site to
4675 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4676 slow_call->set_override_symbolic_info(true);
4677 }
4678 set_arguments_for_java_call(slow_call);
4679 set_edges_for_java_call(slow_call);
4680 return slow_call;
4681 }
4682
4683
4684 /**
4685 * Build special case code for calls to hashCode on an object. This call may
4686 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4687 * slightly different code.
4688 */
4689 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4690 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4691 assert(!(is_virtual && is_static), "either virtual, special, or static");
4692
4693 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4694
4695 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4696 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4697 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4698 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4699 Node* obj = nullptr;
4700 if (!is_static) {
4701 // Check for hashing null object
4702 obj = null_check_receiver();
4703 if (stopped()) return true; // unconditionally null
4704 result_reg->init_req(_null_path, top());
4705 result_val->init_req(_null_path, top());
4706 } else {
4707 // Do a null check, and return zero if null.
4708 // System.identityHashCode(null) == 0
4709 obj = argument(0);
4710 Node* null_ctl = top();
4711 obj = null_check_oop(obj, &null_ctl);
4712 result_reg->init_req(_null_path, null_ctl);
4713 result_val->init_req(_null_path, _gvn.intcon(0));
4714 }
4715
4716 // Unconditionally null? Then return right away.
4717 if (stopped()) {
4718 set_control( result_reg->in(_null_path));
4719 if (!stopped())
4720 set_result(result_val->in(_null_path));
4721 return true;
4722 }
4723
4724 // We only go to the fast case code if we pass a number of guards. The
4725 // paths which do not pass are accumulated in the slow_region.
4726 RegionNode* slow_region = new RegionNode(1);
4727 record_for_igvn(slow_region);
4728
4729 // If this is a virtual call, we generate a funny guard. We pull out
4730 // the vtable entry corresponding to hashCode() from the target object.
4731 // If the target method which we are calling happens to be the native
4732 // Object hashCode() method, we pass the guard. We do not need this
4733 // guard for non-virtual calls -- the caller is known to be the native
4734 // Object hashCode().
4735 if (is_virtual) {
4736 // After null check, get the object's klass.
4737 Node* obj_klass = load_object_klass(obj);
4738 generate_virtual_guard(obj_klass, slow_region);
4739 }
4740
4741 // Get the header out of the object, use LoadMarkNode when available
4742 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4743 // The control of the load must be null. Otherwise, the load can move before
4744 // the null check after castPP removal.
4745 Node* no_ctrl = nullptr;
4746 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4747
4748 if (!UseObjectMonitorTable) {
4749 // Test the header to see if it is safe to read w.r.t. locking.
4750 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4751 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4752 if (LockingMode == LM_LIGHTWEIGHT) {
4753 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4754 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4755 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4756
4757 generate_slow_guard(test_monitor, slow_region);
4758 } else {
4759 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4760 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4761 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4762
4763 generate_slow_guard(test_not_unlocked, slow_region);
4764 }
4765 }
4766
4767 // Get the hash value and check to see that it has been properly assigned.
4768 // We depend on hash_mask being at most 32 bits and avoid the use of
4769 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4770 // vm: see markWord.hpp.
4771 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4772 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4773 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4774 // This hack lets the hash bits live anywhere in the mark object now, as long
4775 // as the shift drops the relevant bits into the low 32 bits. Note that
4776 // Java spec says that HashCode is an int so there's no point in capturing
4777 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4778 hshifted_header = ConvX2I(hshifted_header);
4779 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4780
4781 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4782 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4783 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4784
4785 generate_slow_guard(test_assigned, slow_region);
4786
4787 Node* init_mem = reset_memory();
4788 // fill in the rest of the null path:
4789 result_io ->init_req(_null_path, i_o());
4790 result_mem->init_req(_null_path, init_mem);
4791
4792 result_val->init_req(_fast_path, hash_val);
4793 result_reg->init_req(_fast_path, control());
4794 result_io ->init_req(_fast_path, i_o());
4795 result_mem->init_req(_fast_path, init_mem);
4796
4797 // Generate code for the slow case. We make a call to hashCode().
4798 set_control(_gvn.transform(slow_region));
4799 if (!stopped()) {
4800 // No need for PreserveJVMState, because we're using up the present state.
4801 set_all_memory(init_mem);
4802 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4803 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4804 Node* slow_result = set_results_for_java_call(slow_call);
4805 // this->control() comes from set_results_for_java_call
4806 result_reg->init_req(_slow_path, control());
4807 result_val->init_req(_slow_path, slow_result);
4808 result_io ->set_req(_slow_path, i_o());
4809 result_mem ->set_req(_slow_path, reset_memory());
4810 }
4811
4812 // Return the combined state.
4813 set_i_o( _gvn.transform(result_io) );
4814 set_all_memory( _gvn.transform(result_mem));
4815
4816 set_result(result_reg, result_val);
4817 return true;
4818 }
4819
4820 //---------------------------inline_native_getClass----------------------------
4821 // public final native Class<?> java.lang.Object.getClass();
4822 //
4823 // Build special case code for calls to getClass on an object.
4824 bool LibraryCallKit::inline_native_getClass() {
4825 Node* obj = null_check_receiver();
4826 if (stopped()) return true;
4827 set_result(load_mirror_from_klass(load_object_klass(obj)));
4828 return true;
4829 }
4830
4831 //-----------------inline_native_Reflection_getCallerClass---------------------
4832 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4833 //
4834 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4835 //
4836 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4837 // in that it must skip particular security frames and checks for
4838 // caller sensitive methods.
4839 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4840 #ifndef PRODUCT
4841 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4842 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4843 }
4844 #endif
4845
4846 if (!jvms()->has_method()) {
4847 #ifndef PRODUCT
4848 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4849 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4850 }
4851 #endif
4852 return false;
4853 }
4854
4855 // Walk back up the JVM state to find the caller at the required
4856 // depth.
4857 JVMState* caller_jvms = jvms();
4858
4859 // Cf. JVM_GetCallerClass
4860 // NOTE: Start the loop at depth 1 because the current JVM state does
4861 // not include the Reflection.getCallerClass() frame.
4862 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4863 ciMethod* m = caller_jvms->method();
4864 switch (n) {
4865 case 0:
4866 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4867 break;
4868 case 1:
4869 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4870 if (!m->caller_sensitive()) {
4871 #ifndef PRODUCT
4872 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4873 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4874 }
4875 #endif
4876 return false; // bail-out; let JVM_GetCallerClass do the work
4877 }
4878 break;
4879 default:
4880 if (!m->is_ignored_by_security_stack_walk()) {
4881 // We have reached the desired frame; return the holder class.
4882 // Acquire method holder as java.lang.Class and push as constant.
4883 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4884 ciInstance* caller_mirror = caller_klass->java_mirror();
4885 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4886
4887 #ifndef PRODUCT
4888 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4889 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());
4890 tty->print_cr(" JVM state at this point:");
4891 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4892 ciMethod* m = jvms()->of_depth(i)->method();
4893 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4894 }
4895 }
4896 #endif
4897 return true;
4898 }
4899 break;
4900 }
4901 }
4902
4903 #ifndef PRODUCT
4904 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4905 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4906 tty->print_cr(" JVM state at this point:");
4907 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4908 ciMethod* m = jvms()->of_depth(i)->method();
4909 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4910 }
4911 }
4912 #endif
4913
4914 return false; // bail-out; let JVM_GetCallerClass do the work
4915 }
4916
4917 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4918 Node* arg = argument(0);
4919 Node* result = nullptr;
4920
4921 switch (id) {
4922 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4923 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4924 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4925 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4926 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4927 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4928
4929 case vmIntrinsics::_doubleToLongBits: {
4930 // two paths (plus control) merge in a wood
4931 RegionNode *r = new RegionNode(3);
4932 Node *phi = new PhiNode(r, TypeLong::LONG);
4933
4934 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4935 // Build the boolean node
4936 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4937
4938 // Branch either way.
4939 // NaN case is less traveled, which makes all the difference.
4940 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4941 Node *opt_isnan = _gvn.transform(ifisnan);
4942 assert( opt_isnan->is_If(), "Expect an IfNode");
4943 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4944 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4945
4946 set_control(iftrue);
4947
4948 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4949 Node *slow_result = longcon(nan_bits); // return NaN
4950 phi->init_req(1, _gvn.transform( slow_result ));
4951 r->init_req(1, iftrue);
4952
4953 // Else fall through
4954 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4955 set_control(iffalse);
4956
4957 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4958 r->init_req(2, iffalse);
4959
4960 // Post merge
4961 set_control(_gvn.transform(r));
4962 record_for_igvn(r);
4963
4964 C->set_has_split_ifs(true); // Has chance for split-if optimization
4965 result = phi;
4966 assert(result->bottom_type()->isa_long(), "must be");
4967 break;
4968 }
4969
4970 case vmIntrinsics::_floatToIntBits: {
4971 // two paths (plus control) merge in a wood
4972 RegionNode *r = new RegionNode(3);
4973 Node *phi = new PhiNode(r, TypeInt::INT);
4974
4975 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4976 // Build the boolean node
4977 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4978
4979 // Branch either way.
4980 // NaN case is less traveled, which makes all the difference.
4981 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4982 Node *opt_isnan = _gvn.transform(ifisnan);
4983 assert( opt_isnan->is_If(), "Expect an IfNode");
4984 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4985 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4986
4987 set_control(iftrue);
4988
4989 static const jint nan_bits = 0x7fc00000;
4990 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4991 phi->init_req(1, _gvn.transform( slow_result ));
4992 r->init_req(1, iftrue);
4993
4994 // Else fall through
4995 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4996 set_control(iffalse);
4997
4998 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4999 r->init_req(2, iffalse);
5000
5001 // Post merge
5002 set_control(_gvn.transform(r));
5003 record_for_igvn(r);
5004
5005 C->set_has_split_ifs(true); // Has chance for split-if optimization
5006 result = phi;
5007 assert(result->bottom_type()->isa_int(), "must be");
5008 break;
5009 }
5010
5011 default:
5012 fatal_unexpected_iid(id);
5013 break;
5014 }
5015 set_result(_gvn.transform(result));
5016 return true;
5017 }
5018
5019 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5020 Node* arg = argument(0);
5021 Node* result = nullptr;
5022
5023 switch (id) {
5024 case vmIntrinsics::_floatIsInfinite:
5025 result = new IsInfiniteFNode(arg);
5026 break;
5027 case vmIntrinsics::_floatIsFinite:
5028 result = new IsFiniteFNode(arg);
5029 break;
5030 case vmIntrinsics::_doubleIsInfinite:
5031 result = new IsInfiniteDNode(arg);
5032 break;
5033 case vmIntrinsics::_doubleIsFinite:
5034 result = new IsFiniteDNode(arg);
5035 break;
5036 default:
5037 fatal_unexpected_iid(id);
5038 break;
5039 }
5040 set_result(_gvn.transform(result));
5041 return true;
5042 }
5043
5044 //----------------------inline_unsafe_copyMemory-------------------------
5045 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5046
5047 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5048 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5049 const Type* base_t = gvn.type(base);
5050
5051 bool in_native = (base_t == TypePtr::NULL_PTR);
5052 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5053 bool is_mixed = !in_heap && !in_native;
5054
5055 if (is_mixed) {
5056 return true; // mixed accesses can touch both on-heap and off-heap memory
5057 }
5058 if (in_heap) {
5059 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5060 if (!is_prim_array) {
5061 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5062 // there's not enough type information available to determine proper memory slice for it.
5063 return true;
5064 }
5065 }
5066 return false;
5067 }
5068
5069 bool LibraryCallKit::inline_unsafe_copyMemory() {
5070 if (callee()->is_static()) return false; // caller must have the capability!
5071 null_check_receiver(); // null-check receiver
5072 if (stopped()) return true;
5073
5074 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5075
5076 Node* src_base = argument(1); // type: oop
5077 Node* src_off = ConvL2X(argument(2)); // type: long
5078 Node* dst_base = argument(4); // type: oop
5079 Node* dst_off = ConvL2X(argument(5)); // type: long
5080 Node* size = ConvL2X(argument(7)); // type: long
5081
5082 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5083 "fieldOffset must be byte-scaled");
5084
5085 Node* src_addr = make_unsafe_address(src_base, src_off);
5086 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5087
5088 Node* thread = _gvn.transform(new ThreadLocalNode());
5089 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5090 BasicType doing_unsafe_access_bt = T_BYTE;
5091 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5092
5093 // update volatile field
5094 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5095
5096 int flags = RC_LEAF | RC_NO_FP;
5097
5098 const TypePtr* dst_type = TypePtr::BOTTOM;
5099
5100 // Adjust memory effects of the runtime call based on input values.
5101 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5102 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5103 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5104
5105 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5106 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5107 flags |= RC_NARROW_MEM; // narrow in memory
5108 }
5109 }
5110
5111 // Call it. Note that the length argument is not scaled.
5112 make_runtime_call(flags,
5113 OptoRuntime::fast_arraycopy_Type(),
5114 StubRoutines::unsafe_arraycopy(),
5115 "unsafe_arraycopy",
5116 dst_type,
5117 src_addr, dst_addr, size XTOP);
5118
5119 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5120
5121 return true;
5122 }
5123
5124 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5125 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5126 bool LibraryCallKit::inline_unsafe_setMemory() {
5127 if (callee()->is_static()) return false; // caller must have the capability!
5128 null_check_receiver(); // null-check receiver
5129 if (stopped()) return true;
5130
5131 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5132
5133 Node* dst_base = argument(1); // type: oop
5134 Node* dst_off = ConvL2X(argument(2)); // type: long
5135 Node* size = ConvL2X(argument(4)); // type: long
5136 Node* byte = argument(6); // type: byte
5137
5138 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5139 "fieldOffset must be byte-scaled");
5140
5141 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5142
5143 Node* thread = _gvn.transform(new ThreadLocalNode());
5144 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5145 BasicType doing_unsafe_access_bt = T_BYTE;
5146 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5147
5148 // update volatile field
5149 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5150
5151 int flags = RC_LEAF | RC_NO_FP;
5152
5153 const TypePtr* dst_type = TypePtr::BOTTOM;
5154
5155 // Adjust memory effects of the runtime call based on input values.
5156 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5157 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5158
5159 flags |= RC_NARROW_MEM; // narrow in memory
5160 }
5161
5162 // Call it. Note that the length argument is not scaled.
5163 make_runtime_call(flags,
5164 OptoRuntime::unsafe_setmemory_Type(),
5165 StubRoutines::unsafe_setmemory(),
5166 "unsafe_setmemory",
5167 dst_type,
5168 dst_addr, size XTOP, byte);
5169
5170 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5171
5172 return true;
5173 }
5174
5175 #undef XTOP
5176
5177 //------------------------clone_coping-----------------------------------
5178 // Helper function for inline_native_clone.
5179 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5180 assert(obj_size != nullptr, "");
5181 Node* raw_obj = alloc_obj->in(1);
5182 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5183
5184 AllocateNode* alloc = nullptr;
5185 if (ReduceBulkZeroing &&
5186 // If we are implementing an array clone without knowing its source type
5187 // (can happen when compiling the array-guarded branch of a reflective
5188 // Object.clone() invocation), initialize the array within the allocation.
5189 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5190 // to a runtime clone call that assumes fully initialized source arrays.
5191 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5192 // We will be completely responsible for initializing this object -
5193 // mark Initialize node as complete.
5194 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5195 // The object was just allocated - there should be no any stores!
5196 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5197 // Mark as complete_with_arraycopy so that on AllocateNode
5198 // expansion, we know this AllocateNode is initialized by an array
5199 // copy and a StoreStore barrier exists after the array copy.
5200 alloc->initialization()->set_complete_with_arraycopy();
5201 }
5202
5203 Node* size = _gvn.transform(obj_size);
5204 access_clone(obj, alloc_obj, size, is_array);
5205
5206 // Do not let reads from the cloned object float above the arraycopy.
5207 if (alloc != nullptr) {
5208 // Do not let stores that initialize this object be reordered with
5209 // a subsequent store that would make this object accessible by
5210 // other threads.
5211 // Record what AllocateNode this StoreStore protects so that
5212 // escape analysis can go from the MemBarStoreStoreNode to the
5213 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5214 // based on the escape status of the AllocateNode.
5215 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5216 } else {
5217 insert_mem_bar(Op_MemBarCPUOrder);
5218 }
5219 }
5220
5221 //------------------------inline_native_clone----------------------------
5222 // protected native Object java.lang.Object.clone();
5223 //
5224 // Here are the simple edge cases:
5225 // null receiver => normal trap
5226 // virtual and clone was overridden => slow path to out-of-line clone
5227 // not cloneable or finalizer => slow path to out-of-line Object.clone
5228 //
5229 // The general case has two steps, allocation and copying.
5230 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5231 //
5232 // Copying also has two cases, oop arrays and everything else.
5233 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5234 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5235 //
5236 // These steps fold up nicely if and when the cloned object's klass
5237 // can be sharply typed as an object array, a type array, or an instance.
5238 //
5239 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5240 PhiNode* result_val;
5241
5242 // Set the reexecute bit for the interpreter to reexecute
5243 // the bytecode that invokes Object.clone if deoptimization happens.
5244 { PreserveReexecuteState preexecs(this);
5245 jvms()->set_should_reexecute(true);
5246
5247 Node* obj = null_check_receiver();
5248 if (stopped()) return true;
5249
5250 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5251
5252 // If we are going to clone an instance, we need its exact type to
5253 // know the number and types of fields to convert the clone to
5254 // loads/stores. Maybe a speculative type can help us.
5255 if (!obj_type->klass_is_exact() &&
5256 obj_type->speculative_type() != nullptr &&
5257 obj_type->speculative_type()->is_instance_klass()) {
5258 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5259 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5260 !spec_ik->has_injected_fields()) {
5261 if (!obj_type->isa_instptr() ||
5262 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5263 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5264 }
5265 }
5266 }
5267
5268 // Conservatively insert a memory barrier on all memory slices.
5269 // Do not let writes into the original float below the clone.
5270 insert_mem_bar(Op_MemBarCPUOrder);
5271
5272 // paths into result_reg:
5273 enum {
5274 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5275 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5276 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5277 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5278 PATH_LIMIT
5279 };
5280 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5281 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5282 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5283 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5284 record_for_igvn(result_reg);
5285
5286 Node* obj_klass = load_object_klass(obj);
5287 Node* array_obj = obj;
5288 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5289 if (array_ctl != nullptr) {
5290 // It's an array.
5291 PreserveJVMState pjvms(this);
5292 set_control(array_ctl);
5293 Node* obj_length = load_array_length(array_obj);
5294 Node* array_size = nullptr; // Size of the array without object alignment padding.
5295 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5296
5297 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5298 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5299 // If it is an oop array, it requires very special treatment,
5300 // because gc barriers are required when accessing the array.
5301 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5302 if (is_obja != nullptr) {
5303 PreserveJVMState pjvms2(this);
5304 set_control(is_obja);
5305 // Generate a direct call to the right arraycopy function(s).
5306 // Clones are always tightly coupled.
5307 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5308 ac->set_clone_oop_array();
5309 Node* n = _gvn.transform(ac);
5310 assert(n == ac, "cannot disappear");
5311 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5312
5313 result_reg->init_req(_objArray_path, control());
5314 result_val->init_req(_objArray_path, alloc_obj);
5315 result_i_o ->set_req(_objArray_path, i_o());
5316 result_mem ->set_req(_objArray_path, reset_memory());
5317 }
5318 }
5319 // Otherwise, there are no barriers to worry about.
5320 // (We can dispense with card marks if we know the allocation
5321 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5322 // causes the non-eden paths to take compensating steps to
5323 // simulate a fresh allocation, so that no further
5324 // card marks are required in compiled code to initialize
5325 // the object.)
5326
5327 if (!stopped()) {
5328 copy_to_clone(array_obj, alloc_obj, array_size, true);
5329
5330 // Present the results of the copy.
5331 result_reg->init_req(_array_path, control());
5332 result_val->init_req(_array_path, alloc_obj);
5333 result_i_o ->set_req(_array_path, i_o());
5334 result_mem ->set_req(_array_path, reset_memory());
5335 }
5336 }
5337
5338 // We only go to the instance fast case code if we pass a number of guards.
5339 // The paths which do not pass are accumulated in the slow_region.
5340 RegionNode* slow_region = new RegionNode(1);
5341 record_for_igvn(slow_region);
5342 if (!stopped()) {
5343 // It's an instance (we did array above). Make the slow-path tests.
5344 // If this is a virtual call, we generate a funny guard. We grab
5345 // the vtable entry corresponding to clone() from the target object.
5346 // If the target method which we are calling happens to be the
5347 // Object clone() method, we pass the guard. We do not need this
5348 // guard for non-virtual calls; the caller is known to be the native
5349 // Object clone().
5350 if (is_virtual) {
5351 generate_virtual_guard(obj_klass, slow_region);
5352 }
5353
5354 // The object must be easily cloneable and must not have a finalizer.
5355 // Both of these conditions may be checked in a single test.
5356 // We could optimize the test further, but we don't care.
5357 generate_misc_flags_guard(obj_klass,
5358 // Test both conditions:
5359 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5360 // Must be cloneable but not finalizer:
5361 KlassFlags::_misc_is_cloneable_fast,
5362 slow_region);
5363 }
5364
5365 if (!stopped()) {
5366 // It's an instance, and it passed the slow-path tests.
5367 PreserveJVMState pjvms(this);
5368 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5369 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5370 // is reexecuted if deoptimization occurs and there could be problems when merging
5371 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5372 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5373
5374 copy_to_clone(obj, alloc_obj, obj_size, false);
5375
5376 // Present the results of the slow call.
5377 result_reg->init_req(_instance_path, control());
5378 result_val->init_req(_instance_path, alloc_obj);
5379 result_i_o ->set_req(_instance_path, i_o());
5380 result_mem ->set_req(_instance_path, reset_memory());
5381 }
5382
5383 // Generate code for the slow case. We make a call to clone().
5384 set_control(_gvn.transform(slow_region));
5385 if (!stopped()) {
5386 PreserveJVMState pjvms(this);
5387 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5388 // We need to deoptimize on exception (see comment above)
5389 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5390 // this->control() comes from set_results_for_java_call
5391 result_reg->init_req(_slow_path, control());
5392 result_val->init_req(_slow_path, slow_result);
5393 result_i_o ->set_req(_slow_path, i_o());
5394 result_mem ->set_req(_slow_path, reset_memory());
5395 }
5396
5397 // Return the combined state.
5398 set_control( _gvn.transform(result_reg));
5399 set_i_o( _gvn.transform(result_i_o));
5400 set_all_memory( _gvn.transform(result_mem));
5401 } // original reexecute is set back here
5402
5403 set_result(_gvn.transform(result_val));
5404 return true;
5405 }
5406
5407 // If we have a tightly coupled allocation, the arraycopy may take care
5408 // of the array initialization. If one of the guards we insert between
5409 // the allocation and the arraycopy causes a deoptimization, an
5410 // uninitialized array will escape the compiled method. To prevent that
5411 // we set the JVM state for uncommon traps between the allocation and
5412 // the arraycopy to the state before the allocation so, in case of
5413 // deoptimization, we'll reexecute the allocation and the
5414 // initialization.
5415 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5416 if (alloc != nullptr) {
5417 ciMethod* trap_method = alloc->jvms()->method();
5418 int trap_bci = alloc->jvms()->bci();
5419
5420 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5421 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5422 // Make sure there's no store between the allocation and the
5423 // arraycopy otherwise visible side effects could be rexecuted
5424 // in case of deoptimization and cause incorrect execution.
5425 bool no_interfering_store = true;
5426 Node* mem = alloc->in(TypeFunc::Memory);
5427 if (mem->is_MergeMem()) {
5428 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5429 Node* n = mms.memory();
5430 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5431 assert(n->is_Store(), "what else?");
5432 no_interfering_store = false;
5433 break;
5434 }
5435 }
5436 } else {
5437 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5438 Node* n = mms.memory();
5439 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5440 assert(n->is_Store(), "what else?");
5441 no_interfering_store = false;
5442 break;
5443 }
5444 }
5445 }
5446
5447 if (no_interfering_store) {
5448 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5449
5450 JVMState* saved_jvms = jvms();
5451 saved_reexecute_sp = _reexecute_sp;
5452
5453 set_jvms(sfpt->jvms());
5454 _reexecute_sp = jvms()->sp();
5455
5456 return saved_jvms;
5457 }
5458 }
5459 }
5460 return nullptr;
5461 }
5462
5463 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5464 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5465 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5466 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5467 uint size = alloc->req();
5468 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5469 old_jvms->set_map(sfpt);
5470 for (uint i = 0; i < size; i++) {
5471 sfpt->init_req(i, alloc->in(i));
5472 }
5473 // re-push array length for deoptimization
5474 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5475 old_jvms->set_sp(old_jvms->sp()+1);
5476 old_jvms->set_monoff(old_jvms->monoff()+1);
5477 old_jvms->set_scloff(old_jvms->scloff()+1);
5478 old_jvms->set_endoff(old_jvms->endoff()+1);
5479 old_jvms->set_should_reexecute(true);
5480
5481 sfpt->set_i_o(map()->i_o());
5482 sfpt->set_memory(map()->memory());
5483 sfpt->set_control(map()->control());
5484 return sfpt;
5485 }
5486
5487 // In case of a deoptimization, we restart execution at the
5488 // allocation, allocating a new array. We would leave an uninitialized
5489 // array in the heap that GCs wouldn't expect. Move the allocation
5490 // after the traps so we don't allocate the array if we
5491 // deoptimize. This is possible because tightly_coupled_allocation()
5492 // guarantees there's no observer of the allocated array at this point
5493 // and the control flow is simple enough.
5494 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5495 int saved_reexecute_sp, uint new_idx) {
5496 if (saved_jvms_before_guards != nullptr && !stopped()) {
5497 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5498
5499 assert(alloc != nullptr, "only with a tightly coupled allocation");
5500 // restore JVM state to the state at the arraycopy
5501 saved_jvms_before_guards->map()->set_control(map()->control());
5502 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5503 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5504 // If we've improved the types of some nodes (null check) while
5505 // emitting the guards, propagate them to the current state
5506 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5507 set_jvms(saved_jvms_before_guards);
5508 _reexecute_sp = saved_reexecute_sp;
5509
5510 // Remove the allocation from above the guards
5511 CallProjections callprojs;
5512 alloc->extract_projections(&callprojs, true);
5513 InitializeNode* init = alloc->initialization();
5514 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5515 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5516 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5517
5518 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5519 // the allocation (i.e. is only valid if the allocation succeeds):
5520 // 1) replace CastIINode with AllocateArrayNode's length here
5521 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5522 //
5523 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5524 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5525 Node* init_control = init->proj_out(TypeFunc::Control);
5526 Node* alloc_length = alloc->Ideal_length();
5527 #ifdef ASSERT
5528 Node* prev_cast = nullptr;
5529 #endif
5530 for (uint i = 0; i < init_control->outcnt(); i++) {
5531 Node* init_out = init_control->raw_out(i);
5532 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5533 #ifdef ASSERT
5534 if (prev_cast == nullptr) {
5535 prev_cast = init_out;
5536 } else {
5537 if (prev_cast->cmp(*init_out) == false) {
5538 prev_cast->dump();
5539 init_out->dump();
5540 assert(false, "not equal CastIINode");
5541 }
5542 }
5543 #endif
5544 C->gvn_replace_by(init_out, alloc_length);
5545 }
5546 }
5547 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5548
5549 // move the allocation here (after the guards)
5550 _gvn.hash_delete(alloc);
5551 alloc->set_req(TypeFunc::Control, control());
5552 alloc->set_req(TypeFunc::I_O, i_o());
5553 Node *mem = reset_memory();
5554 set_all_memory(mem);
5555 alloc->set_req(TypeFunc::Memory, mem);
5556 set_control(init->proj_out_or_null(TypeFunc::Control));
5557 set_i_o(callprojs.fallthrough_ioproj);
5558
5559 // Update memory as done in GraphKit::set_output_for_allocation()
5560 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5561 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5562 if (ary_type->isa_aryptr() && length_type != nullptr) {
5563 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5564 }
5565 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5566 int elemidx = C->get_alias_index(telemref);
5567 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5568 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5569
5570 Node* allocx = _gvn.transform(alloc);
5571 assert(allocx == alloc, "where has the allocation gone?");
5572 assert(dest->is_CheckCastPP(), "not an allocation result?");
5573
5574 _gvn.hash_delete(dest);
5575 dest->set_req(0, control());
5576 Node* destx = _gvn.transform(dest);
5577 assert(destx == dest, "where has the allocation result gone?");
5578
5579 array_ideal_length(alloc, ary_type, true);
5580 }
5581 }
5582
5583 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5584 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5585 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5586 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5587 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5588 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5589 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5590 JVMState* saved_jvms_before_guards) {
5591 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5592 // There is at least one unrelated uncommon trap which needs to be replaced.
5593 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5594
5595 JVMState* saved_jvms = jvms();
5596 const int saved_reexecute_sp = _reexecute_sp;
5597 set_jvms(sfpt->jvms());
5598 _reexecute_sp = jvms()->sp();
5599
5600 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5601
5602 // Restore state
5603 set_jvms(saved_jvms);
5604 _reexecute_sp = saved_reexecute_sp;
5605 }
5606 }
5607
5608 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5609 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5610 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5611 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5612 while (if_proj->is_IfProj()) {
5613 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5614 if (uncommon_trap != nullptr) {
5615 create_new_uncommon_trap(uncommon_trap);
5616 }
5617 assert(if_proj->in(0)->is_If(), "must be If");
5618 if_proj = if_proj->in(0)->in(0);
5619 }
5620 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5621 "must have reached control projection of init node");
5622 }
5623
5624 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5625 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5626 assert(trap_request != 0, "no valid UCT trap request");
5627 PreserveJVMState pjvms(this);
5628 set_control(uncommon_trap_call->in(0));
5629 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5630 Deoptimization::trap_request_action(trap_request));
5631 assert(stopped(), "Should be stopped");
5632 _gvn.hash_delete(uncommon_trap_call);
5633 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5634 }
5635
5636 // Common checks for array sorting intrinsics arguments.
5637 // Returns `true` if checks passed.
5638 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5639 // check address of the class
5640 if (elementType == nullptr || elementType->is_top()) {
5641 return false; // dead path
5642 }
5643 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5644 if (elem_klass == nullptr) {
5645 return false; // dead path
5646 }
5647 // java_mirror_type() returns non-null for compile-time Class constants only
5648 ciType* elem_type = elem_klass->java_mirror_type();
5649 if (elem_type == nullptr) {
5650 return false;
5651 }
5652 bt = elem_type->basic_type();
5653 // Disable the intrinsic if the CPU does not support SIMD sort
5654 if (!Matcher::supports_simd_sort(bt)) {
5655 return false;
5656 }
5657 // check address of the array
5658 if (obj == nullptr || obj->is_top()) {
5659 return false; // dead path
5660 }
5661 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5662 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5663 return false; // failed input validation
5664 }
5665 return true;
5666 }
5667
5668 //------------------------------inline_array_partition-----------------------
5669 bool LibraryCallKit::inline_array_partition() {
5670 address stubAddr = StubRoutines::select_array_partition_function();
5671 if (stubAddr == nullptr) {
5672 return false; // Intrinsic's stub is not implemented on this platform
5673 }
5674 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5675
5676 // no receiver because it is a static method
5677 Node* elementType = argument(0);
5678 Node* obj = argument(1);
5679 Node* offset = argument(2); // long
5680 Node* fromIndex = argument(4);
5681 Node* toIndex = argument(5);
5682 Node* indexPivot1 = argument(6);
5683 Node* indexPivot2 = argument(7);
5684 // PartitionOperation: argument(8) is ignored
5685
5686 Node* pivotIndices = nullptr;
5687 BasicType bt = T_ILLEGAL;
5688
5689 if (!check_array_sort_arguments(elementType, obj, bt)) {
5690 return false;
5691 }
5692 null_check(obj);
5693 // If obj is dead, only null-path is taken.
5694 if (stopped()) {
5695 return true;
5696 }
5697 // Set the original stack and the reexecute bit for the interpreter to reexecute
5698 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5699 { PreserveReexecuteState preexecs(this);
5700 jvms()->set_should_reexecute(true);
5701
5702 Node* obj_adr = make_unsafe_address(obj, offset);
5703
5704 // create the pivotIndices array of type int and size = 2
5705 Node* size = intcon(2);
5706 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5707 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5708 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5709 guarantee(alloc != nullptr, "created above");
5710 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5711
5712 // pass the basic type enum to the stub
5713 Node* elemType = intcon(bt);
5714
5715 // Call the stub
5716 const char *stubName = "array_partition_stub";
5717 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5718 stubAddr, stubName, TypePtr::BOTTOM,
5719 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5720 indexPivot1, indexPivot2);
5721
5722 } // original reexecute is set back here
5723
5724 if (!stopped()) {
5725 set_result(pivotIndices);
5726 }
5727
5728 return true;
5729 }
5730
5731
5732 //------------------------------inline_array_sort-----------------------
5733 bool LibraryCallKit::inline_array_sort() {
5734 address stubAddr = StubRoutines::select_arraysort_function();
5735 if (stubAddr == nullptr) {
5736 return false; // Intrinsic's stub is not implemented on this platform
5737 }
5738 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5739
5740 // no receiver because it is a static method
5741 Node* elementType = argument(0);
5742 Node* obj = argument(1);
5743 Node* offset = argument(2); // long
5744 Node* fromIndex = argument(4);
5745 Node* toIndex = argument(5);
5746 // SortOperation: argument(6) is ignored
5747
5748 BasicType bt = T_ILLEGAL;
5749
5750 if (!check_array_sort_arguments(elementType, obj, bt)) {
5751 return false;
5752 }
5753 null_check(obj);
5754 // If obj is dead, only null-path is taken.
5755 if (stopped()) {
5756 return true;
5757 }
5758 Node* obj_adr = make_unsafe_address(obj, offset);
5759
5760 // pass the basic type enum to the stub
5761 Node* elemType = intcon(bt);
5762
5763 // Call the stub.
5764 const char *stubName = "arraysort_stub";
5765 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5766 stubAddr, stubName, TypePtr::BOTTOM,
5767 obj_adr, elemType, fromIndex, toIndex);
5768
5769 return true;
5770 }
5771
5772
5773 //------------------------------inline_arraycopy-----------------------
5774 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5775 // Object dest, int destPos,
5776 // int length);
5777 bool LibraryCallKit::inline_arraycopy() {
5778 // Get the arguments.
5779 Node* src = argument(0); // type: oop
5780 Node* src_offset = argument(1); // type: int
5781 Node* dest = argument(2); // type: oop
5782 Node* dest_offset = argument(3); // type: int
5783 Node* length = argument(4); // type: int
5784
5785 uint new_idx = C->unique();
5786
5787 // Check for allocation before we add nodes that would confuse
5788 // tightly_coupled_allocation()
5789 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5790
5791 int saved_reexecute_sp = -1;
5792 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5793 // See arraycopy_restore_alloc_state() comment
5794 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5795 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5796 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5797 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5798
5799 // The following tests must be performed
5800 // (1) src and dest are arrays.
5801 // (2) src and dest arrays must have elements of the same BasicType
5802 // (3) src and dest must not be null.
5803 // (4) src_offset must not be negative.
5804 // (5) dest_offset must not be negative.
5805 // (6) length must not be negative.
5806 // (7) src_offset + length must not exceed length of src.
5807 // (8) dest_offset + length must not exceed length of dest.
5808 // (9) each element of an oop array must be assignable
5809
5810 // (3) src and dest must not be null.
5811 // always do this here because we need the JVM state for uncommon traps
5812 Node* null_ctl = top();
5813 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5814 assert(null_ctl->is_top(), "no null control here");
5815 dest = null_check(dest, T_ARRAY);
5816
5817 if (!can_emit_guards) {
5818 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5819 // guards but the arraycopy node could still take advantage of a
5820 // tightly allocated allocation. tightly_coupled_allocation() is
5821 // called again to make sure it takes the null check above into
5822 // account: the null check is mandatory and if it caused an
5823 // uncommon trap to be emitted then the allocation can't be
5824 // considered tightly coupled in this context.
5825 alloc = tightly_coupled_allocation(dest);
5826 }
5827
5828 bool validated = false;
5829
5830 const Type* src_type = _gvn.type(src);
5831 const Type* dest_type = _gvn.type(dest);
5832 const TypeAryPtr* top_src = src_type->isa_aryptr();
5833 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5834
5835 // Do we have the type of src?
5836 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5837 // Do we have the type of dest?
5838 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5839 // Is the type for src from speculation?
5840 bool src_spec = false;
5841 // Is the type for dest from speculation?
5842 bool dest_spec = false;
5843
5844 if ((!has_src || !has_dest) && can_emit_guards) {
5845 // We don't have sufficient type information, let's see if
5846 // speculative types can help. We need to have types for both src
5847 // and dest so that it pays off.
5848
5849 // Do we already have or could we have type information for src
5850 bool could_have_src = has_src;
5851 // Do we already have or could we have type information for dest
5852 bool could_have_dest = has_dest;
5853
5854 ciKlass* src_k = nullptr;
5855 if (!has_src) {
5856 src_k = src_type->speculative_type_not_null();
5857 if (src_k != nullptr && src_k->is_array_klass()) {
5858 could_have_src = true;
5859 }
5860 }
5861
5862 ciKlass* dest_k = nullptr;
5863 if (!has_dest) {
5864 dest_k = dest_type->speculative_type_not_null();
5865 if (dest_k != nullptr && dest_k->is_array_klass()) {
5866 could_have_dest = true;
5867 }
5868 }
5869
5870 if (could_have_src && could_have_dest) {
5871 // This is going to pay off so emit the required guards
5872 if (!has_src) {
5873 src = maybe_cast_profiled_obj(src, src_k, true);
5874 src_type = _gvn.type(src);
5875 top_src = src_type->isa_aryptr();
5876 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5877 src_spec = true;
5878 }
5879 if (!has_dest) {
5880 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5881 dest_type = _gvn.type(dest);
5882 top_dest = dest_type->isa_aryptr();
5883 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5884 dest_spec = true;
5885 }
5886 }
5887 }
5888
5889 if (has_src && has_dest && can_emit_guards) {
5890 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5891 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5892 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5893 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5894
5895 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5896 // If both arrays are object arrays then having the exact types
5897 // for both will remove the need for a subtype check at runtime
5898 // before the call and may make it possible to pick a faster copy
5899 // routine (without a subtype check on every element)
5900 // Do we have the exact type of src?
5901 bool could_have_src = src_spec;
5902 // Do we have the exact type of dest?
5903 bool could_have_dest = dest_spec;
5904 ciKlass* src_k = nullptr;
5905 ciKlass* dest_k = nullptr;
5906 if (!src_spec) {
5907 src_k = src_type->speculative_type_not_null();
5908 if (src_k != nullptr && src_k->is_array_klass()) {
5909 could_have_src = true;
5910 }
5911 }
5912 if (!dest_spec) {
5913 dest_k = dest_type->speculative_type_not_null();
5914 if (dest_k != nullptr && dest_k->is_array_klass()) {
5915 could_have_dest = true;
5916 }
5917 }
5918 if (could_have_src && could_have_dest) {
5919 // If we can have both exact types, emit the missing guards
5920 if (could_have_src && !src_spec) {
5921 src = maybe_cast_profiled_obj(src, src_k, true);
5922 }
5923 if (could_have_dest && !dest_spec) {
5924 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5925 }
5926 }
5927 }
5928 }
5929
5930 ciMethod* trap_method = method();
5931 int trap_bci = bci();
5932 if (saved_jvms_before_guards != nullptr) {
5933 trap_method = alloc->jvms()->method();
5934 trap_bci = alloc->jvms()->bci();
5935 }
5936
5937 bool negative_length_guard_generated = false;
5938
5939 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5940 can_emit_guards &&
5941 !src->is_top() && !dest->is_top()) {
5942 // validate arguments: enables transformation the ArrayCopyNode
5943 validated = true;
5944
5945 RegionNode* slow_region = new RegionNode(1);
5946 record_for_igvn(slow_region);
5947
5948 // (1) src and dest are arrays.
5949 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5950 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5951
5952 // (2) src and dest arrays must have elements of the same BasicType
5953 // done at macro expansion or at Ideal transformation time
5954
5955 // (4) src_offset must not be negative.
5956 generate_negative_guard(src_offset, slow_region);
5957
5958 // (5) dest_offset must not be negative.
5959 generate_negative_guard(dest_offset, slow_region);
5960
5961 // (7) src_offset + length must not exceed length of src.
5962 generate_limit_guard(src_offset, length,
5963 load_array_length(src),
5964 slow_region);
5965
5966 // (8) dest_offset + length must not exceed length of dest.
5967 generate_limit_guard(dest_offset, length,
5968 load_array_length(dest),
5969 slow_region);
5970
5971 // (6) length must not be negative.
5972 // This is also checked in generate_arraycopy() during macro expansion, but
5973 // we also have to check it here for the case where the ArrayCopyNode will
5974 // be eliminated by Escape Analysis.
5975 if (EliminateAllocations) {
5976 generate_negative_guard(length, slow_region);
5977 negative_length_guard_generated = true;
5978 }
5979
5980 // (9) each element of an oop array must be assignable
5981 Node* dest_klass = load_object_klass(dest);
5982 if (src != dest) {
5983 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5984
5985 if (not_subtype_ctrl != top()) {
5986 PreserveJVMState pjvms(this);
5987 set_control(not_subtype_ctrl);
5988 uncommon_trap(Deoptimization::Reason_intrinsic,
5989 Deoptimization::Action_make_not_entrant);
5990 assert(stopped(), "Should be stopped");
5991 }
5992 }
5993 {
5994 PreserveJVMState pjvms(this);
5995 set_control(_gvn.transform(slow_region));
5996 uncommon_trap(Deoptimization::Reason_intrinsic,
5997 Deoptimization::Action_make_not_entrant);
5998 assert(stopped(), "Should be stopped");
5999 }
6000
6001 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6002 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6003 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6004 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6005 }
6006
6007 if (stopped()) {
6008 return true;
6009 }
6010
6011 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6012 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6013 // so the compiler has a chance to eliminate them: during macro expansion,
6014 // we have to set their control (CastPP nodes are eliminated).
6015 load_object_klass(src), load_object_klass(dest),
6016 load_array_length(src), load_array_length(dest));
6017
6018 ac->set_arraycopy(validated);
6019
6020 Node* n = _gvn.transform(ac);
6021 if (n == ac) {
6022 ac->connect_outputs(this);
6023 } else {
6024 assert(validated, "shouldn't transform if all arguments not validated");
6025 set_all_memory(n);
6026 }
6027 clear_upper_avx();
6028
6029
6030 return true;
6031 }
6032
6033
6034 // Helper function which determines if an arraycopy immediately follows
6035 // an allocation, with no intervening tests or other escapes for the object.
6036 AllocateArrayNode*
6037 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6038 if (stopped()) return nullptr; // no fast path
6039 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6040
6041 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6042 if (alloc == nullptr) return nullptr;
6043
6044 Node* rawmem = memory(Compile::AliasIdxRaw);
6045 // Is the allocation's memory state untouched?
6046 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6047 // Bail out if there have been raw-memory effects since the allocation.
6048 // (Example: There might have been a call or safepoint.)
6049 return nullptr;
6050 }
6051 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6052 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6053 return nullptr;
6054 }
6055
6056 // There must be no unexpected observers of this allocation.
6057 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6058 Node* obs = ptr->fast_out(i);
6059 if (obs != this->map()) {
6060 return nullptr;
6061 }
6062 }
6063
6064 // This arraycopy must unconditionally follow the allocation of the ptr.
6065 Node* alloc_ctl = ptr->in(0);
6066 Node* ctl = control();
6067 while (ctl != alloc_ctl) {
6068 // There may be guards which feed into the slow_region.
6069 // Any other control flow means that we might not get a chance
6070 // to finish initializing the allocated object.
6071 // Various low-level checks bottom out in uncommon traps. These
6072 // are considered safe since we've already checked above that
6073 // there is no unexpected observer of this allocation.
6074 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6075 assert(ctl->in(0)->is_If(), "must be If");
6076 ctl = ctl->in(0)->in(0);
6077 } else {
6078 return nullptr;
6079 }
6080 }
6081
6082 // If we get this far, we have an allocation which immediately
6083 // precedes the arraycopy, and we can take over zeroing the new object.
6084 // The arraycopy will finish the initialization, and provide
6085 // a new control state to which we will anchor the destination pointer.
6086
6087 return alloc;
6088 }
6089
6090 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6091 if (node->is_IfProj()) {
6092 Node* other_proj = node->as_IfProj()->other_if_proj();
6093 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6094 Node* obs = other_proj->fast_out(j);
6095 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6096 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6097 return obs->as_CallStaticJava();
6098 }
6099 }
6100 }
6101 return nullptr;
6102 }
6103
6104 //-------------inline_encodeISOArray-----------------------------------
6105 // encode char[] to byte[] in ISO_8859_1 or ASCII
6106 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6107 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6108 // no receiver since it is static method
6109 Node *src = argument(0);
6110 Node *src_offset = argument(1);
6111 Node *dst = argument(2);
6112 Node *dst_offset = argument(3);
6113 Node *length = argument(4);
6114
6115 src = must_be_not_null(src, true);
6116 dst = must_be_not_null(dst, true);
6117
6118 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6119 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6120 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6121 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6122 // failed array check
6123 return false;
6124 }
6125
6126 // Figure out the size and type of the elements we will be copying.
6127 BasicType src_elem = src_type->elem()->array_element_basic_type();
6128 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6129 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6130 return false;
6131 }
6132
6133 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6134 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6135 // 'src_start' points to src array + scaled offset
6136 // 'dst_start' points to dst array + scaled offset
6137
6138 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6139 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6140 enc = _gvn.transform(enc);
6141 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6142 set_memory(res_mem, mtype);
6143 set_result(enc);
6144 clear_upper_avx();
6145
6146 return true;
6147 }
6148
6149 //-------------inline_multiplyToLen-----------------------------------
6150 bool LibraryCallKit::inline_multiplyToLen() {
6151 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6152
6153 address stubAddr = StubRoutines::multiplyToLen();
6154 if (stubAddr == nullptr) {
6155 return false; // Intrinsic's stub is not implemented on this platform
6156 }
6157 const char* stubName = "multiplyToLen";
6158
6159 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6160
6161 // no receiver because it is a static method
6162 Node* x = argument(0);
6163 Node* xlen = argument(1);
6164 Node* y = argument(2);
6165 Node* ylen = argument(3);
6166 Node* z = argument(4);
6167
6168 x = must_be_not_null(x, true);
6169 y = must_be_not_null(y, true);
6170
6171 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6172 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6173 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6174 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6175 // failed array check
6176 return false;
6177 }
6178
6179 BasicType x_elem = x_type->elem()->array_element_basic_type();
6180 BasicType y_elem = y_type->elem()->array_element_basic_type();
6181 if (x_elem != T_INT || y_elem != T_INT) {
6182 return false;
6183 }
6184
6185 Node* x_start = array_element_address(x, intcon(0), x_elem);
6186 Node* y_start = array_element_address(y, intcon(0), y_elem);
6187 // 'x_start' points to x array + scaled xlen
6188 // 'y_start' points to y array + scaled ylen
6189
6190 Node* z_start = array_element_address(z, intcon(0), T_INT);
6191
6192 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6193 OptoRuntime::multiplyToLen_Type(),
6194 stubAddr, stubName, TypePtr::BOTTOM,
6195 x_start, xlen, y_start, ylen, z_start);
6196
6197 C->set_has_split_ifs(true); // Has chance for split-if optimization
6198 set_result(z);
6199 return true;
6200 }
6201
6202 //-------------inline_squareToLen------------------------------------
6203 bool LibraryCallKit::inline_squareToLen() {
6204 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6205
6206 address stubAddr = StubRoutines::squareToLen();
6207 if (stubAddr == nullptr) {
6208 return false; // Intrinsic's stub is not implemented on this platform
6209 }
6210 const char* stubName = "squareToLen";
6211
6212 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6213
6214 Node* x = argument(0);
6215 Node* len = argument(1);
6216 Node* z = argument(2);
6217 Node* zlen = argument(3);
6218
6219 x = must_be_not_null(x, true);
6220 z = must_be_not_null(z, true);
6221
6222 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6223 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6224 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6225 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6226 // failed array check
6227 return false;
6228 }
6229
6230 BasicType x_elem = x_type->elem()->array_element_basic_type();
6231 BasicType z_elem = z_type->elem()->array_element_basic_type();
6232 if (x_elem != T_INT || z_elem != T_INT) {
6233 return false;
6234 }
6235
6236
6237 Node* x_start = array_element_address(x, intcon(0), x_elem);
6238 Node* z_start = array_element_address(z, intcon(0), z_elem);
6239
6240 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6241 OptoRuntime::squareToLen_Type(),
6242 stubAddr, stubName, TypePtr::BOTTOM,
6243 x_start, len, z_start, zlen);
6244
6245 set_result(z);
6246 return true;
6247 }
6248
6249 //-------------inline_mulAdd------------------------------------------
6250 bool LibraryCallKit::inline_mulAdd() {
6251 assert(UseMulAddIntrinsic, "not implemented on this platform");
6252
6253 address stubAddr = StubRoutines::mulAdd();
6254 if (stubAddr == nullptr) {
6255 return false; // Intrinsic's stub is not implemented on this platform
6256 }
6257 const char* stubName = "mulAdd";
6258
6259 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6260
6261 Node* out = argument(0);
6262 Node* in = argument(1);
6263 Node* offset = argument(2);
6264 Node* len = argument(3);
6265 Node* k = argument(4);
6266
6267 in = must_be_not_null(in, true);
6268 out = must_be_not_null(out, true);
6269
6270 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6271 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6272 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6273 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6274 // failed array check
6275 return false;
6276 }
6277
6278 BasicType out_elem = out_type->elem()->array_element_basic_type();
6279 BasicType in_elem = in_type->elem()->array_element_basic_type();
6280 if (out_elem != T_INT || in_elem != T_INT) {
6281 return false;
6282 }
6283
6284 Node* outlen = load_array_length(out);
6285 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6286 Node* out_start = array_element_address(out, intcon(0), out_elem);
6287 Node* in_start = array_element_address(in, intcon(0), in_elem);
6288
6289 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6290 OptoRuntime::mulAdd_Type(),
6291 stubAddr, stubName, TypePtr::BOTTOM,
6292 out_start,in_start, new_offset, len, k);
6293 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6294 set_result(result);
6295 return true;
6296 }
6297
6298 //-------------inline_montgomeryMultiply-----------------------------------
6299 bool LibraryCallKit::inline_montgomeryMultiply() {
6300 address stubAddr = StubRoutines::montgomeryMultiply();
6301 if (stubAddr == nullptr) {
6302 return false; // Intrinsic's stub is not implemented on this platform
6303 }
6304
6305 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6306 const char* stubName = "montgomery_multiply";
6307
6308 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6309
6310 Node* a = argument(0);
6311 Node* b = argument(1);
6312 Node* n = argument(2);
6313 Node* len = argument(3);
6314 Node* inv = argument(4);
6315 Node* m = argument(6);
6316
6317 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6318 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6319 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6320 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6321 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6322 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6323 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6324 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6325 // failed array check
6326 return false;
6327 }
6328
6329 BasicType a_elem = a_type->elem()->array_element_basic_type();
6330 BasicType b_elem = b_type->elem()->array_element_basic_type();
6331 BasicType n_elem = n_type->elem()->array_element_basic_type();
6332 BasicType m_elem = m_type->elem()->array_element_basic_type();
6333 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6334 return false;
6335 }
6336
6337 // Make the call
6338 {
6339 Node* a_start = array_element_address(a, intcon(0), a_elem);
6340 Node* b_start = array_element_address(b, intcon(0), b_elem);
6341 Node* n_start = array_element_address(n, intcon(0), n_elem);
6342 Node* m_start = array_element_address(m, intcon(0), m_elem);
6343
6344 Node* call = make_runtime_call(RC_LEAF,
6345 OptoRuntime::montgomeryMultiply_Type(),
6346 stubAddr, stubName, TypePtr::BOTTOM,
6347 a_start, b_start, n_start, len, inv, top(),
6348 m_start);
6349 set_result(m);
6350 }
6351
6352 return true;
6353 }
6354
6355 bool LibraryCallKit::inline_montgomerySquare() {
6356 address stubAddr = StubRoutines::montgomerySquare();
6357 if (stubAddr == nullptr) {
6358 return false; // Intrinsic's stub is not implemented on this platform
6359 }
6360
6361 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6362 const char* stubName = "montgomery_square";
6363
6364 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6365
6366 Node* a = argument(0);
6367 Node* n = argument(1);
6368 Node* len = argument(2);
6369 Node* inv = argument(3);
6370 Node* m = argument(5);
6371
6372 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6373 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6374 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6375 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6376 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6377 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6378 // failed array check
6379 return false;
6380 }
6381
6382 BasicType a_elem = a_type->elem()->array_element_basic_type();
6383 BasicType n_elem = n_type->elem()->array_element_basic_type();
6384 BasicType m_elem = m_type->elem()->array_element_basic_type();
6385 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6386 return false;
6387 }
6388
6389 // Make the call
6390 {
6391 Node* a_start = array_element_address(a, intcon(0), a_elem);
6392 Node* n_start = array_element_address(n, intcon(0), n_elem);
6393 Node* m_start = array_element_address(m, intcon(0), m_elem);
6394
6395 Node* call = make_runtime_call(RC_LEAF,
6396 OptoRuntime::montgomerySquare_Type(),
6397 stubAddr, stubName, TypePtr::BOTTOM,
6398 a_start, n_start, len, inv, top(),
6399 m_start);
6400 set_result(m);
6401 }
6402
6403 return true;
6404 }
6405
6406 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6407 address stubAddr = nullptr;
6408 const char* stubName = nullptr;
6409
6410 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6411 if (stubAddr == nullptr) {
6412 return false; // Intrinsic's stub is not implemented on this platform
6413 }
6414
6415 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6416
6417 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6418
6419 Node* newArr = argument(0);
6420 Node* oldArr = argument(1);
6421 Node* newIdx = argument(2);
6422 Node* shiftCount = argument(3);
6423 Node* numIter = argument(4);
6424
6425 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6426 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6427 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6428 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6429 return false;
6430 }
6431
6432 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6433 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6434 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6435 return false;
6436 }
6437
6438 // Make the call
6439 {
6440 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6441 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6442
6443 Node* call = make_runtime_call(RC_LEAF,
6444 OptoRuntime::bigIntegerShift_Type(),
6445 stubAddr,
6446 stubName,
6447 TypePtr::BOTTOM,
6448 newArr_start,
6449 oldArr_start,
6450 newIdx,
6451 shiftCount,
6452 numIter);
6453 }
6454
6455 return true;
6456 }
6457
6458 //-------------inline_vectorizedMismatch------------------------------
6459 bool LibraryCallKit::inline_vectorizedMismatch() {
6460 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6461
6462 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6463 Node* obja = argument(0); // Object
6464 Node* aoffset = argument(1); // long
6465 Node* objb = argument(3); // Object
6466 Node* boffset = argument(4); // long
6467 Node* length = argument(6); // int
6468 Node* scale = argument(7); // int
6469
6470 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6471 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6472 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6473 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6474 scale == top()) {
6475 return false; // failed input validation
6476 }
6477
6478 Node* obja_adr = make_unsafe_address(obja, aoffset);
6479 Node* objb_adr = make_unsafe_address(objb, boffset);
6480
6481 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6482 //
6483 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6484 // if (length <= inline_limit) {
6485 // inline_path:
6486 // vmask = VectorMaskGen length
6487 // vload1 = LoadVectorMasked obja, vmask
6488 // vload2 = LoadVectorMasked objb, vmask
6489 // result1 = VectorCmpMasked vload1, vload2, vmask
6490 // } else {
6491 // call_stub_path:
6492 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6493 // }
6494 // exit_block:
6495 // return Phi(result1, result2);
6496 //
6497 enum { inline_path = 1, // input is small enough to process it all at once
6498 stub_path = 2, // input is too large; call into the VM
6499 PATH_LIMIT = 3
6500 };
6501
6502 Node* exit_block = new RegionNode(PATH_LIMIT);
6503 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6504 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6505
6506 Node* call_stub_path = control();
6507
6508 BasicType elem_bt = T_ILLEGAL;
6509
6510 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6511 if (scale_t->is_con()) {
6512 switch (scale_t->get_con()) {
6513 case 0: elem_bt = T_BYTE; break;
6514 case 1: elem_bt = T_SHORT; break;
6515 case 2: elem_bt = T_INT; break;
6516 case 3: elem_bt = T_LONG; break;
6517
6518 default: elem_bt = T_ILLEGAL; break; // not supported
6519 }
6520 }
6521
6522 int inline_limit = 0;
6523 bool do_partial_inline = false;
6524
6525 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6526 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6527 do_partial_inline = inline_limit >= 16;
6528 }
6529
6530 if (do_partial_inline) {
6531 assert(elem_bt != T_ILLEGAL, "sanity");
6532
6533 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6534 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6535 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6536
6537 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6538 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6539 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6540
6541 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6542
6543 if (!stopped()) {
6544 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6545
6546 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6547 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6548 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6549 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6550
6551 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6552 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6553 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6554 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6555
6556 exit_block->init_req(inline_path, control());
6557 memory_phi->init_req(inline_path, map()->memory());
6558 result_phi->init_req(inline_path, result);
6559
6560 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6561 clear_upper_avx();
6562 }
6563 }
6564 }
6565
6566 if (call_stub_path != nullptr) {
6567 set_control(call_stub_path);
6568
6569 Node* call = make_runtime_call(RC_LEAF,
6570 OptoRuntime::vectorizedMismatch_Type(),
6571 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6572 obja_adr, objb_adr, length, scale);
6573
6574 exit_block->init_req(stub_path, control());
6575 memory_phi->init_req(stub_path, map()->memory());
6576 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6577 }
6578
6579 exit_block = _gvn.transform(exit_block);
6580 memory_phi = _gvn.transform(memory_phi);
6581 result_phi = _gvn.transform(result_phi);
6582
6583 set_control(exit_block);
6584 set_all_memory(memory_phi);
6585 set_result(result_phi);
6586
6587 return true;
6588 }
6589
6590 //------------------------------inline_vectorizedHashcode----------------------------
6591 bool LibraryCallKit::inline_vectorizedHashCode() {
6592 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6593
6594 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6595 Node* array = argument(0);
6596 Node* offset = argument(1);
6597 Node* length = argument(2);
6598 Node* initialValue = argument(3);
6599 Node* basic_type = argument(4);
6600
6601 if (basic_type == top()) {
6602 return false; // failed input validation
6603 }
6604
6605 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6606 if (!basic_type_t->is_con()) {
6607 return false; // Only intrinsify if mode argument is constant
6608 }
6609
6610 array = must_be_not_null(array, true);
6611
6612 BasicType bt = (BasicType)basic_type_t->get_con();
6613
6614 // Resolve address of first element
6615 Node* array_start = array_element_address(array, offset, bt);
6616
6617 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6618 array_start, length, initialValue, basic_type)));
6619 clear_upper_avx();
6620
6621 return true;
6622 }
6623
6624 /**
6625 * Calculate CRC32 for byte.
6626 * int java.util.zip.CRC32.update(int crc, int b)
6627 */
6628 bool LibraryCallKit::inline_updateCRC32() {
6629 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6630 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6631 // no receiver since it is static method
6632 Node* crc = argument(0); // type: int
6633 Node* b = argument(1); // type: int
6634
6635 /*
6636 * int c = ~ crc;
6637 * b = timesXtoThe32[(b ^ c) & 0xFF];
6638 * b = b ^ (c >>> 8);
6639 * crc = ~b;
6640 */
6641
6642 Node* M1 = intcon(-1);
6643 crc = _gvn.transform(new XorINode(crc, M1));
6644 Node* result = _gvn.transform(new XorINode(crc, b));
6645 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6646
6647 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6648 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6649 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6650 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6651
6652 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6653 result = _gvn.transform(new XorINode(crc, result));
6654 result = _gvn.transform(new XorINode(result, M1));
6655 set_result(result);
6656 return true;
6657 }
6658
6659 /**
6660 * Calculate CRC32 for byte[] array.
6661 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6662 */
6663 bool LibraryCallKit::inline_updateBytesCRC32() {
6664 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6665 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6666 // no receiver since it is static method
6667 Node* crc = argument(0); // type: int
6668 Node* src = argument(1); // type: oop
6669 Node* offset = argument(2); // type: int
6670 Node* length = argument(3); // type: int
6671
6672 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6673 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6674 // failed array check
6675 return false;
6676 }
6677
6678 // Figure out the size and type of the elements we will be copying.
6679 BasicType src_elem = src_type->elem()->array_element_basic_type();
6680 if (src_elem != T_BYTE) {
6681 return false;
6682 }
6683
6684 // 'src_start' points to src array + scaled offset
6685 src = must_be_not_null(src, true);
6686 Node* src_start = array_element_address(src, offset, src_elem);
6687
6688 // We assume that range check is done by caller.
6689 // TODO: generate range check (offset+length < src.length) in debug VM.
6690
6691 // Call the stub.
6692 address stubAddr = StubRoutines::updateBytesCRC32();
6693 const char *stubName = "updateBytesCRC32";
6694
6695 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6696 stubAddr, stubName, TypePtr::BOTTOM,
6697 crc, src_start, length);
6698 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6699 set_result(result);
6700 return true;
6701 }
6702
6703 /**
6704 * Calculate CRC32 for ByteBuffer.
6705 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6706 */
6707 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6708 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6709 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6710 // no receiver since it is static method
6711 Node* crc = argument(0); // type: int
6712 Node* src = argument(1); // type: long
6713 Node* offset = argument(3); // type: int
6714 Node* length = argument(4); // type: int
6715
6716 src = ConvL2X(src); // adjust Java long to machine word
6717 Node* base = _gvn.transform(new CastX2PNode(src));
6718 offset = ConvI2X(offset);
6719
6720 // 'src_start' points to src array + scaled offset
6721 Node* src_start = basic_plus_adr(top(), base, offset);
6722
6723 // Call the stub.
6724 address stubAddr = StubRoutines::updateBytesCRC32();
6725 const char *stubName = "updateBytesCRC32";
6726
6727 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6728 stubAddr, stubName, TypePtr::BOTTOM,
6729 crc, src_start, length);
6730 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6731 set_result(result);
6732 return true;
6733 }
6734
6735 //------------------------------get_table_from_crc32c_class-----------------------
6736 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6737 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6738 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6739
6740 return table;
6741 }
6742
6743 //------------------------------inline_updateBytesCRC32C-----------------------
6744 //
6745 // Calculate CRC32C for byte[] array.
6746 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6747 //
6748 bool LibraryCallKit::inline_updateBytesCRC32C() {
6749 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6750 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6751 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6752 // no receiver since it is a static method
6753 Node* crc = argument(0); // type: int
6754 Node* src = argument(1); // type: oop
6755 Node* offset = argument(2); // type: int
6756 Node* end = argument(3); // type: int
6757
6758 Node* length = _gvn.transform(new SubINode(end, offset));
6759
6760 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6761 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6762 // failed array check
6763 return false;
6764 }
6765
6766 // Figure out the size and type of the elements we will be copying.
6767 BasicType src_elem = src_type->elem()->array_element_basic_type();
6768 if (src_elem != T_BYTE) {
6769 return false;
6770 }
6771
6772 // 'src_start' points to src array + scaled offset
6773 src = must_be_not_null(src, true);
6774 Node* src_start = array_element_address(src, offset, src_elem);
6775
6776 // static final int[] byteTable in class CRC32C
6777 Node* table = get_table_from_crc32c_class(callee()->holder());
6778 table = must_be_not_null(table, true);
6779 Node* table_start = array_element_address(table, intcon(0), T_INT);
6780
6781 // We assume that range check is done by caller.
6782 // TODO: generate range check (offset+length < src.length) in debug VM.
6783
6784 // Call the stub.
6785 address stubAddr = StubRoutines::updateBytesCRC32C();
6786 const char *stubName = "updateBytesCRC32C";
6787
6788 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6789 stubAddr, stubName, TypePtr::BOTTOM,
6790 crc, src_start, length, table_start);
6791 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6792 set_result(result);
6793 return true;
6794 }
6795
6796 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6797 //
6798 // Calculate CRC32C for DirectByteBuffer.
6799 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6800 //
6801 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6802 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6803 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6804 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6805 // no receiver since it is a static method
6806 Node* crc = argument(0); // type: int
6807 Node* src = argument(1); // type: long
6808 Node* offset = argument(3); // type: int
6809 Node* end = argument(4); // type: int
6810
6811 Node* length = _gvn.transform(new SubINode(end, offset));
6812
6813 src = ConvL2X(src); // adjust Java long to machine word
6814 Node* base = _gvn.transform(new CastX2PNode(src));
6815 offset = ConvI2X(offset);
6816
6817 // 'src_start' points to src array + scaled offset
6818 Node* src_start = basic_plus_adr(top(), base, offset);
6819
6820 // static final int[] byteTable in class CRC32C
6821 Node* table = get_table_from_crc32c_class(callee()->holder());
6822 table = must_be_not_null(table, true);
6823 Node* table_start = array_element_address(table, intcon(0), T_INT);
6824
6825 // Call the stub.
6826 address stubAddr = StubRoutines::updateBytesCRC32C();
6827 const char *stubName = "updateBytesCRC32C";
6828
6829 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6830 stubAddr, stubName, TypePtr::BOTTOM,
6831 crc, src_start, length, table_start);
6832 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6833 set_result(result);
6834 return true;
6835 }
6836
6837 //------------------------------inline_updateBytesAdler32----------------------
6838 //
6839 // Calculate Adler32 checksum for byte[] array.
6840 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6841 //
6842 bool LibraryCallKit::inline_updateBytesAdler32() {
6843 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6844 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6845 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6846 // no receiver since it is static method
6847 Node* crc = argument(0); // type: int
6848 Node* src = argument(1); // type: oop
6849 Node* offset = argument(2); // type: int
6850 Node* length = argument(3); // type: int
6851
6852 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6853 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6854 // failed array check
6855 return false;
6856 }
6857
6858 // Figure out the size and type of the elements we will be copying.
6859 BasicType src_elem = src_type->elem()->array_element_basic_type();
6860 if (src_elem != T_BYTE) {
6861 return false;
6862 }
6863
6864 // 'src_start' points to src array + scaled offset
6865 Node* src_start = array_element_address(src, offset, src_elem);
6866
6867 // We assume that range check is done by caller.
6868 // TODO: generate range check (offset+length < src.length) in debug VM.
6869
6870 // Call the stub.
6871 address stubAddr = StubRoutines::updateBytesAdler32();
6872 const char *stubName = "updateBytesAdler32";
6873
6874 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6875 stubAddr, stubName, TypePtr::BOTTOM,
6876 crc, src_start, length);
6877 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6878 set_result(result);
6879 return true;
6880 }
6881
6882 //------------------------------inline_updateByteBufferAdler32---------------
6883 //
6884 // Calculate Adler32 checksum for DirectByteBuffer.
6885 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6886 //
6887 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6888 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6889 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6890 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6891 // no receiver since it is static method
6892 Node* crc = argument(0); // type: int
6893 Node* src = argument(1); // type: long
6894 Node* offset = argument(3); // type: int
6895 Node* length = argument(4); // type: int
6896
6897 src = ConvL2X(src); // adjust Java long to machine word
6898 Node* base = _gvn.transform(new CastX2PNode(src));
6899 offset = ConvI2X(offset);
6900
6901 // 'src_start' points to src array + scaled offset
6902 Node* src_start = basic_plus_adr(top(), base, offset);
6903
6904 // Call the stub.
6905 address stubAddr = StubRoutines::updateBytesAdler32();
6906 const char *stubName = "updateBytesAdler32";
6907
6908 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6909 stubAddr, stubName, TypePtr::BOTTOM,
6910 crc, src_start, length);
6911
6912 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6913 set_result(result);
6914 return true;
6915 }
6916
6917 //----------------------------inline_reference_get----------------------------
6918 // public T java.lang.ref.Reference.get();
6919 bool LibraryCallKit::inline_reference_get() {
6920 const int referent_offset = java_lang_ref_Reference::referent_offset();
6921
6922 // Get the argument:
6923 Node* reference_obj = null_check_receiver();
6924 if (stopped()) return true;
6925
6926 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6927 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6928 decorators, /*is_static*/ false, nullptr);
6929 if (result == nullptr) return false;
6930
6931 // Add memory barrier to prevent commoning reads from this field
6932 // across safepoint since GC can change its value.
6933 insert_mem_bar(Op_MemBarCPUOrder);
6934
6935 set_result(result);
6936 return true;
6937 }
6938
6939 //----------------------------inline_reference_refersTo0----------------------------
6940 // bool java.lang.ref.Reference.refersTo0();
6941 // bool java.lang.ref.PhantomReference.refersTo0();
6942 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6943 // Get arguments:
6944 Node* reference_obj = null_check_receiver();
6945 Node* other_obj = argument(1);
6946 if (stopped()) return true;
6947
6948 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6949 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6950 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6951 decorators, /*is_static*/ false, nullptr);
6952 if (referent == nullptr) return false;
6953
6954 // Add memory barrier to prevent commoning reads from this field
6955 // across safepoint since GC can change its value.
6956 insert_mem_bar(Op_MemBarCPUOrder);
6957
6958 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6959 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6960 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6961
6962 RegionNode* region = new RegionNode(3);
6963 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6964
6965 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6966 region->init_req(1, if_true);
6967 phi->init_req(1, intcon(1));
6968
6969 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6970 region->init_req(2, if_false);
6971 phi->init_req(2, intcon(0));
6972
6973 set_control(_gvn.transform(region));
6974 record_for_igvn(region);
6975 set_result(_gvn.transform(phi));
6976 return true;
6977 }
6978
6979 //----------------------------inline_reference_clear0----------------------------
6980 // void java.lang.ref.Reference.clear0();
6981 // void java.lang.ref.PhantomReference.clear0();
6982 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6983 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6984
6985 // Get arguments
6986 Node* reference_obj = null_check_receiver();
6987 if (stopped()) return true;
6988
6989 // Common access parameters
6990 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6991 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6992 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6993 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6994 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6995
6996 Node* referent = access_load_at(reference_obj,
6997 referent_field_addr,
6998 referent_field_addr_type,
6999 val_type,
7000 T_OBJECT,
7001 decorators);
7002
7003 IdealKit ideal(this);
7004 #define __ ideal.
7005 __ if_then(referent, BoolTest::ne, null());
7006 sync_kit(ideal);
7007 access_store_at(reference_obj,
7008 referent_field_addr,
7009 referent_field_addr_type,
7010 null(),
7011 val_type,
7012 T_OBJECT,
7013 decorators);
7014 __ sync_kit(this);
7015 __ end_if();
7016 final_sync(ideal);
7017 #undef __
7018
7019 return true;
7020 }
7021
7022 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7023 DecoratorSet decorators, bool is_static,
7024 ciInstanceKlass* fromKls) {
7025 if (fromKls == nullptr) {
7026 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7027 assert(tinst != nullptr, "obj is null");
7028 assert(tinst->is_loaded(), "obj is not loaded");
7029 fromKls = tinst->instance_klass();
7030 } else {
7031 assert(is_static, "only for static field access");
7032 }
7033 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7034 ciSymbol::make(fieldTypeString),
7035 is_static);
7036
7037 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7038 if (field == nullptr) return (Node *) nullptr;
7039
7040 if (is_static) {
7041 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7042 fromObj = makecon(tip);
7043 }
7044
7045 // Next code copied from Parse::do_get_xxx():
7046
7047 // Compute address and memory type.
7048 int offset = field->offset_in_bytes();
7049 bool is_vol = field->is_volatile();
7050 ciType* field_klass = field->type();
7051 assert(field_klass->is_loaded(), "should be loaded");
7052 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7053 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7054 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7055 "slice of address and input slice don't match");
7056 BasicType bt = field->layout_type();
7057
7058 // Build the resultant type of the load
7059 const Type *type;
7060 if (bt == T_OBJECT) {
7061 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7062 } else {
7063 type = Type::get_const_basic_type(bt);
7064 }
7065
7066 if (is_vol) {
7067 decorators |= MO_SEQ_CST;
7068 }
7069
7070 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7071 }
7072
7073 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7074 bool is_exact /* true */, bool is_static /* false */,
7075 ciInstanceKlass * fromKls /* nullptr */) {
7076 if (fromKls == nullptr) {
7077 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7078 assert(tinst != nullptr, "obj is null");
7079 assert(tinst->is_loaded(), "obj is not loaded");
7080 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7081 fromKls = tinst->instance_klass();
7082 }
7083 else {
7084 assert(is_static, "only for static field access");
7085 }
7086 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7087 ciSymbol::make(fieldTypeString),
7088 is_static);
7089
7090 assert(field != nullptr, "undefined field");
7091 assert(!field->is_volatile(), "not defined for volatile fields");
7092
7093 if (is_static) {
7094 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7095 fromObj = makecon(tip);
7096 }
7097
7098 // Next code copied from Parse::do_get_xxx():
7099
7100 // Compute address and memory type.
7101 int offset = field->offset_in_bytes();
7102 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7103
7104 return adr;
7105 }
7106
7107 //------------------------------inline_aescrypt_Block-----------------------
7108 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7109 address stubAddr = nullptr;
7110 const char *stubName;
7111 assert(UseAES, "need AES instruction support");
7112
7113 switch(id) {
7114 case vmIntrinsics::_aescrypt_encryptBlock:
7115 stubAddr = StubRoutines::aescrypt_encryptBlock();
7116 stubName = "aescrypt_encryptBlock";
7117 break;
7118 case vmIntrinsics::_aescrypt_decryptBlock:
7119 stubAddr = StubRoutines::aescrypt_decryptBlock();
7120 stubName = "aescrypt_decryptBlock";
7121 break;
7122 default:
7123 break;
7124 }
7125 if (stubAddr == nullptr) return false;
7126
7127 Node* aescrypt_object = argument(0);
7128 Node* src = argument(1);
7129 Node* src_offset = argument(2);
7130 Node* dest = argument(3);
7131 Node* dest_offset = argument(4);
7132
7133 src = must_be_not_null(src, true);
7134 dest = must_be_not_null(dest, true);
7135
7136 // (1) src and dest are arrays.
7137 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7138 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7139 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7140 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7141
7142 // for the quick and dirty code we will skip all the checks.
7143 // we are just trying to get the call to be generated.
7144 Node* src_start = src;
7145 Node* dest_start = dest;
7146 if (src_offset != nullptr || dest_offset != nullptr) {
7147 assert(src_offset != nullptr && dest_offset != nullptr, "");
7148 src_start = array_element_address(src, src_offset, T_BYTE);
7149 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7150 }
7151
7152 // now need to get the start of its expanded key array
7153 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7154 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7155 if (k_start == nullptr) return false;
7156
7157 // Call the stub.
7158 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7159 stubAddr, stubName, TypePtr::BOTTOM,
7160 src_start, dest_start, k_start);
7161
7162 return true;
7163 }
7164
7165 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7166 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7167 address stubAddr = nullptr;
7168 const char *stubName = nullptr;
7169
7170 assert(UseAES, "need AES instruction support");
7171
7172 switch(id) {
7173 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7174 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7175 stubName = "cipherBlockChaining_encryptAESCrypt";
7176 break;
7177 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7178 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7179 stubName = "cipherBlockChaining_decryptAESCrypt";
7180 break;
7181 default:
7182 break;
7183 }
7184 if (stubAddr == nullptr) return false;
7185
7186 Node* cipherBlockChaining_object = argument(0);
7187 Node* src = argument(1);
7188 Node* src_offset = argument(2);
7189 Node* len = argument(3);
7190 Node* dest = argument(4);
7191 Node* dest_offset = argument(5);
7192
7193 src = must_be_not_null(src, false);
7194 dest = must_be_not_null(dest, false);
7195
7196 // (1) src and dest are arrays.
7197 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7198 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7199 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7200 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7201
7202 // checks are the responsibility of the caller
7203 Node* src_start = src;
7204 Node* dest_start = dest;
7205 if (src_offset != nullptr || dest_offset != nullptr) {
7206 assert(src_offset != nullptr && dest_offset != nullptr, "");
7207 src_start = array_element_address(src, src_offset, T_BYTE);
7208 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7209 }
7210
7211 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7212 // (because of the predicated logic executed earlier).
7213 // so we cast it here safely.
7214 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7215
7216 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7217 if (embeddedCipherObj == nullptr) return false;
7218
7219 // cast it to what we know it will be at runtime
7220 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7221 assert(tinst != nullptr, "CBC obj is null");
7222 assert(tinst->is_loaded(), "CBC obj is not loaded");
7223 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7224 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7225
7226 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7227 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7228 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7229 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7230 aescrypt_object = _gvn.transform(aescrypt_object);
7231
7232 // we need to get the start of the aescrypt_object's expanded key array
7233 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7234 if (k_start == nullptr) return false;
7235
7236 // similarly, get the start address of the r vector
7237 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7238 if (objRvec == nullptr) return false;
7239 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7240
7241 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7242 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7243 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7244 stubAddr, stubName, TypePtr::BOTTOM,
7245 src_start, dest_start, k_start, r_start, len);
7246
7247 // return cipher length (int)
7248 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7249 set_result(retvalue);
7250 return true;
7251 }
7252
7253 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7254 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7255 address stubAddr = nullptr;
7256 const char *stubName = nullptr;
7257
7258 assert(UseAES, "need AES instruction support");
7259
7260 switch (id) {
7261 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7262 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7263 stubName = "electronicCodeBook_encryptAESCrypt";
7264 break;
7265 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7266 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7267 stubName = "electronicCodeBook_decryptAESCrypt";
7268 break;
7269 default:
7270 break;
7271 }
7272
7273 if (stubAddr == nullptr) return false;
7274
7275 Node* electronicCodeBook_object = argument(0);
7276 Node* src = argument(1);
7277 Node* src_offset = argument(2);
7278 Node* len = argument(3);
7279 Node* dest = argument(4);
7280 Node* dest_offset = argument(5);
7281
7282 // (1) src and dest are arrays.
7283 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7284 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7285 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7286 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7287
7288 // checks are the responsibility of the caller
7289 Node* src_start = src;
7290 Node* dest_start = dest;
7291 if (src_offset != nullptr || dest_offset != nullptr) {
7292 assert(src_offset != nullptr && dest_offset != nullptr, "");
7293 src_start = array_element_address(src, src_offset, T_BYTE);
7294 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7295 }
7296
7297 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7298 // (because of the predicated logic executed earlier).
7299 // so we cast it here safely.
7300 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7301
7302 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7303 if (embeddedCipherObj == nullptr) return false;
7304
7305 // cast it to what we know it will be at runtime
7306 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7307 assert(tinst != nullptr, "ECB obj is null");
7308 assert(tinst->is_loaded(), "ECB obj is not loaded");
7309 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7310 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7311
7312 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7313 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7314 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7315 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7316 aescrypt_object = _gvn.transform(aescrypt_object);
7317
7318 // we need to get the start of the aescrypt_object's expanded key array
7319 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7320 if (k_start == nullptr) return false;
7321
7322 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7323 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7324 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7325 stubAddr, stubName, TypePtr::BOTTOM,
7326 src_start, dest_start, k_start, len);
7327
7328 // return cipher length (int)
7329 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7330 set_result(retvalue);
7331 return true;
7332 }
7333
7334 //------------------------------inline_counterMode_AESCrypt-----------------------
7335 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7336 assert(UseAES, "need AES instruction support");
7337 if (!UseAESCTRIntrinsics) return false;
7338
7339 address stubAddr = nullptr;
7340 const char *stubName = nullptr;
7341 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7342 stubAddr = StubRoutines::counterMode_AESCrypt();
7343 stubName = "counterMode_AESCrypt";
7344 }
7345 if (stubAddr == nullptr) return false;
7346
7347 Node* counterMode_object = argument(0);
7348 Node* src = argument(1);
7349 Node* src_offset = argument(2);
7350 Node* len = argument(3);
7351 Node* dest = argument(4);
7352 Node* dest_offset = argument(5);
7353
7354 // (1) src and dest are arrays.
7355 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7356 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7357 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7358 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7359
7360 // checks are the responsibility of the caller
7361 Node* src_start = src;
7362 Node* dest_start = dest;
7363 if (src_offset != nullptr || dest_offset != nullptr) {
7364 assert(src_offset != nullptr && dest_offset != nullptr, "");
7365 src_start = array_element_address(src, src_offset, T_BYTE);
7366 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7367 }
7368
7369 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7370 // (because of the predicated logic executed earlier).
7371 // so we cast it here safely.
7372 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7373 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7374 if (embeddedCipherObj == nullptr) return false;
7375 // cast it to what we know it will be at runtime
7376 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7377 assert(tinst != nullptr, "CTR obj is null");
7378 assert(tinst->is_loaded(), "CTR obj is not loaded");
7379 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7380 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7381 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7382 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7383 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7384 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7385 aescrypt_object = _gvn.transform(aescrypt_object);
7386 // we need to get the start of the aescrypt_object's expanded key array
7387 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7388 if (k_start == nullptr) return false;
7389 // similarly, get the start address of the r vector
7390 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7391 if (obj_counter == nullptr) return false;
7392 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7393
7394 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7395 if (saved_encCounter == nullptr) return false;
7396 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7397 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7398
7399 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7400 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7401 OptoRuntime::counterMode_aescrypt_Type(),
7402 stubAddr, stubName, TypePtr::BOTTOM,
7403 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7404
7405 // return cipher length (int)
7406 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7407 set_result(retvalue);
7408 return true;
7409 }
7410
7411 //------------------------------get_key_start_from_aescrypt_object-----------------------
7412 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7413 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7414 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7415 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7416 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7417 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7418 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7419 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7420 if (objSessionK == nullptr) {
7421 return (Node *) nullptr;
7422 }
7423 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7424 #else
7425 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7426 #endif // PPC64
7427 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7428 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7429
7430 // now have the array, need to get the start address of the K array
7431 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7432 return k_start;
7433 }
7434
7435 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7436 // Return node representing slow path of predicate check.
7437 // the pseudo code we want to emulate with this predicate is:
7438 // for encryption:
7439 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7440 // for decryption:
7441 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7442 // note cipher==plain is more conservative than the original java code but that's OK
7443 //
7444 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7445 // The receiver was checked for null already.
7446 Node* objCBC = argument(0);
7447
7448 Node* src = argument(1);
7449 Node* dest = argument(4);
7450
7451 // Load embeddedCipher field of CipherBlockChaining object.
7452 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7453
7454 // get AESCrypt klass for instanceOf check
7455 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7456 // will have same classloader as CipherBlockChaining object
7457 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7458 assert(tinst != nullptr, "CBCobj is null");
7459 assert(tinst->is_loaded(), "CBCobj is not loaded");
7460
7461 // we want to do an instanceof comparison against the AESCrypt class
7462 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7463 if (!klass_AESCrypt->is_loaded()) {
7464 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7465 Node* ctrl = control();
7466 set_control(top()); // no regular fast path
7467 return ctrl;
7468 }
7469
7470 src = must_be_not_null(src, true);
7471 dest = must_be_not_null(dest, true);
7472
7473 // Resolve oops to stable for CmpP below.
7474 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7475
7476 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7477 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7478 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7479
7480 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7481
7482 // for encryption, we are done
7483 if (!decrypting)
7484 return instof_false; // even if it is null
7485
7486 // for decryption, we need to add a further check to avoid
7487 // taking the intrinsic path when cipher and plain are the same
7488 // see the original java code for why.
7489 RegionNode* region = new RegionNode(3);
7490 region->init_req(1, instof_false);
7491
7492 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7493 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7494 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7495 region->init_req(2, src_dest_conjoint);
7496
7497 record_for_igvn(region);
7498 return _gvn.transform(region);
7499 }
7500
7501 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7502 // Return node representing slow path of predicate check.
7503 // the pseudo code we want to emulate with this predicate is:
7504 // for encryption:
7505 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7506 // for decryption:
7507 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7508 // note cipher==plain is more conservative than the original java code but that's OK
7509 //
7510 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7511 // The receiver was checked for null already.
7512 Node* objECB = argument(0);
7513
7514 // Load embeddedCipher field of ElectronicCodeBook object.
7515 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7516
7517 // get AESCrypt klass for instanceOf check
7518 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7519 // will have same classloader as ElectronicCodeBook object
7520 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7521 assert(tinst != nullptr, "ECBobj is null");
7522 assert(tinst->is_loaded(), "ECBobj is not loaded");
7523
7524 // we want to do an instanceof comparison against the AESCrypt class
7525 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7526 if (!klass_AESCrypt->is_loaded()) {
7527 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7528 Node* ctrl = control();
7529 set_control(top()); // no regular fast path
7530 return ctrl;
7531 }
7532 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7533
7534 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7535 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7536 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7537
7538 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7539
7540 // for encryption, we are done
7541 if (!decrypting)
7542 return instof_false; // even if it is null
7543
7544 // for decryption, we need to add a further check to avoid
7545 // taking the intrinsic path when cipher and plain are the same
7546 // see the original java code for why.
7547 RegionNode* region = new RegionNode(3);
7548 region->init_req(1, instof_false);
7549 Node* src = argument(1);
7550 Node* dest = argument(4);
7551 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7552 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7553 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7554 region->init_req(2, src_dest_conjoint);
7555
7556 record_for_igvn(region);
7557 return _gvn.transform(region);
7558 }
7559
7560 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7561 // Return node representing slow path of predicate check.
7562 // the pseudo code we want to emulate with this predicate is:
7563 // for encryption:
7564 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7565 // for decryption:
7566 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7567 // note cipher==plain is more conservative than the original java code but that's OK
7568 //
7569
7570 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7571 // The receiver was checked for null already.
7572 Node* objCTR = argument(0);
7573
7574 // Load embeddedCipher field of CipherBlockChaining object.
7575 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7576
7577 // get AESCrypt klass for instanceOf check
7578 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7579 // will have same classloader as CipherBlockChaining object
7580 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7581 assert(tinst != nullptr, "CTRobj is null");
7582 assert(tinst->is_loaded(), "CTRobj is not loaded");
7583
7584 // we want to do an instanceof comparison against the AESCrypt class
7585 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7586 if (!klass_AESCrypt->is_loaded()) {
7587 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7588 Node* ctrl = control();
7589 set_control(top()); // no regular fast path
7590 return ctrl;
7591 }
7592
7593 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7594 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7595 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7596 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7597 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7598
7599 return instof_false; // even if it is null
7600 }
7601
7602 //------------------------------inline_ghash_processBlocks
7603 bool LibraryCallKit::inline_ghash_processBlocks() {
7604 address stubAddr;
7605 const char *stubName;
7606 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7607
7608 stubAddr = StubRoutines::ghash_processBlocks();
7609 stubName = "ghash_processBlocks";
7610
7611 Node* data = argument(0);
7612 Node* offset = argument(1);
7613 Node* len = argument(2);
7614 Node* state = argument(3);
7615 Node* subkeyH = argument(4);
7616
7617 state = must_be_not_null(state, true);
7618 subkeyH = must_be_not_null(subkeyH, true);
7619 data = must_be_not_null(data, true);
7620
7621 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7622 assert(state_start, "state is null");
7623 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7624 assert(subkeyH_start, "subkeyH is null");
7625 Node* data_start = array_element_address(data, offset, T_BYTE);
7626 assert(data_start, "data is null");
7627
7628 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7629 OptoRuntime::ghash_processBlocks_Type(),
7630 stubAddr, stubName, TypePtr::BOTTOM,
7631 state_start, subkeyH_start, data_start, len);
7632 return true;
7633 }
7634
7635 //------------------------------inline_chacha20Block
7636 bool LibraryCallKit::inline_chacha20Block() {
7637 address stubAddr;
7638 const char *stubName;
7639 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7640
7641 stubAddr = StubRoutines::chacha20Block();
7642 stubName = "chacha20Block";
7643
7644 Node* state = argument(0);
7645 Node* result = argument(1);
7646
7647 state = must_be_not_null(state, true);
7648 result = must_be_not_null(result, true);
7649
7650 Node* state_start = array_element_address(state, intcon(0), T_INT);
7651 assert(state_start, "state is null");
7652 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7653 assert(result_start, "result is null");
7654
7655 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7656 OptoRuntime::chacha20Block_Type(),
7657 stubAddr, stubName, TypePtr::BOTTOM,
7658 state_start, result_start);
7659 // return key stream length (int)
7660 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7661 set_result(retvalue);
7662 return true;
7663 }
7664
7665 //------------------------------inline_kyberNtt
7666 bool LibraryCallKit::inline_kyberNtt() {
7667 address stubAddr;
7668 const char *stubName;
7669 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7670 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7671
7672 stubAddr = StubRoutines::kyberNtt();
7673 stubName = "kyberNtt";
7674 if (!stubAddr) return false;
7675
7676 Node* coeffs = argument(0);
7677 Node* ntt_zetas = argument(1);
7678
7679 coeffs = must_be_not_null(coeffs, true);
7680 ntt_zetas = must_be_not_null(ntt_zetas, true);
7681
7682 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7683 assert(coeffs_start, "coeffs is null");
7684 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7685 assert(ntt_zetas_start, "ntt_zetas is null");
7686 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7687 OptoRuntime::kyberNtt_Type(),
7688 stubAddr, stubName, TypePtr::BOTTOM,
7689 coeffs_start, ntt_zetas_start);
7690 // return an int
7691 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7692 set_result(retvalue);
7693 return true;
7694 }
7695
7696 //------------------------------inline_kyberInverseNtt
7697 bool LibraryCallKit::inline_kyberInverseNtt() {
7698 address stubAddr;
7699 const char *stubName;
7700 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7701 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7702
7703 stubAddr = StubRoutines::kyberInverseNtt();
7704 stubName = "kyberInverseNtt";
7705 if (!stubAddr) return false;
7706
7707 Node* coeffs = argument(0);
7708 Node* zetas = argument(1);
7709
7710 coeffs = must_be_not_null(coeffs, true);
7711 zetas = must_be_not_null(zetas, true);
7712
7713 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7714 assert(coeffs_start, "coeffs is null");
7715 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7716 assert(zetas_start, "inverseNtt_zetas is null");
7717 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7718 OptoRuntime::kyberInverseNtt_Type(),
7719 stubAddr, stubName, TypePtr::BOTTOM,
7720 coeffs_start, zetas_start);
7721
7722 // return an int
7723 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7724 set_result(retvalue);
7725 return true;
7726 }
7727
7728 //------------------------------inline_kyberNttMult
7729 bool LibraryCallKit::inline_kyberNttMult() {
7730 address stubAddr;
7731 const char *stubName;
7732 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7733 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7734
7735 stubAddr = StubRoutines::kyberNttMult();
7736 stubName = "kyberNttMult";
7737 if (!stubAddr) return false;
7738
7739 Node* result = argument(0);
7740 Node* ntta = argument(1);
7741 Node* nttb = argument(2);
7742 Node* zetas = argument(3);
7743
7744 result = must_be_not_null(result, true);
7745 ntta = must_be_not_null(ntta, true);
7746 nttb = must_be_not_null(nttb, true);
7747 zetas = must_be_not_null(zetas, true);
7748
7749 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7750 assert(result_start, "result is null");
7751 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7752 assert(ntta_start, "ntta is null");
7753 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7754 assert(nttb_start, "nttb is null");
7755 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7756 assert(zetas_start, "nttMult_zetas is null");
7757 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7758 OptoRuntime::kyberNttMult_Type(),
7759 stubAddr, stubName, TypePtr::BOTTOM,
7760 result_start, ntta_start, nttb_start,
7761 zetas_start);
7762
7763 // return an int
7764 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7765 set_result(retvalue);
7766
7767 return true;
7768 }
7769
7770 //------------------------------inline_kyberAddPoly_2
7771 bool LibraryCallKit::inline_kyberAddPoly_2() {
7772 address stubAddr;
7773 const char *stubName;
7774 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7775 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7776
7777 stubAddr = StubRoutines::kyberAddPoly_2();
7778 stubName = "kyberAddPoly_2";
7779 if (!stubAddr) return false;
7780
7781 Node* result = argument(0);
7782 Node* a = argument(1);
7783 Node* b = argument(2);
7784
7785 result = must_be_not_null(result, true);
7786 a = must_be_not_null(a, true);
7787 b = must_be_not_null(b, true);
7788
7789 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7790 assert(result_start, "result is null");
7791 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7792 assert(a_start, "a is null");
7793 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7794 assert(b_start, "b is null");
7795 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7796 OptoRuntime::kyberAddPoly_2_Type(),
7797 stubAddr, stubName, TypePtr::BOTTOM,
7798 result_start, a_start, b_start);
7799 // return an int
7800 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7801 set_result(retvalue);
7802 return true;
7803 }
7804
7805 //------------------------------inline_kyberAddPoly_3
7806 bool LibraryCallKit::inline_kyberAddPoly_3() {
7807 address stubAddr;
7808 const char *stubName;
7809 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7810 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7811
7812 stubAddr = StubRoutines::kyberAddPoly_3();
7813 stubName = "kyberAddPoly_3";
7814 if (!stubAddr) return false;
7815
7816 Node* result = argument(0);
7817 Node* a = argument(1);
7818 Node* b = argument(2);
7819 Node* c = argument(3);
7820
7821 result = must_be_not_null(result, true);
7822 a = must_be_not_null(a, true);
7823 b = must_be_not_null(b, true);
7824 c = must_be_not_null(c, true);
7825
7826 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7827 assert(result_start, "result is null");
7828 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7829 assert(a_start, "a is null");
7830 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7831 assert(b_start, "b is null");
7832 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7833 assert(c_start, "c is null");
7834 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7835 OptoRuntime::kyberAddPoly_3_Type(),
7836 stubAddr, stubName, TypePtr::BOTTOM,
7837 result_start, a_start, b_start, c_start);
7838 // return an int
7839 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7840 set_result(retvalue);
7841 return true;
7842 }
7843
7844 //------------------------------inline_kyber12To16
7845 bool LibraryCallKit::inline_kyber12To16() {
7846 address stubAddr;
7847 const char *stubName;
7848 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7849 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7850
7851 stubAddr = StubRoutines::kyber12To16();
7852 stubName = "kyber12To16";
7853 if (!stubAddr) return false;
7854
7855 Node* condensed = argument(0);
7856 Node* condensedOffs = argument(1);
7857 Node* parsed = argument(2);
7858 Node* parsedLength = argument(3);
7859
7860 condensed = must_be_not_null(condensed, true);
7861 parsed = must_be_not_null(parsed, true);
7862
7863 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7864 assert(condensed_start, "condensed is null");
7865 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7866 assert(parsed_start, "parsed is null");
7867 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7868 OptoRuntime::kyber12To16_Type(),
7869 stubAddr, stubName, TypePtr::BOTTOM,
7870 condensed_start, condensedOffs, parsed_start, parsedLength);
7871 // return an int
7872 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7873 set_result(retvalue);
7874 return true;
7875
7876 }
7877
7878 //------------------------------inline_kyberBarrettReduce
7879 bool LibraryCallKit::inline_kyberBarrettReduce() {
7880 address stubAddr;
7881 const char *stubName;
7882 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7883 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7884
7885 stubAddr = StubRoutines::kyberBarrettReduce();
7886 stubName = "kyberBarrettReduce";
7887 if (!stubAddr) return false;
7888
7889 Node* coeffs = argument(0);
7890
7891 coeffs = must_be_not_null(coeffs, true);
7892
7893 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7894 assert(coeffs_start, "coeffs is null");
7895 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7896 OptoRuntime::kyberBarrettReduce_Type(),
7897 stubAddr, stubName, TypePtr::BOTTOM,
7898 coeffs_start);
7899 // return an int
7900 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7901 set_result(retvalue);
7902 return true;
7903 }
7904
7905 //------------------------------inline_dilithiumAlmostNtt
7906 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7907 address stubAddr;
7908 const char *stubName;
7909 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7910 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7911
7912 stubAddr = StubRoutines::dilithiumAlmostNtt();
7913 stubName = "dilithiumAlmostNtt";
7914 if (!stubAddr) return false;
7915
7916 Node* coeffs = argument(0);
7917 Node* ntt_zetas = argument(1);
7918
7919 coeffs = must_be_not_null(coeffs, true);
7920 ntt_zetas = must_be_not_null(ntt_zetas, true);
7921
7922 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7923 assert(coeffs_start, "coeffs is null");
7924 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7925 assert(ntt_zetas_start, "ntt_zetas is null");
7926 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7927 OptoRuntime::dilithiumAlmostNtt_Type(),
7928 stubAddr, stubName, TypePtr::BOTTOM,
7929 coeffs_start, ntt_zetas_start);
7930 // return an int
7931 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7932 set_result(retvalue);
7933 return true;
7934 }
7935
7936 //------------------------------inline_dilithiumAlmostInverseNtt
7937 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7938 address stubAddr;
7939 const char *stubName;
7940 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7941 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7942
7943 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7944 stubName = "dilithiumAlmostInverseNtt";
7945 if (!stubAddr) return false;
7946
7947 Node* coeffs = argument(0);
7948 Node* zetas = argument(1);
7949
7950 coeffs = must_be_not_null(coeffs, true);
7951 zetas = must_be_not_null(zetas, true);
7952
7953 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7954 assert(coeffs_start, "coeffs is null");
7955 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
7956 assert(zetas_start, "inverseNtt_zetas is null");
7957 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7958 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
7959 stubAddr, stubName, TypePtr::BOTTOM,
7960 coeffs_start, zetas_start);
7961 // return an int
7962 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
7963 set_result(retvalue);
7964 return true;
7965 }
7966
7967 //------------------------------inline_dilithiumNttMult
7968 bool LibraryCallKit::inline_dilithiumNttMult() {
7969 address stubAddr;
7970 const char *stubName;
7971 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7972 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
7973
7974 stubAddr = StubRoutines::dilithiumNttMult();
7975 stubName = "dilithiumNttMult";
7976 if (!stubAddr) return false;
7977
7978 Node* result = argument(0);
7979 Node* ntta = argument(1);
7980 Node* nttb = argument(2);
7981 Node* zetas = argument(3);
7982
7983 result = must_be_not_null(result, true);
7984 ntta = must_be_not_null(ntta, true);
7985 nttb = must_be_not_null(nttb, true);
7986 zetas = must_be_not_null(zetas, true);
7987
7988 Node* result_start = array_element_address(result, intcon(0), T_INT);
7989 assert(result_start, "result is null");
7990 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
7991 assert(ntta_start, "ntta is null");
7992 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
7993 assert(nttb_start, "nttb is null");
7994 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7995 OptoRuntime::dilithiumNttMult_Type(),
7996 stubAddr, stubName, TypePtr::BOTTOM,
7997 result_start, ntta_start, nttb_start);
7998
7999 // return an int
8000 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8001 set_result(retvalue);
8002
8003 return true;
8004 }
8005
8006 //------------------------------inline_dilithiumMontMulByConstant
8007 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8008 address stubAddr;
8009 const char *stubName;
8010 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8011 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8012
8013 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8014 stubName = "dilithiumMontMulByConstant";
8015 if (!stubAddr) return false;
8016
8017 Node* coeffs = argument(0);
8018 Node* constant = argument(1);
8019
8020 coeffs = must_be_not_null(coeffs, true);
8021
8022 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8023 assert(coeffs_start, "coeffs is null");
8024 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8025 OptoRuntime::dilithiumMontMulByConstant_Type(),
8026 stubAddr, stubName, TypePtr::BOTTOM,
8027 coeffs_start, constant);
8028
8029 // return an int
8030 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8031 set_result(retvalue);
8032 return true;
8033 }
8034
8035
8036 //------------------------------inline_dilithiumDecomposePoly
8037 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8038 address stubAddr;
8039 const char *stubName;
8040 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8041 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8042
8043 stubAddr = StubRoutines::dilithiumDecomposePoly();
8044 stubName = "dilithiumDecomposePoly";
8045 if (!stubAddr) return false;
8046
8047 Node* input = argument(0);
8048 Node* lowPart = argument(1);
8049 Node* highPart = argument(2);
8050 Node* twoGamma2 = argument(3);
8051 Node* multiplier = argument(4);
8052
8053 input = must_be_not_null(input, true);
8054 lowPart = must_be_not_null(lowPart, true);
8055 highPart = must_be_not_null(highPart, true);
8056
8057 Node* input_start = array_element_address(input, intcon(0), T_INT);
8058 assert(input_start, "input is null");
8059 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8060 assert(lowPart_start, "lowPart is null");
8061 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8062 assert(highPart_start, "highPart is null");
8063
8064 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8065 OptoRuntime::dilithiumDecomposePoly_Type(),
8066 stubAddr, stubName, TypePtr::BOTTOM,
8067 input_start, lowPart_start, highPart_start,
8068 twoGamma2, multiplier);
8069
8070 // return an int
8071 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8072 set_result(retvalue);
8073 return true;
8074 }
8075
8076 bool LibraryCallKit::inline_base64_encodeBlock() {
8077 address stubAddr;
8078 const char *stubName;
8079 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8080 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8081 stubAddr = StubRoutines::base64_encodeBlock();
8082 stubName = "encodeBlock";
8083
8084 if (!stubAddr) return false;
8085 Node* base64obj = argument(0);
8086 Node* src = argument(1);
8087 Node* offset = argument(2);
8088 Node* len = argument(3);
8089 Node* dest = argument(4);
8090 Node* dp = argument(5);
8091 Node* isURL = argument(6);
8092
8093 src = must_be_not_null(src, true);
8094 dest = must_be_not_null(dest, true);
8095
8096 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8097 assert(src_start, "source array is null");
8098 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8099 assert(dest_start, "destination array is null");
8100
8101 Node* base64 = make_runtime_call(RC_LEAF,
8102 OptoRuntime::base64_encodeBlock_Type(),
8103 stubAddr, stubName, TypePtr::BOTTOM,
8104 src_start, offset, len, dest_start, dp, isURL);
8105 return true;
8106 }
8107
8108 bool LibraryCallKit::inline_base64_decodeBlock() {
8109 address stubAddr;
8110 const char *stubName;
8111 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8112 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8113 stubAddr = StubRoutines::base64_decodeBlock();
8114 stubName = "decodeBlock";
8115
8116 if (!stubAddr) return false;
8117 Node* base64obj = argument(0);
8118 Node* src = argument(1);
8119 Node* src_offset = argument(2);
8120 Node* len = argument(3);
8121 Node* dest = argument(4);
8122 Node* dest_offset = argument(5);
8123 Node* isURL = argument(6);
8124 Node* isMIME = argument(7);
8125
8126 src = must_be_not_null(src, true);
8127 dest = must_be_not_null(dest, true);
8128
8129 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8130 assert(src_start, "source array is null");
8131 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8132 assert(dest_start, "destination array is null");
8133
8134 Node* call = make_runtime_call(RC_LEAF,
8135 OptoRuntime::base64_decodeBlock_Type(),
8136 stubAddr, stubName, TypePtr::BOTTOM,
8137 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8138 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8139 set_result(result);
8140 return true;
8141 }
8142
8143 bool LibraryCallKit::inline_poly1305_processBlocks() {
8144 address stubAddr;
8145 const char *stubName;
8146 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8147 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8148 stubAddr = StubRoutines::poly1305_processBlocks();
8149 stubName = "poly1305_processBlocks";
8150
8151 if (!stubAddr) return false;
8152 null_check_receiver(); // null-check receiver
8153 if (stopped()) return true;
8154
8155 Node* input = argument(1);
8156 Node* input_offset = argument(2);
8157 Node* len = argument(3);
8158 Node* alimbs = argument(4);
8159 Node* rlimbs = argument(5);
8160
8161 input = must_be_not_null(input, true);
8162 alimbs = must_be_not_null(alimbs, true);
8163 rlimbs = must_be_not_null(rlimbs, true);
8164
8165 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8166 assert(input_start, "input array is null");
8167 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8168 assert(acc_start, "acc array is null");
8169 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8170 assert(r_start, "r array is null");
8171
8172 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8173 OptoRuntime::poly1305_processBlocks_Type(),
8174 stubAddr, stubName, TypePtr::BOTTOM,
8175 input_start, len, acc_start, r_start);
8176 return true;
8177 }
8178
8179 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8180 address stubAddr;
8181 const char *stubName;
8182 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8183 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8184 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8185 stubName = "intpoly_montgomeryMult_P256";
8186
8187 if (!stubAddr) return false;
8188 null_check_receiver(); // null-check receiver
8189 if (stopped()) return true;
8190
8191 Node* a = argument(1);
8192 Node* b = argument(2);
8193 Node* r = argument(3);
8194
8195 a = must_be_not_null(a, true);
8196 b = must_be_not_null(b, true);
8197 r = must_be_not_null(r, true);
8198
8199 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8200 assert(a_start, "a array is null");
8201 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8202 assert(b_start, "b array is null");
8203 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8204 assert(r_start, "r array is null");
8205
8206 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8207 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8208 stubAddr, stubName, TypePtr::BOTTOM,
8209 a_start, b_start, r_start);
8210 return true;
8211 }
8212
8213 bool LibraryCallKit::inline_intpoly_assign() {
8214 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8215 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8216 const char *stubName = "intpoly_assign";
8217 address stubAddr = StubRoutines::intpoly_assign();
8218 if (!stubAddr) return false;
8219
8220 Node* set = argument(0);
8221 Node* a = argument(1);
8222 Node* b = argument(2);
8223 Node* arr_length = load_array_length(a);
8224
8225 a = must_be_not_null(a, true);
8226 b = must_be_not_null(b, true);
8227
8228 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8229 assert(a_start, "a array is null");
8230 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8231 assert(b_start, "b array is null");
8232
8233 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8234 OptoRuntime::intpoly_assign_Type(),
8235 stubAddr, stubName, TypePtr::BOTTOM,
8236 set, a_start, b_start, arr_length);
8237 return true;
8238 }
8239
8240 //------------------------------inline_digestBase_implCompress-----------------------
8241 //
8242 // Calculate MD5 for single-block byte[] array.
8243 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8244 //
8245 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8246 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8247 //
8248 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8249 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8250 //
8251 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8252 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8253 //
8254 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8255 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8256 //
8257 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8258 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8259
8260 Node* digestBase_obj = argument(0);
8261 Node* src = argument(1); // type oop
8262 Node* ofs = argument(2); // type int
8263
8264 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8265 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8266 // failed array check
8267 return false;
8268 }
8269 // Figure out the size and type of the elements we will be copying.
8270 BasicType src_elem = src_type->elem()->array_element_basic_type();
8271 if (src_elem != T_BYTE) {
8272 return false;
8273 }
8274 // 'src_start' points to src array + offset
8275 src = must_be_not_null(src, true);
8276 Node* src_start = array_element_address(src, ofs, src_elem);
8277 Node* state = nullptr;
8278 Node* block_size = nullptr;
8279 address stubAddr;
8280 const char *stubName;
8281
8282 switch(id) {
8283 case vmIntrinsics::_md5_implCompress:
8284 assert(UseMD5Intrinsics, "need MD5 instruction support");
8285 state = get_state_from_digest_object(digestBase_obj, T_INT);
8286 stubAddr = StubRoutines::md5_implCompress();
8287 stubName = "md5_implCompress";
8288 break;
8289 case vmIntrinsics::_sha_implCompress:
8290 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8291 state = get_state_from_digest_object(digestBase_obj, T_INT);
8292 stubAddr = StubRoutines::sha1_implCompress();
8293 stubName = "sha1_implCompress";
8294 break;
8295 case vmIntrinsics::_sha2_implCompress:
8296 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8297 state = get_state_from_digest_object(digestBase_obj, T_INT);
8298 stubAddr = StubRoutines::sha256_implCompress();
8299 stubName = "sha256_implCompress";
8300 break;
8301 case vmIntrinsics::_sha5_implCompress:
8302 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8303 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8304 stubAddr = StubRoutines::sha512_implCompress();
8305 stubName = "sha512_implCompress";
8306 break;
8307 case vmIntrinsics::_sha3_implCompress:
8308 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8309 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8310 stubAddr = StubRoutines::sha3_implCompress();
8311 stubName = "sha3_implCompress";
8312 block_size = get_block_size_from_digest_object(digestBase_obj);
8313 if (block_size == nullptr) return false;
8314 break;
8315 default:
8316 fatal_unexpected_iid(id);
8317 return false;
8318 }
8319 if (state == nullptr) return false;
8320
8321 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8322 if (stubAddr == nullptr) return false;
8323
8324 // Call the stub.
8325 Node* call;
8326 if (block_size == nullptr) {
8327 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8328 stubAddr, stubName, TypePtr::BOTTOM,
8329 src_start, state);
8330 } else {
8331 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8332 stubAddr, stubName, TypePtr::BOTTOM,
8333 src_start, state, block_size);
8334 }
8335
8336 return true;
8337 }
8338
8339 //------------------------------inline_double_keccak
8340 bool LibraryCallKit::inline_double_keccak() {
8341 address stubAddr;
8342 const char *stubName;
8343 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8344 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8345
8346 stubAddr = StubRoutines::double_keccak();
8347 stubName = "double_keccak";
8348 if (!stubAddr) return false;
8349
8350 Node* status0 = argument(0);
8351 Node* status1 = argument(1);
8352
8353 status0 = must_be_not_null(status0, true);
8354 status1 = must_be_not_null(status1, true);
8355
8356 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8357 assert(status0_start, "status0 is null");
8358 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8359 assert(status1_start, "status1 is null");
8360 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8361 OptoRuntime::double_keccak_Type(),
8362 stubAddr, stubName, TypePtr::BOTTOM,
8363 status0_start, status1_start);
8364 // return an int
8365 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8366 set_result(retvalue);
8367 return true;
8368 }
8369
8370
8371 //------------------------------inline_digestBase_implCompressMB-----------------------
8372 //
8373 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8374 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8375 //
8376 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8377 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8378 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8379 assert((uint)predicate < 5, "sanity");
8380 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8381
8382 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8383 Node* src = argument(1); // byte[] array
8384 Node* ofs = argument(2); // type int
8385 Node* limit = argument(3); // type int
8386
8387 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8388 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8389 // failed array check
8390 return false;
8391 }
8392 // Figure out the size and type of the elements we will be copying.
8393 BasicType src_elem = src_type->elem()->array_element_basic_type();
8394 if (src_elem != T_BYTE) {
8395 return false;
8396 }
8397 // 'src_start' points to src array + offset
8398 src = must_be_not_null(src, false);
8399 Node* src_start = array_element_address(src, ofs, src_elem);
8400
8401 const char* klass_digestBase_name = nullptr;
8402 const char* stub_name = nullptr;
8403 address stub_addr = nullptr;
8404 BasicType elem_type = T_INT;
8405
8406 switch (predicate) {
8407 case 0:
8408 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8409 klass_digestBase_name = "sun/security/provider/MD5";
8410 stub_name = "md5_implCompressMB";
8411 stub_addr = StubRoutines::md5_implCompressMB();
8412 }
8413 break;
8414 case 1:
8415 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8416 klass_digestBase_name = "sun/security/provider/SHA";
8417 stub_name = "sha1_implCompressMB";
8418 stub_addr = StubRoutines::sha1_implCompressMB();
8419 }
8420 break;
8421 case 2:
8422 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8423 klass_digestBase_name = "sun/security/provider/SHA2";
8424 stub_name = "sha256_implCompressMB";
8425 stub_addr = StubRoutines::sha256_implCompressMB();
8426 }
8427 break;
8428 case 3:
8429 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8430 klass_digestBase_name = "sun/security/provider/SHA5";
8431 stub_name = "sha512_implCompressMB";
8432 stub_addr = StubRoutines::sha512_implCompressMB();
8433 elem_type = T_LONG;
8434 }
8435 break;
8436 case 4:
8437 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8438 klass_digestBase_name = "sun/security/provider/SHA3";
8439 stub_name = "sha3_implCompressMB";
8440 stub_addr = StubRoutines::sha3_implCompressMB();
8441 elem_type = T_LONG;
8442 }
8443 break;
8444 default:
8445 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8446 }
8447 if (klass_digestBase_name != nullptr) {
8448 assert(stub_addr != nullptr, "Stub is generated");
8449 if (stub_addr == nullptr) return false;
8450
8451 // get DigestBase klass to lookup for SHA klass
8452 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8453 assert(tinst != nullptr, "digestBase_obj is not instance???");
8454 assert(tinst->is_loaded(), "DigestBase is not loaded");
8455
8456 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8457 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8458 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8459 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8460 }
8461 return false;
8462 }
8463
8464 //------------------------------inline_digestBase_implCompressMB-----------------------
8465 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8466 BasicType elem_type, address stubAddr, const char *stubName,
8467 Node* src_start, Node* ofs, Node* limit) {
8468 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8469 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8470 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8471 digest_obj = _gvn.transform(digest_obj);
8472
8473 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8474 if (state == nullptr) return false;
8475
8476 Node* block_size = nullptr;
8477 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8478 block_size = get_block_size_from_digest_object(digest_obj);
8479 if (block_size == nullptr) return false;
8480 }
8481
8482 // Call the stub.
8483 Node* call;
8484 if (block_size == nullptr) {
8485 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8486 OptoRuntime::digestBase_implCompressMB_Type(false),
8487 stubAddr, stubName, TypePtr::BOTTOM,
8488 src_start, state, ofs, limit);
8489 } else {
8490 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8491 OptoRuntime::digestBase_implCompressMB_Type(true),
8492 stubAddr, stubName, TypePtr::BOTTOM,
8493 src_start, state, block_size, ofs, limit);
8494 }
8495
8496 // return ofs (int)
8497 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8498 set_result(result);
8499
8500 return true;
8501 }
8502
8503 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8504 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8505 assert(UseAES, "need AES instruction support");
8506 address stubAddr = nullptr;
8507 const char *stubName = nullptr;
8508 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8509 stubName = "galoisCounterMode_AESCrypt";
8510
8511 if (stubAddr == nullptr) return false;
8512
8513 Node* in = argument(0);
8514 Node* inOfs = argument(1);
8515 Node* len = argument(2);
8516 Node* ct = argument(3);
8517 Node* ctOfs = argument(4);
8518 Node* out = argument(5);
8519 Node* outOfs = argument(6);
8520 Node* gctr_object = argument(7);
8521 Node* ghash_object = argument(8);
8522
8523 // (1) in, ct and out are arrays.
8524 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8525 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8526 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8527 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8528 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8529 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8530
8531 // checks are the responsibility of the caller
8532 Node* in_start = in;
8533 Node* ct_start = ct;
8534 Node* out_start = out;
8535 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8536 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8537 in_start = array_element_address(in, inOfs, T_BYTE);
8538 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8539 out_start = array_element_address(out, outOfs, T_BYTE);
8540 }
8541
8542 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8543 // (because of the predicated logic executed earlier).
8544 // so we cast it here safely.
8545 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8546 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8547 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8548 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8549 Node* state = load_field_from_object(ghash_object, "state", "[J");
8550
8551 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8552 return false;
8553 }
8554 // cast it to what we know it will be at runtime
8555 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8556 assert(tinst != nullptr, "GCTR obj is null");
8557 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8558 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8559 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8560 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8561 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8562 const TypeOopPtr* xtype = aklass->as_instance_type();
8563 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8564 aescrypt_object = _gvn.transform(aescrypt_object);
8565 // we need to get the start of the aescrypt_object's expanded key array
8566 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8567 if (k_start == nullptr) return false;
8568 // similarly, get the start address of the r vector
8569 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8570 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8571 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8572
8573
8574 // Call the stub, passing params
8575 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8576 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8577 stubAddr, stubName, TypePtr::BOTTOM,
8578 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8579
8580 // return cipher length (int)
8581 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8582 set_result(retvalue);
8583
8584 return true;
8585 }
8586
8587 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8588 // Return node representing slow path of predicate check.
8589 // the pseudo code we want to emulate with this predicate is:
8590 // for encryption:
8591 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8592 // for decryption:
8593 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8594 // note cipher==plain is more conservative than the original java code but that's OK
8595 //
8596
8597 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8598 // The receiver was checked for null already.
8599 Node* objGCTR = argument(7);
8600 // Load embeddedCipher field of GCTR object.
8601 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8602 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8603
8604 // get AESCrypt klass for instanceOf check
8605 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8606 // will have same classloader as CipherBlockChaining object
8607 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8608 assert(tinst != nullptr, "GCTR obj is null");
8609 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8610
8611 // we want to do an instanceof comparison against the AESCrypt class
8612 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8613 if (!klass_AESCrypt->is_loaded()) {
8614 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8615 Node* ctrl = control();
8616 set_control(top()); // no regular fast path
8617 return ctrl;
8618 }
8619
8620 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8621 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8622 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8623 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8624 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8625
8626 return instof_false; // even if it is null
8627 }
8628
8629 //------------------------------get_state_from_digest_object-----------------------
8630 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8631 const char* state_type;
8632 switch (elem_type) {
8633 case T_BYTE: state_type = "[B"; break;
8634 case T_INT: state_type = "[I"; break;
8635 case T_LONG: state_type = "[J"; break;
8636 default: ShouldNotReachHere();
8637 }
8638 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8639 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8640 if (digest_state == nullptr) return (Node *) nullptr;
8641
8642 // now have the array, need to get the start address of the state array
8643 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8644 return state;
8645 }
8646
8647 //------------------------------get_block_size_from_sha3_object----------------------------------
8648 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8649 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8650 assert (block_size != nullptr, "sanity");
8651 return block_size;
8652 }
8653
8654 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8655 // Return node representing slow path of predicate check.
8656 // the pseudo code we want to emulate with this predicate is:
8657 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8658 //
8659 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8660 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8661 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8662 assert((uint)predicate < 5, "sanity");
8663
8664 // The receiver was checked for null already.
8665 Node* digestBaseObj = argument(0);
8666
8667 // get DigestBase klass for instanceOf check
8668 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8669 assert(tinst != nullptr, "digestBaseObj is null");
8670 assert(tinst->is_loaded(), "DigestBase is not loaded");
8671
8672 const char* klass_name = nullptr;
8673 switch (predicate) {
8674 case 0:
8675 if (UseMD5Intrinsics) {
8676 // we want to do an instanceof comparison against the MD5 class
8677 klass_name = "sun/security/provider/MD5";
8678 }
8679 break;
8680 case 1:
8681 if (UseSHA1Intrinsics) {
8682 // we want to do an instanceof comparison against the SHA class
8683 klass_name = "sun/security/provider/SHA";
8684 }
8685 break;
8686 case 2:
8687 if (UseSHA256Intrinsics) {
8688 // we want to do an instanceof comparison against the SHA2 class
8689 klass_name = "sun/security/provider/SHA2";
8690 }
8691 break;
8692 case 3:
8693 if (UseSHA512Intrinsics) {
8694 // we want to do an instanceof comparison against the SHA5 class
8695 klass_name = "sun/security/provider/SHA5";
8696 }
8697 break;
8698 case 4:
8699 if (UseSHA3Intrinsics) {
8700 // we want to do an instanceof comparison against the SHA3 class
8701 klass_name = "sun/security/provider/SHA3";
8702 }
8703 break;
8704 default:
8705 fatal("unknown SHA intrinsic predicate: %d", predicate);
8706 }
8707
8708 ciKlass* klass = nullptr;
8709 if (klass_name != nullptr) {
8710 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8711 }
8712 if ((klass == nullptr) || !klass->is_loaded()) {
8713 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8714 Node* ctrl = control();
8715 set_control(top()); // no intrinsic path
8716 return ctrl;
8717 }
8718 ciInstanceKlass* instklass = klass->as_instance_klass();
8719
8720 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8721 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8722 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8723 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8724
8725 return instof_false; // even if it is null
8726 }
8727
8728 //-------------inline_fma-----------------------------------
8729 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8730 Node *a = nullptr;
8731 Node *b = nullptr;
8732 Node *c = nullptr;
8733 Node* result = nullptr;
8734 switch (id) {
8735 case vmIntrinsics::_fmaD:
8736 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8737 // no receiver since it is static method
8738 a = argument(0);
8739 b = argument(2);
8740 c = argument(4);
8741 result = _gvn.transform(new FmaDNode(a, b, c));
8742 break;
8743 case vmIntrinsics::_fmaF:
8744 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8745 a = argument(0);
8746 b = argument(1);
8747 c = argument(2);
8748 result = _gvn.transform(new FmaFNode(a, b, c));
8749 break;
8750 default:
8751 fatal_unexpected_iid(id); break;
8752 }
8753 set_result(result);
8754 return true;
8755 }
8756
8757 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8758 // argument(0) is receiver
8759 Node* codePoint = argument(1);
8760 Node* n = nullptr;
8761
8762 switch (id) {
8763 case vmIntrinsics::_isDigit :
8764 n = new DigitNode(control(), codePoint);
8765 break;
8766 case vmIntrinsics::_isLowerCase :
8767 n = new LowerCaseNode(control(), codePoint);
8768 break;
8769 case vmIntrinsics::_isUpperCase :
8770 n = new UpperCaseNode(control(), codePoint);
8771 break;
8772 case vmIntrinsics::_isWhitespace :
8773 n = new WhitespaceNode(control(), codePoint);
8774 break;
8775 default:
8776 fatal_unexpected_iid(id);
8777 }
8778
8779 set_result(_gvn.transform(n));
8780 return true;
8781 }
8782
8783 bool LibraryCallKit::inline_profileBoolean() {
8784 Node* counts = argument(1);
8785 const TypeAryPtr* ary = nullptr;
8786 ciArray* aobj = nullptr;
8787 if (counts->is_Con()
8788 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8789 && (aobj = ary->const_oop()->as_array()) != nullptr
8790 && (aobj->length() == 2)) {
8791 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8792 jint false_cnt = aobj->element_value(0).as_int();
8793 jint true_cnt = aobj->element_value(1).as_int();
8794
8795 if (C->log() != nullptr) {
8796 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8797 false_cnt, true_cnt);
8798 }
8799
8800 if (false_cnt + true_cnt == 0) {
8801 // According to profile, never executed.
8802 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8803 Deoptimization::Action_reinterpret);
8804 return true;
8805 }
8806
8807 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8808 // is a number of each value occurrences.
8809 Node* result = argument(0);
8810 if (false_cnt == 0 || true_cnt == 0) {
8811 // According to profile, one value has been never seen.
8812 int expected_val = (false_cnt == 0) ? 1 : 0;
8813
8814 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8815 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8816
8817 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8818 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8819 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8820
8821 { // Slow path: uncommon trap for never seen value and then reexecute
8822 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8823 // the value has been seen at least once.
8824 PreserveJVMState pjvms(this);
8825 PreserveReexecuteState preexecs(this);
8826 jvms()->set_should_reexecute(true);
8827
8828 set_control(slow_path);
8829 set_i_o(i_o());
8830
8831 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8832 Deoptimization::Action_reinterpret);
8833 }
8834 // The guard for never seen value enables sharpening of the result and
8835 // returning a constant. It allows to eliminate branches on the same value
8836 // later on.
8837 set_control(fast_path);
8838 result = intcon(expected_val);
8839 }
8840 // Stop profiling.
8841 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8842 // By replacing method body with profile data (represented as ProfileBooleanNode
8843 // on IR level) we effectively disable profiling.
8844 // It enables full speed execution once optimized code is generated.
8845 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8846 C->record_for_igvn(profile);
8847 set_result(profile);
8848 return true;
8849 } else {
8850 // Continue profiling.
8851 // Profile data isn't available at the moment. So, execute method's bytecode version.
8852 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8853 // is compiled and counters aren't available since corresponding MethodHandle
8854 // isn't a compile-time constant.
8855 return false;
8856 }
8857 }
8858
8859 bool LibraryCallKit::inline_isCompileConstant() {
8860 Node* n = argument(0);
8861 set_result(n->is_Con() ? intcon(1) : intcon(0));
8862 return true;
8863 }
8864
8865 //------------------------------- inline_getObjectSize --------------------------------------
8866 //
8867 // Calculate the runtime size of the object/array.
8868 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8869 //
8870 bool LibraryCallKit::inline_getObjectSize() {
8871 Node* obj = argument(3);
8872 Node* klass_node = load_object_klass(obj);
8873
8874 jint layout_con = Klass::_lh_neutral_value;
8875 Node* layout_val = get_layout_helper(klass_node, layout_con);
8876 int layout_is_con = (layout_val == nullptr);
8877
8878 if (layout_is_con) {
8879 // Layout helper is constant, can figure out things at compile time.
8880
8881 if (Klass::layout_helper_is_instance(layout_con)) {
8882 // Instance case: layout_con contains the size itself.
8883 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8884 set_result(size);
8885 } else {
8886 // Array case: size is round(header + element_size*arraylength).
8887 // Since arraylength is different for every array instance, we have to
8888 // compute the whole thing at runtime.
8889
8890 Node* arr_length = load_array_length(obj);
8891
8892 int round_mask = MinObjAlignmentInBytes - 1;
8893 int hsize = Klass::layout_helper_header_size(layout_con);
8894 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8895
8896 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8897 round_mask = 0; // strength-reduce it if it goes away completely
8898 }
8899 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8900 Node* header_size = intcon(hsize + round_mask);
8901
8902 Node* lengthx = ConvI2X(arr_length);
8903 Node* headerx = ConvI2X(header_size);
8904
8905 Node* abody = lengthx;
8906 if (eshift != 0) {
8907 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8908 }
8909 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8910 if (round_mask != 0) {
8911 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8912 }
8913 size = ConvX2L(size);
8914 set_result(size);
8915 }
8916 } else {
8917 // Layout helper is not constant, need to test for array-ness at runtime.
8918
8919 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8920 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8921 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8922 record_for_igvn(result_reg);
8923
8924 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8925 if (array_ctl != nullptr) {
8926 // Array case: size is round(header + element_size*arraylength).
8927 // Since arraylength is different for every array instance, we have to
8928 // compute the whole thing at runtime.
8929
8930 PreserveJVMState pjvms(this);
8931 set_control(array_ctl);
8932 Node* arr_length = load_array_length(obj);
8933
8934 int round_mask = MinObjAlignmentInBytes - 1;
8935 Node* mask = intcon(round_mask);
8936
8937 Node* hss = intcon(Klass::_lh_header_size_shift);
8938 Node* hsm = intcon(Klass::_lh_header_size_mask);
8939 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8940 header_size = _gvn.transform(new AndINode(header_size, hsm));
8941 header_size = _gvn.transform(new AddINode(header_size, mask));
8942
8943 // There is no need to mask or shift this value.
8944 // The semantics of LShiftINode include an implicit mask to 0x1F.
8945 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8946 Node* elem_shift = layout_val;
8947
8948 Node* lengthx = ConvI2X(arr_length);
8949 Node* headerx = ConvI2X(header_size);
8950
8951 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8952 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8953 if (round_mask != 0) {
8954 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8955 }
8956 size = ConvX2L(size);
8957
8958 result_reg->init_req(_array_path, control());
8959 result_val->init_req(_array_path, size);
8960 }
8961
8962 if (!stopped()) {
8963 // Instance case: the layout helper gives us instance size almost directly,
8964 // but we need to mask out the _lh_instance_slow_path_bit.
8965 Node* size = ConvI2X(layout_val);
8966 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8967 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8968 size = _gvn.transform(new AndXNode(size, mask));
8969 size = ConvX2L(size);
8970
8971 result_reg->init_req(_instance_path, control());
8972 result_val->init_req(_instance_path, size);
8973 }
8974
8975 set_result(result_reg, result_val);
8976 }
8977
8978 return true;
8979 }
8980
8981 //------------------------------- inline_blackhole --------------------------------------
8982 //
8983 // Make sure all arguments to this node are alive.
8984 // This matches methods that were requested to be blackholed through compile commands.
8985 //
8986 bool LibraryCallKit::inline_blackhole() {
8987 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8988 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8989 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8990
8991 // Blackhole node pinches only the control, not memory. This allows
8992 // the blackhole to be pinned in the loop that computes blackholed
8993 // values, but have no other side effects, like breaking the optimizations
8994 // across the blackhole.
8995
8996 Node* bh = _gvn.transform(new BlackholeNode(control()));
8997 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8998
8999 // Bind call arguments as blackhole arguments to keep them alive
9000 uint nargs = callee()->arg_size();
9001 for (uint i = 0; i < nargs; i++) {
9002 bh->add_req(argument(i));
9003 }
9004
9005 return true;
9006 }
9007
9008 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9009 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9010 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9011 return nullptr; // box klass is not Float16
9012 }
9013
9014 // Null check; get notnull casted pointer
9015 Node* null_ctl = top();
9016 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9017 // If not_null_box is dead, only null-path is taken
9018 if (stopped()) {
9019 set_control(null_ctl);
9020 return nullptr;
9021 }
9022 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9023 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9024 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9025 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9026 }
9027
9028 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9029 PreserveReexecuteState preexecs(this);
9030 jvms()->set_should_reexecute(true);
9031
9032 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9033 Node* klass_node = makecon(klass_type);
9034 Node* box = new_instance(klass_node);
9035
9036 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9037 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9038
9039 Node* field_store = _gvn.transform(access_store_at(box,
9040 value_field,
9041 value_adr_type,
9042 value,
9043 TypeInt::SHORT,
9044 T_SHORT,
9045 IN_HEAP));
9046 set_memory(field_store, value_adr_type);
9047 return box;
9048 }
9049
9050 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9051 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9052 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9053 return false;
9054 }
9055
9056 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9057 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9058 return false;
9059 }
9060
9061 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9062 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9063 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9064 ciSymbols::short_signature(),
9065 false);
9066 assert(field != nullptr, "");
9067
9068 // Transformed nodes
9069 Node* fld1 = nullptr;
9070 Node* fld2 = nullptr;
9071 Node* fld3 = nullptr;
9072 switch(num_args) {
9073 case 3:
9074 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9075 if (fld3 == nullptr) {
9076 return false;
9077 }
9078 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9079 // fall-through
9080 case 2:
9081 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9082 if (fld2 == nullptr) {
9083 return false;
9084 }
9085 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9086 // fall-through
9087 case 1:
9088 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9089 if (fld1 == nullptr) {
9090 return false;
9091 }
9092 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9093 break;
9094 default: fatal("Unsupported number of arguments %d", num_args);
9095 }
9096
9097 Node* result = nullptr;
9098 switch (id) {
9099 // Unary operations
9100 case vmIntrinsics::_sqrt_float16:
9101 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9102 break;
9103 // Ternary operations
9104 case vmIntrinsics::_fma_float16:
9105 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9106 break;
9107 default:
9108 fatal_unexpected_iid(id);
9109 break;
9110 }
9111 result = _gvn.transform(new ReinterpretHF2SNode(result));
9112 set_result(box_fp16_value(float16_box_type, field, result));
9113 return true;
9114 }
9115