1 /*
2 * Copyright (c) 1999, 2024, 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 "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dtanh:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
487 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
488
489 #if INCLUDE_JVMTI
490 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
491 "notifyJvmtiStart", true, false);
492 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
493 "notifyJvmtiEnd", false, true);
494 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
495 "notifyJvmtiMount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
497 "notifyJvmtiUnmount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
499 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
500 #endif
501
502 #ifdef JFR_HAVE_INTRINSICS
503 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
504 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
505 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
506 #endif
507 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
508 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
509 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
510 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
511 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
512 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
513 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
514 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
515 case vmIntrinsics::_getLength: return inline_native_getLength();
516 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
517 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
518 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
519 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
520 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
521 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
522 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
523
524 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
525 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
526
527 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
528
529 case vmIntrinsics::_isInstance:
530 case vmIntrinsics::_getModifiers:
531 case vmIntrinsics::_isInterface:
532 case vmIntrinsics::_isArray:
533 case vmIntrinsics::_isPrimitive:
534 case vmIntrinsics::_isHidden:
535 case vmIntrinsics::_getSuperclass:
536 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
537
538 case vmIntrinsics::_floatToRawIntBits:
539 case vmIntrinsics::_floatToIntBits:
540 case vmIntrinsics::_intBitsToFloat:
541 case vmIntrinsics::_doubleToRawLongBits:
542 case vmIntrinsics::_doubleToLongBits:
543 case vmIntrinsics::_longBitsToDouble:
544 case vmIntrinsics::_floatToFloat16:
545 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
546
547 case vmIntrinsics::_floatIsFinite:
548 case vmIntrinsics::_floatIsInfinite:
549 case vmIntrinsics::_doubleIsFinite:
550 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
551
552 case vmIntrinsics::_numberOfLeadingZeros_i:
553 case vmIntrinsics::_numberOfLeadingZeros_l:
554 case vmIntrinsics::_numberOfTrailingZeros_i:
555 case vmIntrinsics::_numberOfTrailingZeros_l:
556 case vmIntrinsics::_bitCount_i:
557 case vmIntrinsics::_bitCount_l:
558 case vmIntrinsics::_reverse_i:
559 case vmIntrinsics::_reverse_l:
560 case vmIntrinsics::_reverseBytes_i:
561 case vmIntrinsics::_reverseBytes_l:
562 case vmIntrinsics::_reverseBytes_s:
563 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
564
565 case vmIntrinsics::_compress_i:
566 case vmIntrinsics::_compress_l:
567 case vmIntrinsics::_expand_i:
568 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
569
570 case vmIntrinsics::_compareUnsigned_i:
571 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
572
573 case vmIntrinsics::_divideUnsigned_i:
574 case vmIntrinsics::_divideUnsigned_l:
575 case vmIntrinsics::_remainderUnsigned_i:
576 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
577
578 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
579
580 case vmIntrinsics::_Reference_get: return inline_reference_get();
581 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
582 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
583 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
584 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
585
586 case vmIntrinsics::_Class_cast: return inline_Class_cast();
587
588 case vmIntrinsics::_aescrypt_encryptBlock:
589 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
590
591 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
592 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
593 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
596 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
597 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
598
599 case vmIntrinsics::_counterMode_AESCrypt:
600 return inline_counterMode_AESCrypt(intrinsic_id());
601
602 case vmIntrinsics::_galoisCounterMode_AESCrypt:
603 return inline_galoisCounterMode_AESCrypt();
604
605 case vmIntrinsics::_md5_implCompress:
606 case vmIntrinsics::_sha_implCompress:
607 case vmIntrinsics::_sha2_implCompress:
608 case vmIntrinsics::_sha5_implCompress:
609 case vmIntrinsics::_sha3_implCompress:
610 return inline_digestBase_implCompress(intrinsic_id());
611
612 case vmIntrinsics::_digestBase_implCompressMB:
613 return inline_digestBase_implCompressMB(predicate);
614
615 case vmIntrinsics::_multiplyToLen:
616 return inline_multiplyToLen();
617
618 case vmIntrinsics::_squareToLen:
619 return inline_squareToLen();
620
621 case vmIntrinsics::_mulAdd:
622 return inline_mulAdd();
623
624 case vmIntrinsics::_montgomeryMultiply:
625 return inline_montgomeryMultiply();
626 case vmIntrinsics::_montgomerySquare:
627 return inline_montgomerySquare();
628
629 case vmIntrinsics::_bigIntegerRightShiftWorker:
630 return inline_bigIntegerShift(true);
631 case vmIntrinsics::_bigIntegerLeftShiftWorker:
632 return inline_bigIntegerShift(false);
633
634 case vmIntrinsics::_vectorizedMismatch:
635 return inline_vectorizedMismatch();
636
637 case vmIntrinsics::_ghash_processBlocks:
638 return inline_ghash_processBlocks();
639 case vmIntrinsics::_chacha20Block:
640 return inline_chacha20Block();
641 case vmIntrinsics::_base64_encodeBlock:
642 return inline_base64_encodeBlock();
643 case vmIntrinsics::_base64_decodeBlock:
644 return inline_base64_decodeBlock();
645 case vmIntrinsics::_poly1305_processBlocks:
646 return inline_poly1305_processBlocks();
647 case vmIntrinsics::_intpoly_montgomeryMult_P256:
648 return inline_intpoly_montgomeryMult_P256();
649 case vmIntrinsics::_intpoly_assign:
650 return inline_intpoly_assign();
651 case vmIntrinsics::_encodeISOArray:
652 case vmIntrinsics::_encodeByteISOArray:
653 return inline_encodeISOArray(false);
654 case vmIntrinsics::_encodeAsciiArray:
655 return inline_encodeISOArray(true);
656
657 case vmIntrinsics::_updateCRC32:
658 return inline_updateCRC32();
659 case vmIntrinsics::_updateBytesCRC32:
660 return inline_updateBytesCRC32();
661 case vmIntrinsics::_updateByteBufferCRC32:
662 return inline_updateByteBufferCRC32();
663
664 case vmIntrinsics::_updateBytesCRC32C:
665 return inline_updateBytesCRC32C();
666 case vmIntrinsics::_updateDirectByteBufferCRC32C:
667 return inline_updateDirectByteBufferCRC32C();
668
669 case vmIntrinsics::_updateBytesAdler32:
670 return inline_updateBytesAdler32();
671 case vmIntrinsics::_updateByteBufferAdler32:
672 return inline_updateByteBufferAdler32();
673
674 case vmIntrinsics::_profileBoolean:
675 return inline_profileBoolean();
676 case vmIntrinsics::_isCompileConstant:
677 return inline_isCompileConstant();
678
679 case vmIntrinsics::_countPositives:
680 return inline_countPositives();
681
682 case vmIntrinsics::_fmaD:
683 case vmIntrinsics::_fmaF:
684 return inline_fma(intrinsic_id());
685
686 case vmIntrinsics::_isDigit:
687 case vmIntrinsics::_isLowerCase:
688 case vmIntrinsics::_isUpperCase:
689 case vmIntrinsics::_isWhitespace:
690 return inline_character_compare(intrinsic_id());
691
692 case vmIntrinsics::_min:
693 case vmIntrinsics::_max:
694 case vmIntrinsics::_min_strict:
695 case vmIntrinsics::_max_strict:
696 return inline_min_max(intrinsic_id());
697
698 case vmIntrinsics::_maxF:
699 case vmIntrinsics::_minF:
700 case vmIntrinsics::_maxD:
701 case vmIntrinsics::_minD:
702 case vmIntrinsics::_maxF_strict:
703 case vmIntrinsics::_minF_strict:
704 case vmIntrinsics::_maxD_strict:
705 case vmIntrinsics::_minD_strict:
706 return inline_fp_min_max(intrinsic_id());
707
708 case vmIntrinsics::_VectorUnaryOp:
709 return inline_vector_nary_operation(1);
710 case vmIntrinsics::_VectorBinaryOp:
711 return inline_vector_nary_operation(2);
712 case vmIntrinsics::_VectorTernaryOp:
713 return inline_vector_nary_operation(3);
714 case vmIntrinsics::_VectorFromBitsCoerced:
715 return inline_vector_frombits_coerced();
716 case vmIntrinsics::_VectorShuffleIota:
717 return inline_vector_shuffle_iota();
718 case vmIntrinsics::_VectorMaskOp:
719 return inline_vector_mask_operation();
720 case vmIntrinsics::_VectorShuffleToVector:
721 return inline_vector_shuffle_to_vector();
722 case vmIntrinsics::_VectorWrapShuffleIndexes:
723 return inline_vector_wrap_shuffle_indexes();
724 case vmIntrinsics::_VectorLoadOp:
725 return inline_vector_mem_operation(/*is_store=*/false);
726 case vmIntrinsics::_VectorLoadMaskedOp:
727 return inline_vector_mem_masked_operation(/*is_store*/false);
728 case vmIntrinsics::_VectorStoreOp:
729 return inline_vector_mem_operation(/*is_store=*/true);
730 case vmIntrinsics::_VectorStoreMaskedOp:
731 return inline_vector_mem_masked_operation(/*is_store=*/true);
732 case vmIntrinsics::_VectorGatherOp:
733 return inline_vector_gather_scatter(/*is_scatter*/ false);
734 case vmIntrinsics::_VectorScatterOp:
735 return inline_vector_gather_scatter(/*is_scatter*/ true);
736 case vmIntrinsics::_VectorReductionCoerced:
737 return inline_vector_reduction();
738 case vmIntrinsics::_VectorTest:
739 return inline_vector_test();
740 case vmIntrinsics::_VectorBlend:
741 return inline_vector_blend();
742 case vmIntrinsics::_VectorRearrange:
743 return inline_vector_rearrange();
744 case vmIntrinsics::_VectorSelectFrom:
745 return inline_vector_select_from();
746 case vmIntrinsics::_VectorCompare:
747 return inline_vector_compare();
748 case vmIntrinsics::_VectorBroadcastInt:
749 return inline_vector_broadcast_int();
750 case vmIntrinsics::_VectorConvert:
751 return inline_vector_convert();
752 case vmIntrinsics::_VectorInsert:
753 return inline_vector_insert();
754 case vmIntrinsics::_VectorExtract:
755 return inline_vector_extract();
756 case vmIntrinsics::_VectorCompressExpand:
757 return inline_vector_compress_expand();
758 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
759 return inline_vector_select_from_two_vectors();
760 case vmIntrinsics::_IndexVector:
761 return inline_index_vector();
762 case vmIntrinsics::_IndexPartiallyInUpperRange:
763 return inline_index_partially_in_upper_range();
764
765 case vmIntrinsics::_getObjectSize:
766 return inline_getObjectSize();
767
768 case vmIntrinsics::_blackhole:
769 return inline_blackhole();
770
771 default:
772 // If you get here, it may be that someone has added a new intrinsic
773 // to the list in vmIntrinsics.hpp without implementing it here.
774 #ifndef PRODUCT
775 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
776 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
777 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
778 }
779 #endif
780 return false;
781 }
782 }
783
784 Node* LibraryCallKit::try_to_predicate(int predicate) {
785 if (!jvms()->has_method()) {
786 // Root JVMState has a null method.
787 assert(map()->memory()->Opcode() == Op_Parm, "");
788 // Insert the memory aliasing node
789 set_all_memory(reset_memory());
790 }
791 assert(merged_memory(), "");
792
793 switch (intrinsic_id()) {
794 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
795 return inline_cipherBlockChaining_AESCrypt_predicate(false);
796 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
797 return inline_cipherBlockChaining_AESCrypt_predicate(true);
798 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
799 return inline_electronicCodeBook_AESCrypt_predicate(false);
800 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
801 return inline_electronicCodeBook_AESCrypt_predicate(true);
802 case vmIntrinsics::_counterMode_AESCrypt:
803 return inline_counterMode_AESCrypt_predicate();
804 case vmIntrinsics::_digestBase_implCompressMB:
805 return inline_digestBase_implCompressMB_predicate(predicate);
806 case vmIntrinsics::_galoisCounterMode_AESCrypt:
807 return inline_galoisCounterMode_AESCrypt_predicate();
808
809 default:
810 // If you get here, it may be that someone has added a new intrinsic
811 // to the list in vmIntrinsics.hpp without implementing it here.
812 #ifndef PRODUCT
813 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
814 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
815 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
816 }
817 #endif
818 Node* slow_ctl = control();
819 set_control(top()); // No fast path intrinsic
820 return slow_ctl;
821 }
822 }
823
824 //------------------------------set_result-------------------------------
825 // Helper function for finishing intrinsics.
826 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
827 record_for_igvn(region);
828 set_control(_gvn.transform(region));
829 set_result( _gvn.transform(value));
830 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
831 }
832
833 //------------------------------generate_guard---------------------------
834 // Helper function for generating guarded fast-slow graph structures.
835 // The given 'test', if true, guards a slow path. If the test fails
836 // then a fast path can be taken. (We generally hope it fails.)
837 // In all cases, GraphKit::control() is updated to the fast path.
838 // The returned value represents the control for the slow path.
839 // The return value is never 'top'; it is either a valid control
840 // or null if it is obvious that the slow path can never be taken.
841 // Also, if region and the slow control are not null, the slow edge
842 // is appended to the region.
843 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
844 if (stopped()) {
845 // Already short circuited.
846 return nullptr;
847 }
848
849 // Build an if node and its projections.
850 // If test is true we take the slow path, which we assume is uncommon.
851 if (_gvn.type(test) == TypeInt::ZERO) {
852 // The slow branch is never taken. No need to build this guard.
853 return nullptr;
854 }
855
856 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
857
858 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
859 if (if_slow == top()) {
860 // The slow branch is never taken. No need to build this guard.
861 return nullptr;
862 }
863
864 if (region != nullptr)
865 region->add_req(if_slow);
866
867 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
868 set_control(if_fast);
869
870 return if_slow;
871 }
872
873 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
874 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
875 }
876 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
877 return generate_guard(test, region, PROB_FAIR);
878 }
879
880 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
881 Node* *pos_index) {
882 if (stopped())
883 return nullptr; // already stopped
884 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
885 return nullptr; // index is already adequately typed
886 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
887 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
888 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
889 if (is_neg != nullptr && pos_index != nullptr) {
890 // Emulate effect of Parse::adjust_map_after_if.
891 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
892 (*pos_index) = _gvn.transform(ccast);
893 }
894 return is_neg;
895 }
896
897 // Make sure that 'position' is a valid limit index, in [0..length].
898 // There are two equivalent plans for checking this:
899 // A. (offset + copyLength) unsigned<= arrayLength
900 // B. offset <= (arrayLength - copyLength)
901 // We require that all of the values above, except for the sum and
902 // difference, are already known to be non-negative.
903 // Plan A is robust in the face of overflow, if offset and copyLength
904 // are both hugely positive.
905 //
906 // Plan B is less direct and intuitive, but it does not overflow at
907 // all, since the difference of two non-negatives is always
908 // representable. Whenever Java methods must perform the equivalent
909 // check they generally use Plan B instead of Plan A.
910 // For the moment we use Plan A.
911 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
912 Node* subseq_length,
913 Node* array_length,
914 RegionNode* region) {
915 if (stopped())
916 return nullptr; // already stopped
917 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
918 if (zero_offset && subseq_length->eqv_uncast(array_length))
919 return nullptr; // common case of whole-array copy
920 Node* last = subseq_length;
921 if (!zero_offset) // last += offset
922 last = _gvn.transform(new AddINode(last, offset));
923 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
924 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
925 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
926 return is_over;
927 }
928
929 // Emit range checks for the given String.value byte array
930 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
931 if (stopped()) {
932 return; // already stopped
933 }
934 RegionNode* bailout = new RegionNode(1);
935 record_for_igvn(bailout);
936 if (char_count) {
937 // Convert char count to byte count
938 count = _gvn.transform(new LShiftINode(count, intcon(1)));
939 }
940
941 // Offset and count must not be negative
942 generate_negative_guard(offset, bailout);
943 generate_negative_guard(count, bailout);
944 // Offset + count must not exceed length of array
945 generate_limit_guard(offset, count, load_array_length(array), bailout);
946
947 if (bailout->req() > 1) {
948 PreserveJVMState pjvms(this);
949 set_control(_gvn.transform(bailout));
950 uncommon_trap(Deoptimization::Reason_intrinsic,
951 Deoptimization::Action_maybe_recompile);
952 }
953 }
954
955 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
956 bool is_immutable) {
957 ciKlass* thread_klass = env()->Thread_klass();
958 const Type* thread_type
959 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
960
961 Node* thread = _gvn.transform(new ThreadLocalNode());
962 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
963 tls_output = thread;
964
965 Node* thread_obj_handle
966 = (is_immutable
967 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
968 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
969 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
970 thread_obj_handle = _gvn.transform(thread_obj_handle);
971
972 DecoratorSet decorators = IN_NATIVE;
973 if (is_immutable) {
974 decorators |= C2_IMMUTABLE_MEMORY;
975 }
976 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
977 }
978
979 //--------------------------generate_current_thread--------------------
980 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
981 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
982 /*is_immutable*/false);
983 }
984
985 //--------------------------generate_virtual_thread--------------------
986 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
987 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
988 !C->method()->changes_current_thread());
989 }
990
991 //------------------------------make_string_method_node------------------------
992 // Helper method for String intrinsic functions. This version is called with
993 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
994 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
995 // containing the lengths of str1 and str2.
996 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
997 Node* result = nullptr;
998 switch (opcode) {
999 case Op_StrIndexOf:
1000 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1001 str1_start, cnt1, str2_start, cnt2, ae);
1002 break;
1003 case Op_StrComp:
1004 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1005 str1_start, cnt1, str2_start, cnt2, ae);
1006 break;
1007 case Op_StrEquals:
1008 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1009 // Use the constant length if there is one because optimized match rule may exist.
1010 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1011 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1012 break;
1013 default:
1014 ShouldNotReachHere();
1015 return nullptr;
1016 }
1017
1018 // All these intrinsics have checks.
1019 C->set_has_split_ifs(true); // Has chance for split-if optimization
1020 clear_upper_avx();
1021
1022 return _gvn.transform(result);
1023 }
1024
1025 //------------------------------inline_string_compareTo------------------------
1026 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1027 Node* arg1 = argument(0);
1028 Node* arg2 = argument(1);
1029
1030 arg1 = must_be_not_null(arg1, true);
1031 arg2 = must_be_not_null(arg2, true);
1032
1033 // Get start addr and length of first argument
1034 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1035 Node* arg1_cnt = load_array_length(arg1);
1036
1037 // Get start addr and length of second argument
1038 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1039 Node* arg2_cnt = load_array_length(arg2);
1040
1041 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1042 set_result(result);
1043 return true;
1044 }
1045
1046 //------------------------------inline_string_equals------------------------
1047 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1048 Node* arg1 = argument(0);
1049 Node* arg2 = argument(1);
1050
1051 // paths (plus control) merge
1052 RegionNode* region = new RegionNode(3);
1053 Node* phi = new PhiNode(region, TypeInt::BOOL);
1054
1055 if (!stopped()) {
1056
1057 arg1 = must_be_not_null(arg1, true);
1058 arg2 = must_be_not_null(arg2, true);
1059
1060 // Get start addr and length of first argument
1061 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1062 Node* arg1_cnt = load_array_length(arg1);
1063
1064 // Get start addr and length of second argument
1065 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1066 Node* arg2_cnt = load_array_length(arg2);
1067
1068 // Check for arg1_cnt != arg2_cnt
1069 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1070 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1071 Node* if_ne = generate_slow_guard(bol, nullptr);
1072 if (if_ne != nullptr) {
1073 phi->init_req(2, intcon(0));
1074 region->init_req(2, if_ne);
1075 }
1076
1077 // Check for count == 0 is done by assembler code for StrEquals.
1078
1079 if (!stopped()) {
1080 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1081 phi->init_req(1, equals);
1082 region->init_req(1, control());
1083 }
1084 }
1085
1086 // post merge
1087 set_control(_gvn.transform(region));
1088 record_for_igvn(region);
1089
1090 set_result(_gvn.transform(phi));
1091 return true;
1092 }
1093
1094 //------------------------------inline_array_equals----------------------------
1095 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1096 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1097 Node* arg1 = argument(0);
1098 Node* arg2 = argument(1);
1099
1100 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1101 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1102 clear_upper_avx();
1103
1104 return true;
1105 }
1106
1107
1108 //------------------------------inline_countPositives------------------------------
1109 bool LibraryCallKit::inline_countPositives() {
1110 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1111 return false;
1112 }
1113
1114 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1115 // no receiver since it is static method
1116 Node* ba = argument(0);
1117 Node* offset = argument(1);
1118 Node* len = argument(2);
1119
1120 ba = must_be_not_null(ba, true);
1121
1122 // Range checks
1123 generate_string_range_check(ba, offset, len, false);
1124 if (stopped()) {
1125 return true;
1126 }
1127 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1128 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1129 set_result(_gvn.transform(result));
1130 clear_upper_avx();
1131 return true;
1132 }
1133
1134 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1135 Node* index = argument(0);
1136 Node* length = bt == T_INT ? argument(1) : argument(2);
1137 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1138 return false;
1139 }
1140
1141 // check that length is positive
1142 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1143 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1144
1145 {
1146 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1147 uncommon_trap(Deoptimization::Reason_intrinsic,
1148 Deoptimization::Action_make_not_entrant);
1149 }
1150
1151 if (stopped()) {
1152 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1153 return true;
1154 }
1155
1156 // length is now known positive, add a cast node to make this explicit
1157 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1158 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1159 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1160 ConstraintCastNode::RegularDependency, bt);
1161 casted_length = _gvn.transform(casted_length);
1162 replace_in_map(length, casted_length);
1163 length = casted_length;
1164
1165 // Use an unsigned comparison for the range check itself
1166 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1167 BoolTest::mask btest = BoolTest::lt;
1168 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1169 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1170 _gvn.set_type(rc, rc->Value(&_gvn));
1171 if (!rc_bool->is_Con()) {
1172 record_for_igvn(rc);
1173 }
1174 set_control(_gvn.transform(new IfTrueNode(rc)));
1175 {
1176 PreserveJVMState pjvms(this);
1177 set_control(_gvn.transform(new IfFalseNode(rc)));
1178 uncommon_trap(Deoptimization::Reason_range_check,
1179 Deoptimization::Action_make_not_entrant);
1180 }
1181
1182 if (stopped()) {
1183 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1184 return true;
1185 }
1186
1187 // index is now known to be >= 0 and < length, cast it
1188 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1189 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1190 ConstraintCastNode::RegularDependency, bt);
1191 result = _gvn.transform(result);
1192 set_result(result);
1193 replace_in_map(index, result);
1194 return true;
1195 }
1196
1197 //------------------------------inline_string_indexOf------------------------
1198 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1199 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1200 return false;
1201 }
1202 Node* src = argument(0);
1203 Node* tgt = argument(1);
1204
1205 // Make the merge point
1206 RegionNode* result_rgn = new RegionNode(4);
1207 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1208
1209 src = must_be_not_null(src, true);
1210 tgt = must_be_not_null(tgt, true);
1211
1212 // Get start addr and length of source string
1213 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1214 Node* src_count = load_array_length(src);
1215
1216 // Get start addr and length of substring
1217 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1218 Node* tgt_count = load_array_length(tgt);
1219
1220 Node* result = nullptr;
1221 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1222
1223 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1224 // Divide src size by 2 if String is UTF16 encoded
1225 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1226 }
1227 if (ae == StrIntrinsicNode::UU) {
1228 // Divide substring size by 2 if String is UTF16 encoded
1229 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1230 }
1231
1232 if (call_opt_stub) {
1233 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1234 StubRoutines::_string_indexof_array[ae],
1235 "stringIndexOf", TypePtr::BOTTOM, src_start,
1236 src_count, tgt_start, tgt_count);
1237 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1238 } else {
1239 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1240 result_rgn, result_phi, ae);
1241 }
1242 if (result != nullptr) {
1243 result_phi->init_req(3, result);
1244 result_rgn->init_req(3, control());
1245 }
1246 set_control(_gvn.transform(result_rgn));
1247 record_for_igvn(result_rgn);
1248 set_result(_gvn.transform(result_phi));
1249
1250 return true;
1251 }
1252
1253 //-----------------------------inline_string_indexOfI-----------------------
1254 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1255 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1256 return false;
1257 }
1258 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1259 return false;
1260 }
1261
1262 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1263 Node* src = argument(0); // byte[]
1264 Node* src_count = argument(1); // char count
1265 Node* tgt = argument(2); // byte[]
1266 Node* tgt_count = argument(3); // char count
1267 Node* from_index = argument(4); // char index
1268
1269 src = must_be_not_null(src, true);
1270 tgt = must_be_not_null(tgt, true);
1271
1272 // Multiply byte array index by 2 if String is UTF16 encoded
1273 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1274 src_count = _gvn.transform(new SubINode(src_count, from_index));
1275 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1276 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1277
1278 // Range checks
1279 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1280 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1281 if (stopped()) {
1282 return true;
1283 }
1284
1285 RegionNode* region = new RegionNode(5);
1286 Node* phi = new PhiNode(region, TypeInt::INT);
1287 Node* result = nullptr;
1288
1289 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1290
1291 if (call_opt_stub) {
1292 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1293 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1294 StubRoutines::_string_indexof_array[ae],
1295 "stringIndexOf", TypePtr::BOTTOM, src_start,
1296 src_count, tgt_start, tgt_count);
1297 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1298 } else {
1299 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1300 region, phi, ae);
1301 }
1302 if (result != nullptr) {
1303 // The result is index relative to from_index if substring was found, -1 otherwise.
1304 // Generate code which will fold into cmove.
1305 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1306 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1307
1308 Node* if_lt = generate_slow_guard(bol, nullptr);
1309 if (if_lt != nullptr) {
1310 // result == -1
1311 phi->init_req(3, result);
1312 region->init_req(3, if_lt);
1313 }
1314 if (!stopped()) {
1315 result = _gvn.transform(new AddINode(result, from_index));
1316 phi->init_req(4, result);
1317 region->init_req(4, control());
1318 }
1319 }
1320
1321 set_control(_gvn.transform(region));
1322 record_for_igvn(region);
1323 set_result(_gvn.transform(phi));
1324 clear_upper_avx();
1325
1326 return true;
1327 }
1328
1329 // Create StrIndexOfNode with fast path checks
1330 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1331 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1332 // Check for substr count > string count
1333 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1334 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1335 Node* if_gt = generate_slow_guard(bol, nullptr);
1336 if (if_gt != nullptr) {
1337 phi->init_req(1, intcon(-1));
1338 region->init_req(1, if_gt);
1339 }
1340 if (!stopped()) {
1341 // Check for substr count == 0
1342 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1343 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1344 Node* if_zero = generate_slow_guard(bol, nullptr);
1345 if (if_zero != nullptr) {
1346 phi->init_req(2, intcon(0));
1347 region->init_req(2, if_zero);
1348 }
1349 }
1350 if (!stopped()) {
1351 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1352 }
1353 return nullptr;
1354 }
1355
1356 //-----------------------------inline_string_indexOfChar-----------------------
1357 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1358 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1359 return false;
1360 }
1361 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1362 return false;
1363 }
1364 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1365 Node* src = argument(0); // byte[]
1366 Node* int_ch = argument(1);
1367 Node* from_index = argument(2);
1368 Node* max = argument(3);
1369
1370 src = must_be_not_null(src, true);
1371
1372 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1373 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1374 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1375
1376 // Range checks
1377 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1378
1379 // Check for int_ch >= 0
1380 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1381 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1382 {
1383 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1384 uncommon_trap(Deoptimization::Reason_intrinsic,
1385 Deoptimization::Action_maybe_recompile);
1386 }
1387 if (stopped()) {
1388 return true;
1389 }
1390
1391 RegionNode* region = new RegionNode(3);
1392 Node* phi = new PhiNode(region, TypeInt::INT);
1393
1394 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1395 C->set_has_split_ifs(true); // Has chance for split-if optimization
1396 _gvn.transform(result);
1397
1398 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1399 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1400
1401 Node* if_lt = generate_slow_guard(bol, nullptr);
1402 if (if_lt != nullptr) {
1403 // result == -1
1404 phi->init_req(2, result);
1405 region->init_req(2, if_lt);
1406 }
1407 if (!stopped()) {
1408 result = _gvn.transform(new AddINode(result, from_index));
1409 phi->init_req(1, result);
1410 region->init_req(1, control());
1411 }
1412 set_control(_gvn.transform(region));
1413 record_for_igvn(region);
1414 set_result(_gvn.transform(phi));
1415 clear_upper_avx();
1416
1417 return true;
1418 }
1419 //---------------------------inline_string_copy---------------------
1420 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1421 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1422 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1423 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1424 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1425 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1426 bool LibraryCallKit::inline_string_copy(bool compress) {
1427 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1428 return false;
1429 }
1430 int nargs = 5; // 2 oops, 3 ints
1431 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1432
1433 Node* src = argument(0);
1434 Node* src_offset = argument(1);
1435 Node* dst = argument(2);
1436 Node* dst_offset = argument(3);
1437 Node* length = argument(4);
1438
1439 // Check for allocation before we add nodes that would confuse
1440 // tightly_coupled_allocation()
1441 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1442
1443 // Figure out the size and type of the elements we will be copying.
1444 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1445 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1446 if (src_type == nullptr || dst_type == nullptr) {
1447 return false;
1448 }
1449 BasicType src_elem = src_type->elem()->array_element_basic_type();
1450 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1451 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1452 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1453 "Unsupported array types for inline_string_copy");
1454
1455 src = must_be_not_null(src, true);
1456 dst = must_be_not_null(dst, true);
1457
1458 // Convert char[] offsets to byte[] offsets
1459 bool convert_src = (compress && src_elem == T_BYTE);
1460 bool convert_dst = (!compress && dst_elem == T_BYTE);
1461 if (convert_src) {
1462 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1463 } else if (convert_dst) {
1464 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1465 }
1466
1467 // Range checks
1468 generate_string_range_check(src, src_offset, length, convert_src);
1469 generate_string_range_check(dst, dst_offset, length, convert_dst);
1470 if (stopped()) {
1471 return true;
1472 }
1473
1474 Node* src_start = array_element_address(src, src_offset, src_elem);
1475 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1476 // 'src_start' points to src array + scaled offset
1477 // 'dst_start' points to dst array + scaled offset
1478 Node* count = nullptr;
1479 if (compress) {
1480 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1481 } else {
1482 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1483 }
1484
1485 if (alloc != nullptr) {
1486 if (alloc->maybe_set_complete(&_gvn)) {
1487 // "You break it, you buy it."
1488 InitializeNode* init = alloc->initialization();
1489 assert(init->is_complete(), "we just did this");
1490 init->set_complete_with_arraycopy();
1491 assert(dst->is_CheckCastPP(), "sanity");
1492 assert(dst->in(0)->in(0) == init, "dest pinned");
1493 }
1494 // Do not let stores that initialize this object be reordered with
1495 // a subsequent store that would make this object accessible by
1496 // other threads.
1497 // Record what AllocateNode this StoreStore protects so that
1498 // escape analysis can go from the MemBarStoreStoreNode to the
1499 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1500 // based on the escape status of the AllocateNode.
1501 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1502 }
1503 if (compress) {
1504 set_result(_gvn.transform(count));
1505 }
1506 clear_upper_avx();
1507
1508 return true;
1509 }
1510
1511 #ifdef _LP64
1512 #define XTOP ,top() /*additional argument*/
1513 #else //_LP64
1514 #define XTOP /*no additional argument*/
1515 #endif //_LP64
1516
1517 //------------------------inline_string_toBytesU--------------------------
1518 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1519 bool LibraryCallKit::inline_string_toBytesU() {
1520 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1521 return false;
1522 }
1523 // Get the arguments.
1524 Node* value = argument(0);
1525 Node* offset = argument(1);
1526 Node* length = argument(2);
1527
1528 Node* newcopy = nullptr;
1529
1530 // Set the original stack and the reexecute bit for the interpreter to reexecute
1531 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1532 { PreserveReexecuteState preexecs(this);
1533 jvms()->set_should_reexecute(true);
1534
1535 // Check if a null path was taken unconditionally.
1536 value = null_check(value);
1537
1538 RegionNode* bailout = new RegionNode(1);
1539 record_for_igvn(bailout);
1540
1541 // Range checks
1542 generate_negative_guard(offset, bailout);
1543 generate_negative_guard(length, bailout);
1544 generate_limit_guard(offset, length, load_array_length(value), bailout);
1545 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1546 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1547
1548 if (bailout->req() > 1) {
1549 PreserveJVMState pjvms(this);
1550 set_control(_gvn.transform(bailout));
1551 uncommon_trap(Deoptimization::Reason_intrinsic,
1552 Deoptimization::Action_maybe_recompile);
1553 }
1554 if (stopped()) {
1555 return true;
1556 }
1557
1558 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1559 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1560 newcopy = new_array(klass_node, size, 0); // no arguments to push
1561 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1562 guarantee(alloc != nullptr, "created above");
1563
1564 // Calculate starting addresses.
1565 Node* src_start = array_element_address(value, offset, T_CHAR);
1566 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1567
1568 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1569 const TypeInt* toffset = gvn().type(offset)->is_int();
1570 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1571
1572 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1573 const char* copyfunc_name = "arraycopy";
1574 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1575 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1576 OptoRuntime::fast_arraycopy_Type(),
1577 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1578 src_start, dst_start, ConvI2X(length) XTOP);
1579 // Do not let reads from the cloned object float above the arraycopy.
1580 if (alloc->maybe_set_complete(&_gvn)) {
1581 // "You break it, you buy it."
1582 InitializeNode* init = alloc->initialization();
1583 assert(init->is_complete(), "we just did this");
1584 init->set_complete_with_arraycopy();
1585 assert(newcopy->is_CheckCastPP(), "sanity");
1586 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1587 }
1588 // Do not let stores that initialize this object be reordered with
1589 // a subsequent store that would make this object accessible by
1590 // other threads.
1591 // Record what AllocateNode this StoreStore protects so that
1592 // escape analysis can go from the MemBarStoreStoreNode to the
1593 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1594 // based on the escape status of the AllocateNode.
1595 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1596 } // original reexecute is set back here
1597
1598 C->set_has_split_ifs(true); // Has chance for split-if optimization
1599 if (!stopped()) {
1600 set_result(newcopy);
1601 }
1602 clear_upper_avx();
1603
1604 return true;
1605 }
1606
1607 //------------------------inline_string_getCharsU--------------------------
1608 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1609 bool LibraryCallKit::inline_string_getCharsU() {
1610 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1611 return false;
1612 }
1613
1614 // Get the arguments.
1615 Node* src = argument(0);
1616 Node* src_begin = argument(1);
1617 Node* src_end = argument(2); // exclusive offset (i < src_end)
1618 Node* dst = argument(3);
1619 Node* dst_begin = argument(4);
1620
1621 // Check for allocation before we add nodes that would confuse
1622 // tightly_coupled_allocation()
1623 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1624
1625 // Check if a null path was taken unconditionally.
1626 src = null_check(src);
1627 dst = null_check(dst);
1628 if (stopped()) {
1629 return true;
1630 }
1631
1632 // Get length and convert char[] offset to byte[] offset
1633 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1634 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1635
1636 // Range checks
1637 generate_string_range_check(src, src_begin, length, true);
1638 generate_string_range_check(dst, dst_begin, length, false);
1639 if (stopped()) {
1640 return true;
1641 }
1642
1643 if (!stopped()) {
1644 // Calculate starting addresses.
1645 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1646 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1647
1648 // Check if array addresses are aligned to HeapWordSize
1649 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1650 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1651 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1652 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1653
1654 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1655 const char* copyfunc_name = "arraycopy";
1656 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1657 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1658 OptoRuntime::fast_arraycopy_Type(),
1659 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1660 src_start, dst_start, ConvI2X(length) XTOP);
1661 // Do not let reads from the cloned object float above the arraycopy.
1662 if (alloc != nullptr) {
1663 if (alloc->maybe_set_complete(&_gvn)) {
1664 // "You break it, you buy it."
1665 InitializeNode* init = alloc->initialization();
1666 assert(init->is_complete(), "we just did this");
1667 init->set_complete_with_arraycopy();
1668 assert(dst->is_CheckCastPP(), "sanity");
1669 assert(dst->in(0)->in(0) == init, "dest pinned");
1670 }
1671 // Do not let stores that initialize this object be reordered with
1672 // a subsequent store that would make this object accessible by
1673 // other threads.
1674 // Record what AllocateNode this StoreStore protects so that
1675 // escape analysis can go from the MemBarStoreStoreNode to the
1676 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1677 // based on the escape status of the AllocateNode.
1678 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1679 } else {
1680 insert_mem_bar(Op_MemBarCPUOrder);
1681 }
1682 }
1683
1684 C->set_has_split_ifs(true); // Has chance for split-if optimization
1685 return true;
1686 }
1687
1688 //----------------------inline_string_char_access----------------------------
1689 // Store/Load char to/from byte[] array.
1690 // static void StringUTF16.putChar(byte[] val, int index, int c)
1691 // static char StringUTF16.getChar(byte[] val, int index)
1692 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1693 Node* value = argument(0);
1694 Node* index = argument(1);
1695 Node* ch = is_store ? argument(2) : nullptr;
1696
1697 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1698 // correctly requires matched array shapes.
1699 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1700 "sanity: byte[] and char[] bases agree");
1701 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1702 "sanity: byte[] and char[] scales agree");
1703
1704 // Bail when getChar over constants is requested: constant folding would
1705 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1706 // Java method would constant fold nicely instead.
1707 if (!is_store && value->is_Con() && index->is_Con()) {
1708 return false;
1709 }
1710
1711 // Save state and restore on bailout
1712 uint old_sp = sp();
1713 SafePointNode* old_map = clone_map();
1714
1715 value = must_be_not_null(value, true);
1716
1717 Node* adr = array_element_address(value, index, T_CHAR);
1718 if (adr->is_top()) {
1719 set_map(old_map);
1720 set_sp(old_sp);
1721 return false;
1722 }
1723 destruct_map_clone(old_map);
1724 if (is_store) {
1725 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1726 } else {
1727 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);
1728 set_result(ch);
1729 }
1730 return true;
1731 }
1732
1733 //--------------------------round_double_node--------------------------------
1734 // Round a double node if necessary.
1735 Node* LibraryCallKit::round_double_node(Node* n) {
1736 if (Matcher::strict_fp_requires_explicit_rounding) {
1737 #ifdef IA32
1738 if (UseSSE < 2) {
1739 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1740 }
1741 #else
1742 Unimplemented();
1743 #endif // IA32
1744 }
1745 return n;
1746 }
1747
1748 //------------------------------inline_math-----------------------------------
1749 // public static double Math.abs(double)
1750 // public static double Math.sqrt(double)
1751 // public static double Math.log(double)
1752 // public static double Math.log10(double)
1753 // public static double Math.round(double)
1754 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1755 Node* arg = round_double_node(argument(0));
1756 Node* n = nullptr;
1757 switch (id) {
1758 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1759 case vmIntrinsics::_dsqrt:
1760 case vmIntrinsics::_dsqrt_strict:
1761 n = new SqrtDNode(C, control(), arg); break;
1762 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1763 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1764 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1765 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1766 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1767 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1768 default: fatal_unexpected_iid(id); break;
1769 }
1770 set_result(_gvn.transform(n));
1771 return true;
1772 }
1773
1774 //------------------------------inline_math-----------------------------------
1775 // public static float Math.abs(float)
1776 // public static int Math.abs(int)
1777 // public static long Math.abs(long)
1778 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1779 Node* arg = argument(0);
1780 Node* n = nullptr;
1781 switch (id) {
1782 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1783 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1784 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1785 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1786 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1787 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1788 default: fatal_unexpected_iid(id); break;
1789 }
1790 set_result(_gvn.transform(n));
1791 return true;
1792 }
1793
1794 //------------------------------runtime_math-----------------------------
1795 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1796 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1797 "must be (DD)D or (D)D type");
1798
1799 // Inputs
1800 Node* a = round_double_node(argument(0));
1801 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1802
1803 const TypePtr* no_memory_effects = nullptr;
1804 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1805 no_memory_effects,
1806 a, top(), b, b ? top() : nullptr);
1807 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1808 #ifdef ASSERT
1809 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1810 assert(value_top == top(), "second value must be top");
1811 #endif
1812
1813 set_result(value);
1814 return true;
1815 }
1816
1817 //------------------------------inline_math_pow-----------------------------
1818 bool LibraryCallKit::inline_math_pow() {
1819 Node* exp = round_double_node(argument(2));
1820 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1821 if (d != nullptr) {
1822 if (d->getd() == 2.0) {
1823 // Special case: pow(x, 2.0) => x * x
1824 Node* base = round_double_node(argument(0));
1825 set_result(_gvn.transform(new MulDNode(base, base)));
1826 return true;
1827 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1828 // Special case: pow(x, 0.5) => sqrt(x)
1829 Node* base = round_double_node(argument(0));
1830 Node* zero = _gvn.zerocon(T_DOUBLE);
1831
1832 RegionNode* region = new RegionNode(3);
1833 Node* phi = new PhiNode(region, Type::DOUBLE);
1834
1835 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1836 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1837 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1838 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1839 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1840
1841 Node* if_pow = generate_slow_guard(test, nullptr);
1842 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1843 phi->init_req(1, value_sqrt);
1844 region->init_req(1, control());
1845
1846 if (if_pow != nullptr) {
1847 set_control(if_pow);
1848 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1849 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1850 const TypePtr* no_memory_effects = nullptr;
1851 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1852 no_memory_effects, base, top(), exp, top());
1853 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1854 #ifdef ASSERT
1855 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1856 assert(value_top == top(), "second value must be top");
1857 #endif
1858 phi->init_req(2, value_pow);
1859 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1860 }
1861
1862 C->set_has_split_ifs(true); // Has chance for split-if optimization
1863 set_control(_gvn.transform(region));
1864 record_for_igvn(region);
1865 set_result(_gvn.transform(phi));
1866
1867 return true;
1868 }
1869 }
1870
1871 return StubRoutines::dpow() != nullptr ?
1872 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1873 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1874 }
1875
1876 //------------------------------inline_math_native-----------------------------
1877 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1878 switch (id) {
1879 case vmIntrinsics::_dsin:
1880 return StubRoutines::dsin() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1883 case vmIntrinsics::_dcos:
1884 return StubRoutines::dcos() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1887 case vmIntrinsics::_dtan:
1888 return StubRoutines::dtan() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1890 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1891 case vmIntrinsics::_dtanh:
1892 return StubRoutines::dtanh() != nullptr ?
1893 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1894 case vmIntrinsics::_dexp:
1895 return StubRoutines::dexp() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1898 case vmIntrinsics::_dlog:
1899 return StubRoutines::dlog() != nullptr ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1901 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1902 case vmIntrinsics::_dlog10:
1903 return StubRoutines::dlog10() != nullptr ?
1904 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1905 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1906
1907 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1908 case vmIntrinsics::_ceil:
1909 case vmIntrinsics::_floor:
1910 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1911
1912 case vmIntrinsics::_dsqrt:
1913 case vmIntrinsics::_dsqrt_strict:
1914 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1915 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1916 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1917 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1918 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1919
1920 case vmIntrinsics::_dpow: return inline_math_pow();
1921 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1922 case vmIntrinsics::_fcopySign: return inline_math(id);
1923 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1924 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1925 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1926
1927 // These intrinsics are not yet correctly implemented
1928 case vmIntrinsics::_datan2:
1929 return false;
1930
1931 default:
1932 fatal_unexpected_iid(id);
1933 return false;
1934 }
1935 }
1936
1937 //----------------------------inline_notify-----------------------------------*
1938 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1939 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1940 address func;
1941 if (id == vmIntrinsics::_notify) {
1942 func = OptoRuntime::monitor_notify_Java();
1943 } else {
1944 func = OptoRuntime::monitor_notifyAll_Java();
1945 }
1946 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1947 make_slow_call_ex(call, env()->Throwable_klass(), false);
1948 return true;
1949 }
1950
1951
1952 //----------------------------inline_min_max-----------------------------------
1953 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1954 set_result(generate_min_max(id, argument(0), argument(1)));
1955 return true;
1956 }
1957
1958 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1959 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1960 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1961 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1962 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1963
1964 {
1965 PreserveJVMState pjvms(this);
1966 PreserveReexecuteState preexecs(this);
1967 jvms()->set_should_reexecute(true);
1968
1969 set_control(slow_path);
1970 set_i_o(i_o());
1971
1972 uncommon_trap(Deoptimization::Reason_intrinsic,
1973 Deoptimization::Action_none);
1974 }
1975
1976 set_control(fast_path);
1977 set_result(math);
1978 }
1979
1980 template <typename OverflowOp>
1981 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1982 typedef typename OverflowOp::MathOp MathOp;
1983
1984 MathOp* mathOp = new MathOp(arg1, arg2);
1985 Node* operation = _gvn.transform( mathOp );
1986 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1987 inline_math_mathExact(operation, ofcheck);
1988 return true;
1989 }
1990
1991 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1992 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1993 }
1994
1995 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1996 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1997 }
1998
1999 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2000 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2001 }
2002
2003 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2004 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2005 }
2006
2007 bool LibraryCallKit::inline_math_negateExactI() {
2008 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2009 }
2010
2011 bool LibraryCallKit::inline_math_negateExactL() {
2012 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2013 }
2014
2015 bool LibraryCallKit::inline_math_multiplyExactI() {
2016 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2017 }
2018
2019 bool LibraryCallKit::inline_math_multiplyExactL() {
2020 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2021 }
2022
2023 bool LibraryCallKit::inline_math_multiplyHigh() {
2024 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2025 return true;
2026 }
2027
2028 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2029 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2030 return true;
2031 }
2032
2033 Node*
2034 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2035 Node* result_val = nullptr;
2036 switch (id) {
2037 case vmIntrinsics::_min:
2038 case vmIntrinsics::_min_strict:
2039 result_val = _gvn.transform(new MinINode(x0, y0));
2040 break;
2041 case vmIntrinsics::_max:
2042 case vmIntrinsics::_max_strict:
2043 result_val = _gvn.transform(new MaxINode(x0, y0));
2044 break;
2045 default:
2046 fatal_unexpected_iid(id);
2047 break;
2048 }
2049 return result_val;
2050 }
2051
2052 inline int
2053 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2054 const TypePtr* base_type = TypePtr::NULL_PTR;
2055 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2056 if (base_type == nullptr) {
2057 // Unknown type.
2058 return Type::AnyPtr;
2059 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2060 // Since this is a null+long form, we have to switch to a rawptr.
2061 base = _gvn.transform(new CastX2PNode(offset));
2062 offset = MakeConX(0);
2063 return Type::RawPtr;
2064 } else if (base_type->base() == Type::RawPtr) {
2065 return Type::RawPtr;
2066 } else if (base_type->isa_oopptr()) {
2067 // Base is never null => always a heap address.
2068 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2069 return Type::OopPtr;
2070 }
2071 // Offset is small => always a heap address.
2072 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2073 if (offset_type != nullptr &&
2074 base_type->offset() == 0 && // (should always be?)
2075 offset_type->_lo >= 0 &&
2076 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2077 return Type::OopPtr;
2078 } else if (type == T_OBJECT) {
2079 // off heap access to an oop doesn't make any sense. Has to be on
2080 // heap.
2081 return Type::OopPtr;
2082 }
2083 // Otherwise, it might either be oop+off or null+addr.
2084 return Type::AnyPtr;
2085 } else {
2086 // No information:
2087 return Type::AnyPtr;
2088 }
2089 }
2090
2091 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2092 Node* uncasted_base = base;
2093 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2094 if (kind == Type::RawPtr) {
2095 return basic_plus_adr(top(), uncasted_base, offset);
2096 } else if (kind == Type::AnyPtr) {
2097 assert(base == uncasted_base, "unexpected base change");
2098 if (can_cast) {
2099 if (!_gvn.type(base)->speculative_maybe_null() &&
2100 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2101 // According to profiling, this access is always on
2102 // heap. Casting the base to not null and thus avoiding membars
2103 // around the access should allow better optimizations
2104 Node* null_ctl = top();
2105 base = null_check_oop(base, &null_ctl, true, true, true);
2106 assert(null_ctl->is_top(), "no null control here");
2107 return basic_plus_adr(base, offset);
2108 } else if (_gvn.type(base)->speculative_always_null() &&
2109 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2110 // According to profiling, this access is always off
2111 // heap.
2112 base = null_assert(base);
2113 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2114 offset = MakeConX(0);
2115 return basic_plus_adr(top(), raw_base, offset);
2116 }
2117 }
2118 // We don't know if it's an on heap or off heap access. Fall back
2119 // to raw memory access.
2120 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2121 return basic_plus_adr(top(), raw, offset);
2122 } else {
2123 assert(base == uncasted_base, "unexpected base change");
2124 // We know it's an on heap access so base can't be null
2125 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2126 base = must_be_not_null(base, true);
2127 }
2128 return basic_plus_adr(base, offset);
2129 }
2130 }
2131
2132 //--------------------------inline_number_methods-----------------------------
2133 // inline int Integer.numberOfLeadingZeros(int)
2134 // inline int Long.numberOfLeadingZeros(long)
2135 //
2136 // inline int Integer.numberOfTrailingZeros(int)
2137 // inline int Long.numberOfTrailingZeros(long)
2138 //
2139 // inline int Integer.bitCount(int)
2140 // inline int Long.bitCount(long)
2141 //
2142 // inline char Character.reverseBytes(char)
2143 // inline short Short.reverseBytes(short)
2144 // inline int Integer.reverseBytes(int)
2145 // inline long Long.reverseBytes(long)
2146 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2147 Node* arg = argument(0);
2148 Node* n = nullptr;
2149 switch (id) {
2150 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2151 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2152 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2153 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2154 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2155 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2156 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2157 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2158 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2159 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2160 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2161 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2162 default: fatal_unexpected_iid(id); break;
2163 }
2164 set_result(_gvn.transform(n));
2165 return true;
2166 }
2167
2168 //--------------------------inline_bitshuffle_methods-----------------------------
2169 // inline int Integer.compress(int, int)
2170 // inline int Integer.expand(int, int)
2171 // inline long Long.compress(long, long)
2172 // inline long Long.expand(long, long)
2173 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2174 Node* n = nullptr;
2175 switch (id) {
2176 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2177 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2178 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2179 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2180 default: fatal_unexpected_iid(id); break;
2181 }
2182 set_result(_gvn.transform(n));
2183 return true;
2184 }
2185
2186 //--------------------------inline_number_methods-----------------------------
2187 // inline int Integer.compareUnsigned(int, int)
2188 // inline int Long.compareUnsigned(long, long)
2189 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2190 Node* arg1 = argument(0);
2191 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2192 Node* n = nullptr;
2193 switch (id) {
2194 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2195 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2196 default: fatal_unexpected_iid(id); break;
2197 }
2198 set_result(_gvn.transform(n));
2199 return true;
2200 }
2201
2202 //--------------------------inline_unsigned_divmod_methods-----------------------------
2203 // inline int Integer.divideUnsigned(int, int)
2204 // inline int Integer.remainderUnsigned(int, int)
2205 // inline long Long.divideUnsigned(long, long)
2206 // inline long Long.remainderUnsigned(long, long)
2207 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2208 Node* n = nullptr;
2209 switch (id) {
2210 case vmIntrinsics::_divideUnsigned_i: {
2211 zero_check_int(argument(1));
2212 // Compile-time detect of null-exception
2213 if (stopped()) {
2214 return true; // keep the graph constructed so far
2215 }
2216 n = new UDivINode(control(), argument(0), argument(1));
2217 break;
2218 }
2219 case vmIntrinsics::_divideUnsigned_l: {
2220 zero_check_long(argument(2));
2221 // Compile-time detect of null-exception
2222 if (stopped()) {
2223 return true; // keep the graph constructed so far
2224 }
2225 n = new UDivLNode(control(), argument(0), argument(2));
2226 break;
2227 }
2228 case vmIntrinsics::_remainderUnsigned_i: {
2229 zero_check_int(argument(1));
2230 // Compile-time detect of null-exception
2231 if (stopped()) {
2232 return true; // keep the graph constructed so far
2233 }
2234 n = new UModINode(control(), argument(0), argument(1));
2235 break;
2236 }
2237 case vmIntrinsics::_remainderUnsigned_l: {
2238 zero_check_long(argument(2));
2239 // Compile-time detect of null-exception
2240 if (stopped()) {
2241 return true; // keep the graph constructed so far
2242 }
2243 n = new UModLNode(control(), argument(0), argument(2));
2244 break;
2245 }
2246 default: fatal_unexpected_iid(id); break;
2247 }
2248 set_result(_gvn.transform(n));
2249 return true;
2250 }
2251
2252 //----------------------------inline_unsafe_access----------------------------
2253
2254 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2255 // Attempt to infer a sharper value type from the offset and base type.
2256 ciKlass* sharpened_klass = nullptr;
2257
2258 // See if it is an instance field, with an object type.
2259 if (alias_type->field() != nullptr) {
2260 if (alias_type->field()->type()->is_klass()) {
2261 sharpened_klass = alias_type->field()->type()->as_klass();
2262 }
2263 }
2264
2265 const TypeOopPtr* result = nullptr;
2266 // See if it is a narrow oop array.
2267 if (adr_type->isa_aryptr()) {
2268 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2269 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2270 if (elem_type != nullptr && elem_type->is_loaded()) {
2271 // Sharpen the value type.
2272 result = elem_type;
2273 }
2274 }
2275 }
2276
2277 // The sharpened class might be unloaded if there is no class loader
2278 // contraint in place.
2279 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2280 // Sharpen the value type.
2281 result = TypeOopPtr::make_from_klass(sharpened_klass);
2282 }
2283 if (result != nullptr) {
2284 #ifndef PRODUCT
2285 if (C->print_intrinsics() || C->print_inlining()) {
2286 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2287 tty->print(" sharpened value: "); result->dump(); tty->cr();
2288 }
2289 #endif
2290 }
2291 return result;
2292 }
2293
2294 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2295 switch (kind) {
2296 case Relaxed:
2297 return MO_UNORDERED;
2298 case Opaque:
2299 return MO_RELAXED;
2300 case Acquire:
2301 return MO_ACQUIRE;
2302 case Release:
2303 return MO_RELEASE;
2304 case Volatile:
2305 return MO_SEQ_CST;
2306 default:
2307 ShouldNotReachHere();
2308 return 0;
2309 }
2310 }
2311
2312 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2313 if (callee()->is_static()) return false; // caller must have the capability!
2314 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2315 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2316 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2317 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2318
2319 if (is_reference_type(type)) {
2320 decorators |= ON_UNKNOWN_OOP_REF;
2321 }
2322
2323 if (unaligned) {
2324 decorators |= C2_UNALIGNED;
2325 }
2326
2327 #ifndef PRODUCT
2328 {
2329 ResourceMark rm;
2330 // Check the signatures.
2331 ciSignature* sig = callee()->signature();
2332 #ifdef ASSERT
2333 if (!is_store) {
2334 // Object getReference(Object base, int/long offset), etc.
2335 BasicType rtype = sig->return_type()->basic_type();
2336 assert(rtype == type, "getter must return the expected value");
2337 assert(sig->count() == 2, "oop getter has 2 arguments");
2338 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2339 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2340 } else {
2341 // void putReference(Object base, int/long offset, Object x), etc.
2342 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2343 assert(sig->count() == 3, "oop putter has 3 arguments");
2344 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2345 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2346 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2347 assert(vtype == type, "putter must accept the expected value");
2348 }
2349 #endif // ASSERT
2350 }
2351 #endif //PRODUCT
2352
2353 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2354
2355 Node* receiver = argument(0); // type: oop
2356
2357 // Build address expression.
2358 Node* heap_base_oop = top();
2359
2360 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2361 Node* base = argument(1); // type: oop
2362 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2363 Node* offset = argument(2); // type: long
2364 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2365 // to be plain byte offsets, which are also the same as those accepted
2366 // by oopDesc::field_addr.
2367 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2368 "fieldOffset must be byte-scaled");
2369 // 32-bit machines ignore the high half!
2370 offset = ConvL2X(offset);
2371
2372 // Save state and restore on bailout
2373 uint old_sp = sp();
2374 SafePointNode* old_map = clone_map();
2375
2376 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2377 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2378
2379 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2380 if (type != T_OBJECT) {
2381 decorators |= IN_NATIVE; // off-heap primitive access
2382 } else {
2383 set_map(old_map);
2384 set_sp(old_sp);
2385 return false; // off-heap oop accesses are not supported
2386 }
2387 } else {
2388 heap_base_oop = base; // on-heap or mixed access
2389 }
2390
2391 // Can base be null? Otherwise, always on-heap access.
2392 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2393
2394 if (!can_access_non_heap) {
2395 decorators |= IN_HEAP;
2396 }
2397
2398 Node* val = is_store ? argument(4) : nullptr;
2399
2400 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2401 if (adr_type == TypePtr::NULL_PTR) {
2402 set_map(old_map);
2403 set_sp(old_sp);
2404 return false; // off-heap access with zero address
2405 }
2406
2407 // Try to categorize the address.
2408 Compile::AliasType* alias_type = C->alias_type(adr_type);
2409 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2410
2411 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2412 alias_type->adr_type() == TypeAryPtr::RANGE) {
2413 set_map(old_map);
2414 set_sp(old_sp);
2415 return false; // not supported
2416 }
2417
2418 bool mismatched = false;
2419 BasicType bt = alias_type->basic_type();
2420 if (bt != T_ILLEGAL) {
2421 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2422 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2423 // Alias type doesn't differentiate between byte[] and boolean[]).
2424 // Use address type to get the element type.
2425 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2426 }
2427 if (is_reference_type(bt, true)) {
2428 // accessing an array field with getReference is not a mismatch
2429 bt = T_OBJECT;
2430 }
2431 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2432 // Don't intrinsify mismatched object accesses
2433 set_map(old_map);
2434 set_sp(old_sp);
2435 return false;
2436 }
2437 mismatched = (bt != type);
2438 } else if (alias_type->adr_type()->isa_oopptr()) {
2439 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2440 }
2441
2442 destruct_map_clone(old_map);
2443 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2444
2445 if (mismatched) {
2446 decorators |= C2_MISMATCHED;
2447 }
2448
2449 // First guess at the value type.
2450 const Type *value_type = Type::get_const_basic_type(type);
2451
2452 // Figure out the memory ordering.
2453 decorators |= mo_decorator_for_access_kind(kind);
2454
2455 if (!is_store && type == T_OBJECT) {
2456 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2457 if (tjp != nullptr) {
2458 value_type = tjp;
2459 }
2460 }
2461
2462 receiver = null_check(receiver);
2463 if (stopped()) {
2464 return true;
2465 }
2466 // Heap pointers get a null-check from the interpreter,
2467 // as a courtesy. However, this is not guaranteed by Unsafe,
2468 // and it is not possible to fully distinguish unintended nulls
2469 // from intended ones in this API.
2470
2471 if (!is_store) {
2472 Node* p = nullptr;
2473 // Try to constant fold a load from a constant field
2474 ciField* field = alias_type->field();
2475 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2476 // final or stable field
2477 p = make_constant_from_field(field, heap_base_oop);
2478 }
2479
2480 if (p == nullptr) { // Could not constant fold the load
2481 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2482 // Normalize the value returned by getBoolean in the following cases
2483 if (type == T_BOOLEAN &&
2484 (mismatched ||
2485 heap_base_oop == top() || // - heap_base_oop is null or
2486 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2487 // and the unsafe access is made to large offset
2488 // (i.e., larger than the maximum offset necessary for any
2489 // field access)
2490 ) {
2491 IdealKit ideal = IdealKit(this);
2492 #define __ ideal.
2493 IdealVariable normalized_result(ideal);
2494 __ declarations_done();
2495 __ set(normalized_result, p);
2496 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2497 __ set(normalized_result, ideal.ConI(1));
2498 ideal.end_if();
2499 final_sync(ideal);
2500 p = __ value(normalized_result);
2501 #undef __
2502 }
2503 }
2504 if (type == T_ADDRESS) {
2505 p = gvn().transform(new CastP2XNode(nullptr, p));
2506 p = ConvX2UL(p);
2507 }
2508 // The load node has the control of the preceding MemBarCPUOrder. All
2509 // following nodes will have the control of the MemBarCPUOrder inserted at
2510 // the end of this method. So, pushing the load onto the stack at a later
2511 // point is fine.
2512 set_result(p);
2513 } else {
2514 if (bt == T_ADDRESS) {
2515 // Repackage the long as a pointer.
2516 val = ConvL2X(val);
2517 val = gvn().transform(new CastX2PNode(val));
2518 }
2519 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2520 }
2521
2522 return true;
2523 }
2524
2525 //----------------------------inline_unsafe_load_store----------------------------
2526 // This method serves a couple of different customers (depending on LoadStoreKind):
2527 //
2528 // LS_cmp_swap:
2529 //
2530 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2531 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2532 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2533 //
2534 // LS_cmp_swap_weak:
2535 //
2536 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2537 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2538 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2539 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2540 //
2541 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2542 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2543 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2544 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2545 //
2546 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2547 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2548 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2549 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2550 //
2551 // LS_cmp_exchange:
2552 //
2553 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2554 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2555 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2556 //
2557 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2558 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2559 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2560 //
2561 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2562 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2563 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2564 //
2565 // LS_get_add:
2566 //
2567 // int getAndAddInt( Object o, long offset, int delta)
2568 // long getAndAddLong(Object o, long offset, long delta)
2569 //
2570 // LS_get_set:
2571 //
2572 // int getAndSet(Object o, long offset, int newValue)
2573 // long getAndSet(Object o, long offset, long newValue)
2574 // Object getAndSet(Object o, long offset, Object newValue)
2575 //
2576 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2577 // This basic scheme here is the same as inline_unsafe_access, but
2578 // differs in enough details that combining them would make the code
2579 // overly confusing. (This is a true fact! I originally combined
2580 // them, but even I was confused by it!) As much code/comments as
2581 // possible are retained from inline_unsafe_access though to make
2582 // the correspondences clearer. - dl
2583
2584 if (callee()->is_static()) return false; // caller must have the capability!
2585
2586 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2587 decorators |= mo_decorator_for_access_kind(access_kind);
2588
2589 #ifndef PRODUCT
2590 BasicType rtype;
2591 {
2592 ResourceMark rm;
2593 // Check the signatures.
2594 ciSignature* sig = callee()->signature();
2595 rtype = sig->return_type()->basic_type();
2596 switch(kind) {
2597 case LS_get_add:
2598 case LS_get_set: {
2599 // Check the signatures.
2600 #ifdef ASSERT
2601 assert(rtype == type, "get and set must return the expected type");
2602 assert(sig->count() == 3, "get and set has 3 arguments");
2603 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2604 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2605 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2606 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2607 #endif // ASSERT
2608 break;
2609 }
2610 case LS_cmp_swap:
2611 case LS_cmp_swap_weak: {
2612 // Check the signatures.
2613 #ifdef ASSERT
2614 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2615 assert(sig->count() == 4, "CAS has 4 arguments");
2616 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2617 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2618 #endif // ASSERT
2619 break;
2620 }
2621 case LS_cmp_exchange: {
2622 // Check the signatures.
2623 #ifdef ASSERT
2624 assert(rtype == type, "CAS must return the expected type");
2625 assert(sig->count() == 4, "CAS has 4 arguments");
2626 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2627 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2628 #endif // ASSERT
2629 break;
2630 }
2631 default:
2632 ShouldNotReachHere();
2633 }
2634 }
2635 #endif //PRODUCT
2636
2637 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2638
2639 // Get arguments:
2640 Node* receiver = nullptr;
2641 Node* base = nullptr;
2642 Node* offset = nullptr;
2643 Node* oldval = nullptr;
2644 Node* newval = nullptr;
2645 switch(kind) {
2646 case LS_cmp_swap:
2647 case LS_cmp_swap_weak:
2648 case LS_cmp_exchange: {
2649 const bool two_slot_type = type2size[type] == 2;
2650 receiver = argument(0); // type: oop
2651 base = argument(1); // type: oop
2652 offset = argument(2); // type: long
2653 oldval = argument(4); // type: oop, int, or long
2654 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2655 break;
2656 }
2657 case LS_get_add:
2658 case LS_get_set: {
2659 receiver = argument(0); // type: oop
2660 base = argument(1); // type: oop
2661 offset = argument(2); // type: long
2662 oldval = nullptr;
2663 newval = argument(4); // type: oop, int, or long
2664 break;
2665 }
2666 default:
2667 ShouldNotReachHere();
2668 }
2669
2670 // Build field offset expression.
2671 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2672 // to be plain byte offsets, which are also the same as those accepted
2673 // by oopDesc::field_addr.
2674 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2675 // 32-bit machines ignore the high half of long offsets
2676 offset = ConvL2X(offset);
2677 // Save state and restore on bailout
2678 uint old_sp = sp();
2679 SafePointNode* old_map = clone_map();
2680 Node* adr = make_unsafe_address(base, offset,type, false);
2681 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2682
2683 Compile::AliasType* alias_type = C->alias_type(adr_type);
2684 BasicType bt = alias_type->basic_type();
2685 if (bt != T_ILLEGAL &&
2686 (is_reference_type(bt) != (type == T_OBJECT))) {
2687 // Don't intrinsify mismatched object accesses.
2688 set_map(old_map);
2689 set_sp(old_sp);
2690 return false;
2691 }
2692
2693 destruct_map_clone(old_map);
2694
2695 // For CAS, unlike inline_unsafe_access, there seems no point in
2696 // trying to refine types. Just use the coarse types here.
2697 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2698 const Type *value_type = Type::get_const_basic_type(type);
2699
2700 switch (kind) {
2701 case LS_get_set:
2702 case LS_cmp_exchange: {
2703 if (type == T_OBJECT) {
2704 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2705 if (tjp != nullptr) {
2706 value_type = tjp;
2707 }
2708 }
2709 break;
2710 }
2711 case LS_cmp_swap:
2712 case LS_cmp_swap_weak:
2713 case LS_get_add:
2714 break;
2715 default:
2716 ShouldNotReachHere();
2717 }
2718
2719 // Null check receiver.
2720 receiver = null_check(receiver);
2721 if (stopped()) {
2722 return true;
2723 }
2724
2725 int alias_idx = C->get_alias_index(adr_type);
2726
2727 if (is_reference_type(type)) {
2728 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2729
2730 // Transformation of a value which could be null pointer (CastPP #null)
2731 // could be delayed during Parse (for example, in adjust_map_after_if()).
2732 // Execute transformation here to avoid barrier generation in such case.
2733 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2734 newval = _gvn.makecon(TypePtr::NULL_PTR);
2735
2736 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2737 // Refine the value to a null constant, when it is known to be null
2738 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2739 }
2740 }
2741
2742 Node* result = nullptr;
2743 switch (kind) {
2744 case LS_cmp_exchange: {
2745 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2746 oldval, newval, value_type, type, decorators);
2747 break;
2748 }
2749 case LS_cmp_swap_weak:
2750 decorators |= C2_WEAK_CMPXCHG;
2751 case LS_cmp_swap: {
2752 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2753 oldval, newval, value_type, type, decorators);
2754 break;
2755 }
2756 case LS_get_set: {
2757 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2758 newval, value_type, type, decorators);
2759 break;
2760 }
2761 case LS_get_add: {
2762 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2763 newval, value_type, type, decorators);
2764 break;
2765 }
2766 default:
2767 ShouldNotReachHere();
2768 }
2769
2770 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2771 set_result(result);
2772 return true;
2773 }
2774
2775 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2776 // Regardless of form, don't allow previous ld/st to move down,
2777 // then issue acquire, release, or volatile mem_bar.
2778 insert_mem_bar(Op_MemBarCPUOrder);
2779 switch(id) {
2780 case vmIntrinsics::_loadFence:
2781 insert_mem_bar(Op_LoadFence);
2782 return true;
2783 case vmIntrinsics::_storeFence:
2784 insert_mem_bar(Op_StoreFence);
2785 return true;
2786 case vmIntrinsics::_storeStoreFence:
2787 insert_mem_bar(Op_StoreStoreFence);
2788 return true;
2789 case vmIntrinsics::_fullFence:
2790 insert_mem_bar(Op_MemBarVolatile);
2791 return true;
2792 default:
2793 fatal_unexpected_iid(id);
2794 return false;
2795 }
2796 }
2797
2798 bool LibraryCallKit::inline_onspinwait() {
2799 insert_mem_bar(Op_OnSpinWait);
2800 return true;
2801 }
2802
2803 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2804 if (!kls->is_Con()) {
2805 return true;
2806 }
2807 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2808 if (klsptr == nullptr) {
2809 return true;
2810 }
2811 ciInstanceKlass* ik = klsptr->instance_klass();
2812 // don't need a guard for a klass that is already initialized
2813 return !ik->is_initialized();
2814 }
2815
2816 //----------------------------inline_unsafe_writeback0-------------------------
2817 // public native void Unsafe.writeback0(long address)
2818 bool LibraryCallKit::inline_unsafe_writeback0() {
2819 if (!Matcher::has_match_rule(Op_CacheWB)) {
2820 return false;
2821 }
2822 #ifndef PRODUCT
2823 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2824 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2825 ciSignature* sig = callee()->signature();
2826 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2827 #endif
2828 null_check_receiver(); // null-check, then ignore
2829 Node *addr = argument(1);
2830 addr = new CastX2PNode(addr);
2831 addr = _gvn.transform(addr);
2832 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2833 flush = _gvn.transform(flush);
2834 set_memory(flush, TypeRawPtr::BOTTOM);
2835 return true;
2836 }
2837
2838 //----------------------------inline_unsafe_writeback0-------------------------
2839 // public native void Unsafe.writeback0(long address)
2840 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2841 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2842 return false;
2843 }
2844 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2845 return false;
2846 }
2847 #ifndef PRODUCT
2848 assert(Matcher::has_match_rule(Op_CacheWB),
2849 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2850 : "found match rule for CacheWBPostSync but not CacheWB"));
2851
2852 #endif
2853 null_check_receiver(); // null-check, then ignore
2854 Node *sync;
2855 if (is_pre) {
2856 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2857 } else {
2858 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2859 }
2860 sync = _gvn.transform(sync);
2861 set_memory(sync, TypeRawPtr::BOTTOM);
2862 return true;
2863 }
2864
2865 //----------------------------inline_unsafe_allocate---------------------------
2866 // public native Object Unsafe.allocateInstance(Class<?> cls);
2867 bool LibraryCallKit::inline_unsafe_allocate() {
2868
2869 #if INCLUDE_JVMTI
2870 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2871 return false;
2872 }
2873 #endif //INCLUDE_JVMTI
2874
2875 if (callee()->is_static()) return false; // caller must have the capability!
2876
2877 null_check_receiver(); // null-check, then ignore
2878 Node* cls = null_check(argument(1));
2879 if (stopped()) return true;
2880
2881 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2882 kls = null_check(kls);
2883 if (stopped()) return true; // argument was like int.class
2884
2885 #if INCLUDE_JVMTI
2886 // Don't try to access new allocated obj in the intrinsic.
2887 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2888 // Deoptimize and allocate in interpreter instead.
2889 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2890 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2891 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2892 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2893 {
2894 BuildCutout unless(this, tst, PROB_MAX);
2895 uncommon_trap(Deoptimization::Reason_intrinsic,
2896 Deoptimization::Action_make_not_entrant);
2897 }
2898 if (stopped()) {
2899 return true;
2900 }
2901 #endif //INCLUDE_JVMTI
2902
2903 Node* test = nullptr;
2904 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2905 // Note: The argument might still be an illegal value like
2906 // Serializable.class or Object[].class. The runtime will handle it.
2907 // But we must make an explicit check for initialization.
2908 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2909 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2910 // can generate code to load it as unsigned byte.
2911 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2912 Node* bits = intcon(InstanceKlass::fully_initialized);
2913 test = _gvn.transform(new SubINode(inst, bits));
2914 // The 'test' is non-zero if we need to take a slow path.
2915 }
2916
2917 Node* obj = new_instance(kls, test);
2918 set_result(obj);
2919 return true;
2920 }
2921
2922 //------------------------inline_native_time_funcs--------------
2923 // inline code for System.currentTimeMillis() and System.nanoTime()
2924 // these have the same type and signature
2925 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2926 const TypeFunc* tf = OptoRuntime::void_long_Type();
2927 const TypePtr* no_memory_effects = nullptr;
2928 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2929 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2930 #ifdef ASSERT
2931 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2932 assert(value_top == top(), "second value must be top");
2933 #endif
2934 set_result(value);
2935 return true;
2936 }
2937
2938
2939 #if INCLUDE_JVMTI
2940
2941 // When notifications are disabled then just update the VTMS transition bit and return.
2942 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2943 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2944 if (!DoJVMTIVirtualThreadTransitions) {
2945 return true;
2946 }
2947 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2948 IdealKit ideal(this);
2949
2950 Node* ONE = ideal.ConI(1);
2951 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2952 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2953 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2954
2955 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2956 sync_kit(ideal);
2957 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2958 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2959 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2960 ideal.sync_kit(this);
2961 } ideal.else_(); {
2962 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2963 Node* thread = ideal.thread();
2964 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2965 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2966
2967 sync_kit(ideal);
2968 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2969 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2970
2971 ideal.sync_kit(this);
2972 } ideal.end_if();
2973 final_sync(ideal);
2974
2975 return true;
2976 }
2977
2978 // Always update the temporary VTMS transition bit.
2979 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2980 if (!DoJVMTIVirtualThreadTransitions) {
2981 return true;
2982 }
2983 IdealKit ideal(this);
2984
2985 {
2986 // unconditionally update the temporary VTMS transition bit in current JavaThread
2987 Node* thread = ideal.thread();
2988 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
2989 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2990 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2991
2992 sync_kit(ideal);
2993 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2994 ideal.sync_kit(this);
2995 }
2996 final_sync(ideal);
2997
2998 return true;
2999 }
3000
3001 // Always update the is_disable_suspend bit.
3002 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3003 if (!DoJVMTIVirtualThreadTransitions) {
3004 return true;
3005 }
3006 IdealKit ideal(this);
3007
3008 {
3009 // unconditionally update the is_disable_suspend bit in current JavaThread
3010 Node* thread = ideal.thread();
3011 Node* arg = _gvn.transform(argument(0)); // argument for notification
3012 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3013 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3014
3015 sync_kit(ideal);
3016 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3017 ideal.sync_kit(this);
3018 }
3019 final_sync(ideal);
3020
3021 return true;
3022 }
3023
3024 #endif // INCLUDE_JVMTI
3025
3026 #ifdef JFR_HAVE_INTRINSICS
3027
3028 /**
3029 * if oop->klass != null
3030 * // normal class
3031 * epoch = _epoch_state ? 2 : 1
3032 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3033 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3034 * }
3035 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3036 * else
3037 * // primitive class
3038 * if oop->array_klass != null
3039 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3040 * else
3041 * id = LAST_TYPE_ID + 1 // void class path
3042 * if (!signaled)
3043 * signaled = true
3044 */
3045 bool LibraryCallKit::inline_native_classID() {
3046 Node* cls = argument(0);
3047
3048 IdealKit ideal(this);
3049 #define __ ideal.
3050 IdealVariable result(ideal); __ declarations_done();
3051 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3052 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3053 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3054
3055
3056 __ if_then(kls, BoolTest::ne, null()); {
3057 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3058 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3059
3060 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3061 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3062 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3063 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3064 mask = _gvn.transform(new OrLNode(mask, epoch));
3065 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3066
3067 float unlikely = PROB_UNLIKELY(0.999);
3068 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3069 sync_kit(ideal);
3070 make_runtime_call(RC_LEAF,
3071 OptoRuntime::class_id_load_barrier_Type(),
3072 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3073 "class id load barrier",
3074 TypePtr::BOTTOM,
3075 kls);
3076 ideal.sync_kit(this);
3077 } __ end_if();
3078
3079 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3080 } __ else_(); {
3081 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3082 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3083 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3084 __ if_then(array_kls, BoolTest::ne, null()); {
3085 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3086 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3087 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3088 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3089 } __ else_(); {
3090 // void class case
3091 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3092 } __ end_if();
3093
3094 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3095 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3096 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3097 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3098 } __ end_if();
3099 } __ end_if();
3100
3101 final_sync(ideal);
3102 set_result(ideal.value(result));
3103 #undef __
3104 return true;
3105 }
3106
3107 //------------------------inline_native_jvm_commit------------------
3108 bool LibraryCallKit::inline_native_jvm_commit() {
3109 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3110
3111 // Save input memory and i_o state.
3112 Node* input_memory_state = reset_memory();
3113 set_all_memory(input_memory_state);
3114 Node* input_io_state = i_o();
3115
3116 // TLS.
3117 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3118 // Jfr java buffer.
3119 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3120 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3121 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3122
3123 // Load the current value of the notified field in the JfrThreadLocal.
3124 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3125 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3126
3127 // Test for notification.
3128 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3129 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3130 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3131
3132 // True branch, is notified.
3133 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3134 set_control(is_notified);
3135
3136 // Reset notified state.
3137 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3138
3139 // 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.
3140 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3141 // Convert the machine-word to a long.
3142 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3143
3144 // False branch, not notified.
3145 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3146 set_control(not_notified);
3147 set_all_memory(input_memory_state);
3148
3149 // Arg is the next position as a long.
3150 Node* arg = argument(0);
3151 // Convert long to machine-word.
3152 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3153
3154 // Store the next_position to the underlying jfr java buffer.
3155 Node* commit_memory;
3156 #ifdef _LP64
3157 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3158 #else
3159 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3160 #endif
3161
3162 // 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.
3163 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3164 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3165 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3166
3167 // And flags with lease constant.
3168 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3169
3170 // Branch on lease to conditionalize returning the leased java buffer.
3171 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3172 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3173 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3174
3175 // False branch, not a lease.
3176 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3177
3178 // True branch, is lease.
3179 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3180 set_control(is_lease);
3181
3182 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3183 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3184 OptoRuntime::void_void_Type(),
3185 SharedRuntime::jfr_return_lease(),
3186 "return_lease", TypePtr::BOTTOM);
3187 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3188
3189 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3190 record_for_igvn(lease_compare_rgn);
3191 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3192 record_for_igvn(lease_compare_mem);
3193 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3194 record_for_igvn(lease_compare_io);
3195 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3196 record_for_igvn(lease_result_value);
3197
3198 // Update control and phi nodes.
3199 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3200 lease_compare_rgn->init_req(_false_path, not_lease);
3201
3202 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3203 lease_compare_mem->init_req(_false_path, commit_memory);
3204
3205 lease_compare_io->init_req(_true_path, i_o());
3206 lease_compare_io->init_req(_false_path, input_io_state);
3207
3208 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3209 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3210
3211 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3212 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3213 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3214 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3215
3216 // Update control and phi nodes.
3217 result_rgn->init_req(_true_path, is_notified);
3218 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3219
3220 result_mem->init_req(_true_path, notified_reset_memory);
3221 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3222
3223 result_io->init_req(_true_path, input_io_state);
3224 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3225
3226 result_value->init_req(_true_path, current_pos);
3227 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3228
3229 // Set output state.
3230 set_control(_gvn.transform(result_rgn));
3231 set_all_memory(_gvn.transform(result_mem));
3232 set_i_o(_gvn.transform(result_io));
3233 set_result(result_rgn, result_value);
3234 return true;
3235 }
3236
3237 /*
3238 * The intrinsic is a model of this pseudo-code:
3239 *
3240 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3241 * jobject h_event_writer = tl->java_event_writer();
3242 * if (h_event_writer == nullptr) {
3243 * return nullptr;
3244 * }
3245 * oop threadObj = Thread::threadObj();
3246 * oop vthread = java_lang_Thread::vthread(threadObj);
3247 * traceid tid;
3248 * bool pinVirtualThread;
3249 * bool excluded;
3250 * if (vthread != threadObj) { // i.e. current thread is virtual
3251 * tid = java_lang_Thread::tid(vthread);
3252 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3253 * pinVirtualThread = VMContinuations;
3254 * excluded = vthread_epoch_raw & excluded_mask;
3255 * if (!excluded) {
3256 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3257 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3258 * if (vthread_epoch != current_epoch) {
3259 * write_checkpoint();
3260 * }
3261 * }
3262 * } else {
3263 * tid = java_lang_Thread::tid(threadObj);
3264 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3265 * pinVirtualThread = false;
3266 * excluded = thread_epoch_raw & excluded_mask;
3267 * }
3268 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3269 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3270 * if (tid_in_event_writer != tid) {
3271 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3272 * setField(event_writer, "excluded", excluded);
3273 * setField(event_writer, "threadID", tid);
3274 * }
3275 * return event_writer
3276 */
3277 bool LibraryCallKit::inline_native_getEventWriter() {
3278 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3279
3280 // Save input memory and i_o state.
3281 Node* input_memory_state = reset_memory();
3282 set_all_memory(input_memory_state);
3283 Node* input_io_state = i_o();
3284
3285 Node* excluded_mask = _gvn.intcon(32768);
3286 Node* epoch_mask = _gvn.intcon(32767);
3287
3288 // TLS
3289 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3290
3291 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3292 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3293
3294 // Load the eventwriter jobject handle.
3295 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3296
3297 // Null check the jobject handle.
3298 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3299 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3300 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3301
3302 // False path, jobj is null.
3303 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3304
3305 // True path, jobj is not null.
3306 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3307
3308 set_control(jobj_is_not_null);
3309
3310 // Load the threadObj for the CarrierThread.
3311 Node* threadObj = generate_current_thread(tls_ptr);
3312
3313 // Load the vthread.
3314 Node* vthread = generate_virtual_thread(tls_ptr);
3315
3316 // If vthread != threadObj, this is a virtual thread.
3317 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3318 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3319 IfNode* iff_vthread_not_equal_threadObj =
3320 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3321
3322 // False branch, fallback to threadObj.
3323 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3324 set_control(vthread_equal_threadObj);
3325
3326 // Load the tid field from the vthread object.
3327 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3328
3329 // Load the raw epoch value from the threadObj.
3330 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3331 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3332 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3333 TypeInt::CHAR, T_CHAR,
3334 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3335
3336 // Mask off the excluded information from the epoch.
3337 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3338
3339 // True branch, this is a virtual thread.
3340 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3341 set_control(vthread_not_equal_threadObj);
3342
3343 // Load the tid field from the vthread object.
3344 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3345
3346 // Continuation support determines if a virtual thread should be pinned.
3347 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3348 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3349
3350 // Load the raw epoch value from the vthread.
3351 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3352 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3353 TypeInt::CHAR, T_CHAR,
3354 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3355
3356 // Mask off the excluded information from the epoch.
3357 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3358
3359 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3360 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3361 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3362 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3363
3364 // False branch, vthread is excluded, no need to write epoch info.
3365 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3366
3367 // True branch, vthread is included, update epoch info.
3368 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3369 set_control(included);
3370
3371 // Get epoch value.
3372 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3373
3374 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3375 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3376 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3377
3378 // Compare the epoch in the vthread to the current epoch generation.
3379 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3380 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3381 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3382
3383 // False path, epoch is equal, checkpoint information is valid.
3384 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3385
3386 // True path, epoch is not equal, write a checkpoint for the vthread.
3387 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3388
3389 set_control(epoch_is_not_equal);
3390
3391 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3392 // The call also updates the native thread local thread id and the vthread with the current epoch.
3393 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3394 OptoRuntime::jfr_write_checkpoint_Type(),
3395 SharedRuntime::jfr_write_checkpoint(),
3396 "write_checkpoint", TypePtr::BOTTOM);
3397 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3398
3399 // vthread epoch != current epoch
3400 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3401 record_for_igvn(epoch_compare_rgn);
3402 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3403 record_for_igvn(epoch_compare_mem);
3404 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3405 record_for_igvn(epoch_compare_io);
3406
3407 // Update control and phi nodes.
3408 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3409 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3410 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3411 epoch_compare_mem->init_req(_false_path, input_memory_state);
3412 epoch_compare_io->init_req(_true_path, i_o());
3413 epoch_compare_io->init_req(_false_path, input_io_state);
3414
3415 // excluded != true
3416 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3417 record_for_igvn(exclude_compare_rgn);
3418 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3419 record_for_igvn(exclude_compare_mem);
3420 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3421 record_for_igvn(exclude_compare_io);
3422
3423 // Update control and phi nodes.
3424 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3425 exclude_compare_rgn->init_req(_false_path, excluded);
3426 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3427 exclude_compare_mem->init_req(_false_path, input_memory_state);
3428 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3429 exclude_compare_io->init_req(_false_path, input_io_state);
3430
3431 // vthread != threadObj
3432 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3433 record_for_igvn(vthread_compare_rgn);
3434 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3435 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3436 record_for_igvn(vthread_compare_io);
3437 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3438 record_for_igvn(tid);
3439 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3440 record_for_igvn(exclusion);
3441 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3442 record_for_igvn(pinVirtualThread);
3443
3444 // Update control and phi nodes.
3445 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3446 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3447 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3448 vthread_compare_mem->init_req(_false_path, input_memory_state);
3449 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3450 vthread_compare_io->init_req(_false_path, input_io_state);
3451 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3452 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3453 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3454 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3455 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3456 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3457
3458 // Update branch state.
3459 set_control(_gvn.transform(vthread_compare_rgn));
3460 set_all_memory(_gvn.transform(vthread_compare_mem));
3461 set_i_o(_gvn.transform(vthread_compare_io));
3462
3463 // Load the event writer oop by dereferencing the jobject handle.
3464 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3465 assert(klass_EventWriter->is_loaded(), "invariant");
3466 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3467 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3468 const TypeOopPtr* const xtype = aklass->as_instance_type();
3469 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3470 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3471
3472 // Load the current thread id from the event writer object.
3473 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3474 // Get the field offset to, conditionally, store an updated tid value later.
3475 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3476 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3477 // Get the field offset to, conditionally, store an updated exclusion value later.
3478 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3479 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3480 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3481 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3482 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3483
3484 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3485 record_for_igvn(event_writer_tid_compare_rgn);
3486 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3487 record_for_igvn(event_writer_tid_compare_mem);
3488 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3489 record_for_igvn(event_writer_tid_compare_io);
3490
3491 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3492 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3493 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3494 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3495
3496 // False path, tids are the same.
3497 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3498
3499 // True path, tid is not equal, need to update the tid in the event writer.
3500 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3501 record_for_igvn(tid_is_not_equal);
3502
3503 // Store the pin state to the event writer.
3504 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3505
3506 // Store the exclusion state to the event writer.
3507 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3508
3509 // Store the tid to the event writer.
3510 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3511
3512 // Update control and phi nodes.
3513 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3514 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3515 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3516 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3517 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3518 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3519
3520 // Result of top level CFG, Memory, IO and Value.
3521 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3522 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3523 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3524 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3525
3526 // Result control.
3527 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3528 result_rgn->init_req(_false_path, jobj_is_null);
3529
3530 // Result memory.
3531 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3532 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3533
3534 // Result IO.
3535 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3536 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3537
3538 // Result value.
3539 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3540 result_value->init_req(_false_path, null()); // return null
3541
3542 // Set output state.
3543 set_control(_gvn.transform(result_rgn));
3544 set_all_memory(_gvn.transform(result_mem));
3545 set_i_o(_gvn.transform(result_io));
3546 set_result(result_rgn, result_value);
3547 return true;
3548 }
3549
3550 /*
3551 * The intrinsic is a model of this pseudo-code:
3552 *
3553 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3554 * if (carrierThread != thread) { // is virtual thread
3555 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3556 * bool excluded = vthread_epoch_raw & excluded_mask;
3557 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3558 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3559 * if (!excluded) {
3560 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3561 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3562 * }
3563 * Atomic::release_store(&tl->_vthread, true);
3564 * return;
3565 * }
3566 * Atomic::release_store(&tl->_vthread, false);
3567 */
3568 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3569 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3570
3571 Node* input_memory_state = reset_memory();
3572 set_all_memory(input_memory_state);
3573
3574 Node* excluded_mask = _gvn.intcon(32768);
3575 Node* epoch_mask = _gvn.intcon(32767);
3576
3577 Node* const carrierThread = generate_current_thread(jt);
3578 // If thread != carrierThread, this is a virtual thread.
3579 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3580 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3581 IfNode* iff_thread_not_equal_carrierThread =
3582 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3583
3584 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3585
3586 // False branch, is carrierThread.
3587 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3588 // Store release
3589 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3590
3591 set_all_memory(input_memory_state);
3592
3593 // True branch, is virtual thread.
3594 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3595 set_control(thread_not_equal_carrierThread);
3596
3597 // Load the raw epoch value from the vthread.
3598 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3599 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3600 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3601
3602 // Mask off the excluded information from the epoch.
3603 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3604
3605 // Load the tid field from the thread.
3606 Node* tid = load_field_from_object(thread, "tid", "J");
3607
3608 // Store the vthread tid to the jfr thread local.
3609 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3610 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3611
3612 // Branch is_excluded to conditionalize updating the epoch .
3613 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3614 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3615 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3616
3617 // True branch, vthread is excluded, no need to write epoch info.
3618 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3619 set_control(excluded);
3620 Node* vthread_is_excluded = _gvn.intcon(1);
3621
3622 // False branch, vthread is included, update epoch info.
3623 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3624 set_control(included);
3625 Node* vthread_is_included = _gvn.intcon(0);
3626
3627 // Get epoch value.
3628 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3629
3630 // Store the vthread epoch to the jfr thread local.
3631 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3632 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3633
3634 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3635 record_for_igvn(excluded_rgn);
3636 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3637 record_for_igvn(excluded_mem);
3638 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3639 record_for_igvn(exclusion);
3640
3641 // Merge the excluded control and memory.
3642 excluded_rgn->init_req(_true_path, excluded);
3643 excluded_rgn->init_req(_false_path, included);
3644 excluded_mem->init_req(_true_path, tid_memory);
3645 excluded_mem->init_req(_false_path, included_memory);
3646 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3647 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3648
3649 // Set intermediate state.
3650 set_control(_gvn.transform(excluded_rgn));
3651 set_all_memory(excluded_mem);
3652
3653 // Store the vthread exclusion state to the jfr thread local.
3654 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3655 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3656
3657 // Store release
3658 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3659
3660 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3661 record_for_igvn(thread_compare_rgn);
3662 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3663 record_for_igvn(thread_compare_mem);
3664 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3665 record_for_igvn(vthread);
3666
3667 // Merge the thread_compare control and memory.
3668 thread_compare_rgn->init_req(_true_path, control());
3669 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3670 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3671 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3672
3673 // Set output state.
3674 set_control(_gvn.transform(thread_compare_rgn));
3675 set_all_memory(_gvn.transform(thread_compare_mem));
3676 }
3677
3678 #endif // JFR_HAVE_INTRINSICS
3679
3680 //------------------------inline_native_currentCarrierThread------------------
3681 bool LibraryCallKit::inline_native_currentCarrierThread() {
3682 Node* junk = nullptr;
3683 set_result(generate_current_thread(junk));
3684 return true;
3685 }
3686
3687 //------------------------inline_native_currentThread------------------
3688 bool LibraryCallKit::inline_native_currentThread() {
3689 Node* junk = nullptr;
3690 set_result(generate_virtual_thread(junk));
3691 return true;
3692 }
3693
3694 //------------------------inline_native_setVthread------------------
3695 bool LibraryCallKit::inline_native_setCurrentThread() {
3696 assert(C->method()->changes_current_thread(),
3697 "method changes current Thread but is not annotated ChangesCurrentThread");
3698 Node* arr = argument(1);
3699 Node* thread = _gvn.transform(new ThreadLocalNode());
3700 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3701 Node* thread_obj_handle
3702 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3703 thread_obj_handle = _gvn.transform(thread_obj_handle);
3704 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3705 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3706 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3707 return true;
3708 }
3709
3710 const Type* LibraryCallKit::scopedValueCache_type() {
3711 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3712 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3713 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3714
3715 // Because we create the scopedValue cache lazily we have to make the
3716 // type of the result BotPTR.
3717 bool xk = etype->klass_is_exact();
3718 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3719 return objects_type;
3720 }
3721
3722 Node* LibraryCallKit::scopedValueCache_helper() {
3723 Node* thread = _gvn.transform(new ThreadLocalNode());
3724 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3725 // We cannot use immutable_memory() because we might flip onto a
3726 // different carrier thread, at which point we'll need to use that
3727 // carrier thread's cache.
3728 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3729 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3730 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3731 }
3732
3733 //------------------------inline_native_scopedValueCache------------------
3734 bool LibraryCallKit::inline_native_scopedValueCache() {
3735 Node* cache_obj_handle = scopedValueCache_helper();
3736 const Type* objects_type = scopedValueCache_type();
3737 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3738
3739 return true;
3740 }
3741
3742 //------------------------inline_native_setScopedValueCache------------------
3743 bool LibraryCallKit::inline_native_setScopedValueCache() {
3744 Node* arr = argument(0);
3745 Node* cache_obj_handle = scopedValueCache_helper();
3746 const Type* objects_type = scopedValueCache_type();
3747
3748 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3749 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3750
3751 return true;
3752 }
3753
3754 //------------------------inline_native_Continuation_pin and unpin-----------
3755
3756 // Shared implementation routine for both pin and unpin.
3757 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3758 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3759
3760 // Save input memory.
3761 Node* input_memory_state = reset_memory();
3762 set_all_memory(input_memory_state);
3763
3764 // TLS
3765 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3766 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3767 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3768
3769 // Null check the last continuation object.
3770 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3771 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3772 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3773
3774 // False path, last continuation is null.
3775 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3776
3777 // True path, last continuation is not null.
3778 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3779
3780 set_control(continuation_is_not_null);
3781
3782 // Load the pin count from the last continuation.
3783 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3784 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3785
3786 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3787 Node* pin_count_rhs;
3788 if (unpin) {
3789 pin_count_rhs = _gvn.intcon(0);
3790 } else {
3791 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3792 }
3793 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3794 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3795 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3796
3797 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3798 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3799 set_control(valid_pin_count);
3800
3801 Node* next_pin_count;
3802 if (unpin) {
3803 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3804 } else {
3805 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3806 }
3807
3808 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
3809
3810 // True branch, pin count over/underflow.
3811 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3812 {
3813 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3814 // which will throw IllegalStateException for pin count over/underflow.
3815 PreserveJVMState pjvms(this);
3816 set_control(pin_count_over_underflow);
3817 set_all_memory(input_memory_state);
3818 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3819 Deoptimization::Action_none);
3820 assert(stopped(), "invariant");
3821 }
3822
3823 // Result of top level CFG and Memory.
3824 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3825 record_for_igvn(result_rgn);
3826 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3827 record_for_igvn(result_mem);
3828
3829 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3830 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3831 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3832 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3833
3834 // Set output state.
3835 set_control(_gvn.transform(result_rgn));
3836 set_all_memory(_gvn.transform(result_mem));
3837
3838 return true;
3839 }
3840
3841 //---------------------------load_mirror_from_klass----------------------------
3842 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3843 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3844 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3845 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3846 // mirror = ((OopHandle)mirror)->resolve();
3847 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3848 }
3849
3850 //-----------------------load_klass_from_mirror_common-------------------------
3851 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3852 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3853 // and branch to the given path on the region.
3854 // If never_see_null, take an uncommon trap on null, so we can optimistically
3855 // compile for the non-null case.
3856 // If the region is null, force never_see_null = true.
3857 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3858 bool never_see_null,
3859 RegionNode* region,
3860 int null_path,
3861 int offset) {
3862 if (region == nullptr) never_see_null = true;
3863 Node* p = basic_plus_adr(mirror, offset);
3864 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3865 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3866 Node* null_ctl = top();
3867 kls = null_check_oop(kls, &null_ctl, never_see_null);
3868 if (region != nullptr) {
3869 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3870 region->init_req(null_path, null_ctl);
3871 } else {
3872 assert(null_ctl == top(), "no loose ends");
3873 }
3874 return kls;
3875 }
3876
3877 //--------------------(inline_native_Class_query helpers)---------------------
3878 // Use this for JVM_ACC_INTERFACE.
3879 // Fall through if (mods & mask) == bits, take the guard otherwise.
3880 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3881 ByteSize offset, const Type* type, BasicType bt) {
3882 // Branch around if the given klass has the given modifier bit set.
3883 // Like generate_guard, adds a new path onto the region.
3884 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3885 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3886 Node* mask = intcon(modifier_mask);
3887 Node* bits = intcon(modifier_bits);
3888 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3889 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3890 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3891 return generate_fair_guard(bol, region);
3892 }
3893 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3894 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3895 Klass::access_flags_offset(), TypeInt::INT, T_INT);
3896 }
3897
3898 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3899 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3900 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3901 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3902 }
3903
3904 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3905 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3906 }
3907
3908 //-------------------------inline_native_Class_query-------------------
3909 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3910 const Type* return_type = TypeInt::BOOL;
3911 Node* prim_return_value = top(); // what happens if it's a primitive class?
3912 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3913 bool expect_prim = false; // most of these guys expect to work on refs
3914
3915 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3916
3917 Node* mirror = argument(0);
3918 Node* obj = top();
3919
3920 switch (id) {
3921 case vmIntrinsics::_isInstance:
3922 // nothing is an instance of a primitive type
3923 prim_return_value = intcon(0);
3924 obj = argument(1);
3925 break;
3926 case vmIntrinsics::_getModifiers:
3927 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3928 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3929 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3930 break;
3931 case vmIntrinsics::_isInterface:
3932 prim_return_value = intcon(0);
3933 break;
3934 case vmIntrinsics::_isArray:
3935 prim_return_value = intcon(0);
3936 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3937 break;
3938 case vmIntrinsics::_isPrimitive:
3939 prim_return_value = intcon(1);
3940 expect_prim = true; // obviously
3941 break;
3942 case vmIntrinsics::_isHidden:
3943 prim_return_value = intcon(0);
3944 break;
3945 case vmIntrinsics::_getSuperclass:
3946 prim_return_value = null();
3947 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3948 break;
3949 case vmIntrinsics::_getClassAccessFlags:
3950 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3951 return_type = TypeInt::INT; // not bool! 6297094
3952 break;
3953 default:
3954 fatal_unexpected_iid(id);
3955 break;
3956 }
3957
3958 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3959 if (mirror_con == nullptr) return false; // cannot happen?
3960
3961 #ifndef PRODUCT
3962 if (C->print_intrinsics() || C->print_inlining()) {
3963 ciType* k = mirror_con->java_mirror_type();
3964 if (k) {
3965 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3966 k->print_name();
3967 tty->cr();
3968 }
3969 }
3970 #endif
3971
3972 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3973 RegionNode* region = new RegionNode(PATH_LIMIT);
3974 record_for_igvn(region);
3975 PhiNode* phi = new PhiNode(region, return_type);
3976
3977 // The mirror will never be null of Reflection.getClassAccessFlags, however
3978 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3979 // if it is. See bug 4774291.
3980
3981 // For Reflection.getClassAccessFlags(), the null check occurs in
3982 // the wrong place; see inline_unsafe_access(), above, for a similar
3983 // situation.
3984 mirror = null_check(mirror);
3985 // If mirror or obj is dead, only null-path is taken.
3986 if (stopped()) return true;
3987
3988 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3989
3990 // Now load the mirror's klass metaobject, and null-check it.
3991 // Side-effects region with the control path if the klass is null.
3992 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3993 // If kls is null, we have a primitive mirror.
3994 phi->init_req(_prim_path, prim_return_value);
3995 if (stopped()) { set_result(region, phi); return true; }
3996 bool safe_for_replace = (region->in(_prim_path) == top());
3997
3998 Node* p; // handy temp
3999 Node* null_ctl;
4000
4001 // Now that we have the non-null klass, we can perform the real query.
4002 // For constant classes, the query will constant-fold in LoadNode::Value.
4003 Node* query_value = top();
4004 switch (id) {
4005 case vmIntrinsics::_isInstance:
4006 // nothing is an instance of a primitive type
4007 query_value = gen_instanceof(obj, kls, safe_for_replace);
4008 break;
4009
4010 case vmIntrinsics::_getModifiers:
4011 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4012 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4013 break;
4014
4015 case vmIntrinsics::_isInterface:
4016 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4017 if (generate_interface_guard(kls, region) != nullptr)
4018 // A guard was added. If the guard is taken, it was an interface.
4019 phi->add_req(intcon(1));
4020 // If we fall through, it's a plain class.
4021 query_value = intcon(0);
4022 break;
4023
4024 case vmIntrinsics::_isArray:
4025 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4026 if (generate_array_guard(kls, region) != nullptr)
4027 // A guard was added. If the guard is taken, it was an array.
4028 phi->add_req(intcon(1));
4029 // If we fall through, it's a plain class.
4030 query_value = intcon(0);
4031 break;
4032
4033 case vmIntrinsics::_isPrimitive:
4034 query_value = intcon(0); // "normal" path produces false
4035 break;
4036
4037 case vmIntrinsics::_isHidden:
4038 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4039 if (generate_hidden_class_guard(kls, region) != nullptr)
4040 // A guard was added. If the guard is taken, it was an hidden class.
4041 phi->add_req(intcon(1));
4042 // If we fall through, it's a plain class.
4043 query_value = intcon(0);
4044 break;
4045
4046
4047 case vmIntrinsics::_getSuperclass:
4048 // The rules here are somewhat unfortunate, but we can still do better
4049 // with random logic than with a JNI call.
4050 // Interfaces store null or Object as _super, but must report null.
4051 // Arrays store an intermediate super as _super, but must report Object.
4052 // Other types can report the actual _super.
4053 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4054 if (generate_interface_guard(kls, region) != nullptr)
4055 // A guard was added. If the guard is taken, it was an interface.
4056 phi->add_req(null());
4057 if (generate_array_guard(kls, region) != nullptr)
4058 // A guard was added. If the guard is taken, it was an array.
4059 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4060 // If we fall through, it's a plain class. Get its _super.
4061 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4062 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4063 null_ctl = top();
4064 kls = null_check_oop(kls, &null_ctl);
4065 if (null_ctl != top()) {
4066 // If the guard is taken, Object.superClass is null (both klass and mirror).
4067 region->add_req(null_ctl);
4068 phi ->add_req(null());
4069 }
4070 if (!stopped()) {
4071 query_value = load_mirror_from_klass(kls);
4072 }
4073 break;
4074
4075 case vmIntrinsics::_getClassAccessFlags:
4076 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4077 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4078 break;
4079
4080 default:
4081 fatal_unexpected_iid(id);
4082 break;
4083 }
4084
4085 // Fall-through is the normal case of a query to a real class.
4086 phi->init_req(1, query_value);
4087 region->init_req(1, control());
4088
4089 C->set_has_split_ifs(true); // Has chance for split-if optimization
4090 set_result(region, phi);
4091 return true;
4092 }
4093
4094 //-------------------------inline_Class_cast-------------------
4095 bool LibraryCallKit::inline_Class_cast() {
4096 Node* mirror = argument(0); // Class
4097 Node* obj = argument(1);
4098 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4099 if (mirror_con == nullptr) {
4100 return false; // dead path (mirror->is_top()).
4101 }
4102 if (obj == nullptr || obj->is_top()) {
4103 return false; // dead path
4104 }
4105 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4106
4107 // First, see if Class.cast() can be folded statically.
4108 // java_mirror_type() returns non-null for compile-time Class constants.
4109 ciType* tm = mirror_con->java_mirror_type();
4110 if (tm != nullptr && tm->is_klass() &&
4111 tp != nullptr) {
4112 if (!tp->is_loaded()) {
4113 // Don't use intrinsic when class is not loaded.
4114 return false;
4115 } else {
4116 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4117 if (static_res == Compile::SSC_always_true) {
4118 // isInstance() is true - fold the code.
4119 set_result(obj);
4120 return true;
4121 } else if (static_res == Compile::SSC_always_false) {
4122 // Don't use intrinsic, have to throw ClassCastException.
4123 // If the reference is null, the non-intrinsic bytecode will
4124 // be optimized appropriately.
4125 return false;
4126 }
4127 }
4128 }
4129
4130 // Bailout intrinsic and do normal inlining if exception path is frequent.
4131 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4132 return false;
4133 }
4134
4135 // Generate dynamic checks.
4136 // Class.cast() is java implementation of _checkcast bytecode.
4137 // Do checkcast (Parse::do_checkcast()) optimizations here.
4138
4139 mirror = null_check(mirror);
4140 // If mirror is dead, only null-path is taken.
4141 if (stopped()) {
4142 return true;
4143 }
4144
4145 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4146 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4147 RegionNode* region = new RegionNode(PATH_LIMIT);
4148 record_for_igvn(region);
4149
4150 // Now load the mirror's klass metaobject, and null-check it.
4151 // If kls is null, we have a primitive mirror and
4152 // nothing is an instance of a primitive type.
4153 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4154
4155 Node* res = top();
4156 if (!stopped()) {
4157 Node* bad_type_ctrl = top();
4158 // Do checkcast optimizations.
4159 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4160 region->init_req(_bad_type_path, bad_type_ctrl);
4161 }
4162 if (region->in(_prim_path) != top() ||
4163 region->in(_bad_type_path) != top()) {
4164 // Let Interpreter throw ClassCastException.
4165 PreserveJVMState pjvms(this);
4166 set_control(_gvn.transform(region));
4167 uncommon_trap(Deoptimization::Reason_intrinsic,
4168 Deoptimization::Action_maybe_recompile);
4169 }
4170 if (!stopped()) {
4171 set_result(res);
4172 }
4173 return true;
4174 }
4175
4176
4177 //--------------------------inline_native_subtype_check------------------------
4178 // This intrinsic takes the JNI calls out of the heart of
4179 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4180 bool LibraryCallKit::inline_native_subtype_check() {
4181 // Pull both arguments off the stack.
4182 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4183 args[0] = argument(0);
4184 args[1] = argument(1);
4185 Node* klasses[2]; // corresponding Klasses: superk, subk
4186 klasses[0] = klasses[1] = top();
4187
4188 enum {
4189 // A full decision tree on {superc is prim, subc is prim}:
4190 _prim_0_path = 1, // {P,N} => false
4191 // {P,P} & superc!=subc => false
4192 _prim_same_path, // {P,P} & superc==subc => true
4193 _prim_1_path, // {N,P} => false
4194 _ref_subtype_path, // {N,N} & subtype check wins => true
4195 _both_ref_path, // {N,N} & subtype check loses => false
4196 PATH_LIMIT
4197 };
4198
4199 RegionNode* region = new RegionNode(PATH_LIMIT);
4200 Node* phi = new PhiNode(region, TypeInt::BOOL);
4201 record_for_igvn(region);
4202
4203 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4204 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4205 int class_klass_offset = java_lang_Class::klass_offset();
4206
4207 // First null-check both mirrors and load each mirror's klass metaobject.
4208 int which_arg;
4209 for (which_arg = 0; which_arg <= 1; which_arg++) {
4210 Node* arg = args[which_arg];
4211 arg = null_check(arg);
4212 if (stopped()) break;
4213 args[which_arg] = arg;
4214
4215 Node* p = basic_plus_adr(arg, class_klass_offset);
4216 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4217 klasses[which_arg] = _gvn.transform(kls);
4218 }
4219
4220 // Having loaded both klasses, test each for null.
4221 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4222 for (which_arg = 0; which_arg <= 1; which_arg++) {
4223 Node* kls = klasses[which_arg];
4224 Node* null_ctl = top();
4225 kls = null_check_oop(kls, &null_ctl, never_see_null);
4226 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4227 region->init_req(prim_path, null_ctl);
4228 if (stopped()) break;
4229 klasses[which_arg] = kls;
4230 }
4231
4232 if (!stopped()) {
4233 // now we have two reference types, in klasses[0..1]
4234 Node* subk = klasses[1]; // the argument to isAssignableFrom
4235 Node* superk = klasses[0]; // the receiver
4236 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4237 // now we have a successful reference subtype check
4238 region->set_req(_ref_subtype_path, control());
4239 }
4240
4241 // If both operands are primitive (both klasses null), then
4242 // we must return true when they are identical primitives.
4243 // It is convenient to test this after the first null klass check.
4244 set_control(region->in(_prim_0_path)); // go back to first null check
4245 if (!stopped()) {
4246 // Since superc is primitive, make a guard for the superc==subc case.
4247 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4248 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4249 generate_guard(bol_eq, region, PROB_FAIR);
4250 if (region->req() == PATH_LIMIT+1) {
4251 // A guard was added. If the added guard is taken, superc==subc.
4252 region->swap_edges(PATH_LIMIT, _prim_same_path);
4253 region->del_req(PATH_LIMIT);
4254 }
4255 region->set_req(_prim_0_path, control()); // Not equal after all.
4256 }
4257
4258 // these are the only paths that produce 'true':
4259 phi->set_req(_prim_same_path, intcon(1));
4260 phi->set_req(_ref_subtype_path, intcon(1));
4261
4262 // pull together the cases:
4263 assert(region->req() == PATH_LIMIT, "sane region");
4264 for (uint i = 1; i < region->req(); i++) {
4265 Node* ctl = region->in(i);
4266 if (ctl == nullptr || ctl == top()) {
4267 region->set_req(i, top());
4268 phi ->set_req(i, top());
4269 } else if (phi->in(i) == nullptr) {
4270 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4271 }
4272 }
4273
4274 set_control(_gvn.transform(region));
4275 set_result(_gvn.transform(phi));
4276 return true;
4277 }
4278
4279 //---------------------generate_array_guard_common------------------------
4280 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4281 bool obj_array, bool not_array) {
4282
4283 if (stopped()) {
4284 return nullptr;
4285 }
4286
4287 // If obj_array/non_array==false/false:
4288 // Branch around if the given klass is in fact an array (either obj or prim).
4289 // If obj_array/non_array==false/true:
4290 // Branch around if the given klass is not an array klass of any kind.
4291 // If obj_array/non_array==true/true:
4292 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4293 // If obj_array/non_array==true/false:
4294 // Branch around if the kls is an oop array (Object[] or subtype)
4295 //
4296 // Like generate_guard, adds a new path onto the region.
4297 jint layout_con = 0;
4298 Node* layout_val = get_layout_helper(kls, layout_con);
4299 if (layout_val == nullptr) {
4300 bool query = (obj_array
4301 ? Klass::layout_helper_is_objArray(layout_con)
4302 : Klass::layout_helper_is_array(layout_con));
4303 if (query == not_array) {
4304 return nullptr; // never a branch
4305 } else { // always a branch
4306 Node* always_branch = control();
4307 if (region != nullptr)
4308 region->add_req(always_branch);
4309 set_control(top());
4310 return always_branch;
4311 }
4312 }
4313 // Now test the correct condition.
4314 jint nval = (obj_array
4315 ? (jint)(Klass::_lh_array_tag_type_value
4316 << Klass::_lh_array_tag_shift)
4317 : Klass::_lh_neutral_value);
4318 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4319 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4320 // invert the test if we are looking for a non-array
4321 if (not_array) btest = BoolTest(btest).negate();
4322 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4323 return generate_fair_guard(bol, region);
4324 }
4325
4326
4327 //-----------------------inline_native_newArray--------------------------
4328 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4329 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4330 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4331 Node* mirror;
4332 Node* count_val;
4333 if (uninitialized) {
4334 null_check_receiver();
4335 mirror = argument(1);
4336 count_val = argument(2);
4337 } else {
4338 mirror = argument(0);
4339 count_val = argument(1);
4340 }
4341
4342 mirror = null_check(mirror);
4343 // If mirror or obj is dead, only null-path is taken.
4344 if (stopped()) return true;
4345
4346 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4347 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4348 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4349 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4350 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4351
4352 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4353 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4354 result_reg, _slow_path);
4355 Node* normal_ctl = control();
4356 Node* no_array_ctl = result_reg->in(_slow_path);
4357
4358 // Generate code for the slow case. We make a call to newArray().
4359 set_control(no_array_ctl);
4360 if (!stopped()) {
4361 // Either the input type is void.class, or else the
4362 // array klass has not yet been cached. Either the
4363 // ensuing call will throw an exception, or else it
4364 // will cache the array klass for next time.
4365 PreserveJVMState pjvms(this);
4366 CallJavaNode* slow_call = nullptr;
4367 if (uninitialized) {
4368 // Generate optimized virtual call (holder class 'Unsafe' is final)
4369 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4370 } else {
4371 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4372 }
4373 Node* slow_result = set_results_for_java_call(slow_call);
4374 // this->control() comes from set_results_for_java_call
4375 result_reg->set_req(_slow_path, control());
4376 result_val->set_req(_slow_path, slow_result);
4377 result_io ->set_req(_slow_path, i_o());
4378 result_mem->set_req(_slow_path, reset_memory());
4379 }
4380
4381 set_control(normal_ctl);
4382 if (!stopped()) {
4383 // Normal case: The array type has been cached in the java.lang.Class.
4384 // The following call works fine even if the array type is polymorphic.
4385 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4386 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4387 result_reg->init_req(_normal_path, control());
4388 result_val->init_req(_normal_path, obj);
4389 result_io ->init_req(_normal_path, i_o());
4390 result_mem->init_req(_normal_path, reset_memory());
4391
4392 if (uninitialized) {
4393 // Mark the allocation so that zeroing is skipped
4394 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4395 alloc->maybe_set_complete(&_gvn);
4396 }
4397 }
4398
4399 // Return the combined state.
4400 set_i_o( _gvn.transform(result_io) );
4401 set_all_memory( _gvn.transform(result_mem));
4402
4403 C->set_has_split_ifs(true); // Has chance for split-if optimization
4404 set_result(result_reg, result_val);
4405 return true;
4406 }
4407
4408 //----------------------inline_native_getLength--------------------------
4409 // public static native int java.lang.reflect.Array.getLength(Object array);
4410 bool LibraryCallKit::inline_native_getLength() {
4411 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4412
4413 Node* array = null_check(argument(0));
4414 // If array is dead, only null-path is taken.
4415 if (stopped()) return true;
4416
4417 // Deoptimize if it is a non-array.
4418 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4419
4420 if (non_array != nullptr) {
4421 PreserveJVMState pjvms(this);
4422 set_control(non_array);
4423 uncommon_trap(Deoptimization::Reason_intrinsic,
4424 Deoptimization::Action_maybe_recompile);
4425 }
4426
4427 // If control is dead, only non-array-path is taken.
4428 if (stopped()) return true;
4429
4430 // The works fine even if the array type is polymorphic.
4431 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4432 Node* result = load_array_length(array);
4433
4434 C->set_has_split_ifs(true); // Has chance for split-if optimization
4435 set_result(result);
4436 return true;
4437 }
4438
4439 //------------------------inline_array_copyOf----------------------------
4440 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4441 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4442 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4443 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4444
4445 // Get the arguments.
4446 Node* original = argument(0);
4447 Node* start = is_copyOfRange? argument(1): intcon(0);
4448 Node* end = is_copyOfRange? argument(2): argument(1);
4449 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4450
4451 Node* newcopy = nullptr;
4452
4453 // Set the original stack and the reexecute bit for the interpreter to reexecute
4454 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4455 { PreserveReexecuteState preexecs(this);
4456 jvms()->set_should_reexecute(true);
4457
4458 array_type_mirror = null_check(array_type_mirror);
4459 original = null_check(original);
4460
4461 // Check if a null path was taken unconditionally.
4462 if (stopped()) return true;
4463
4464 Node* orig_length = load_array_length(original);
4465
4466 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4467 klass_node = null_check(klass_node);
4468
4469 RegionNode* bailout = new RegionNode(1);
4470 record_for_igvn(bailout);
4471
4472 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4473 // Bail out if that is so.
4474 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4475 if (not_objArray != nullptr) {
4476 // Improve the klass node's type from the new optimistic assumption:
4477 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4478 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4479 Node* cast = new CastPPNode(control(), klass_node, akls);
4480 klass_node = _gvn.transform(cast);
4481 }
4482
4483 // Bail out if either start or end is negative.
4484 generate_negative_guard(start, bailout, &start);
4485 generate_negative_guard(end, bailout, &end);
4486
4487 Node* length = end;
4488 if (_gvn.type(start) != TypeInt::ZERO) {
4489 length = _gvn.transform(new SubINode(end, start));
4490 }
4491
4492 // Bail out if length is negative (i.e., if start > end).
4493 // Without this the new_array would throw
4494 // NegativeArraySizeException but IllegalArgumentException is what
4495 // should be thrown
4496 generate_negative_guard(length, bailout, &length);
4497
4498 // Bail out if start is larger than the original length
4499 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4500 generate_negative_guard(orig_tail, bailout, &orig_tail);
4501
4502 if (bailout->req() > 1) {
4503 PreserveJVMState pjvms(this);
4504 set_control(_gvn.transform(bailout));
4505 uncommon_trap(Deoptimization::Reason_intrinsic,
4506 Deoptimization::Action_maybe_recompile);
4507 }
4508
4509 if (!stopped()) {
4510 // How many elements will we copy from the original?
4511 // The answer is MinI(orig_tail, length).
4512 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4513
4514 // Generate a direct call to the right arraycopy function(s).
4515 // We know the copy is disjoint but we might not know if the
4516 // oop stores need checking.
4517 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4518 // This will fail a store-check if x contains any non-nulls.
4519
4520 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4521 // loads/stores but it is legal only if we're sure the
4522 // Arrays.copyOf would succeed. So we need all input arguments
4523 // to the copyOf to be validated, including that the copy to the
4524 // new array won't trigger an ArrayStoreException. That subtype
4525 // check can be optimized if we know something on the type of
4526 // the input array from type speculation.
4527 if (_gvn.type(klass_node)->singleton()) {
4528 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4529 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4530
4531 int test = C->static_subtype_check(superk, subk);
4532 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4533 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4534 if (t_original->speculative_type() != nullptr) {
4535 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4536 }
4537 }
4538 }
4539
4540 bool validated = false;
4541 // Reason_class_check rather than Reason_intrinsic because we
4542 // want to intrinsify even if this traps.
4543 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4544 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4545
4546 if (not_subtype_ctrl != top()) {
4547 PreserveJVMState pjvms(this);
4548 set_control(not_subtype_ctrl);
4549 uncommon_trap(Deoptimization::Reason_class_check,
4550 Deoptimization::Action_make_not_entrant);
4551 assert(stopped(), "Should be stopped");
4552 }
4553 validated = true;
4554 }
4555
4556 if (!stopped()) {
4557 newcopy = new_array(klass_node, length, 0); // no arguments to push
4558
4559 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4560 load_object_klass(original), klass_node);
4561 if (!is_copyOfRange) {
4562 ac->set_copyof(validated);
4563 } else {
4564 ac->set_copyofrange(validated);
4565 }
4566 Node* n = _gvn.transform(ac);
4567 if (n == ac) {
4568 ac->connect_outputs(this);
4569 } else {
4570 assert(validated, "shouldn't transform if all arguments not validated");
4571 set_all_memory(n);
4572 }
4573 }
4574 }
4575 } // original reexecute is set back here
4576
4577 C->set_has_split_ifs(true); // Has chance for split-if optimization
4578 if (!stopped()) {
4579 set_result(newcopy);
4580 }
4581 return true;
4582 }
4583
4584
4585 //----------------------generate_virtual_guard---------------------------
4586 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4587 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4588 RegionNode* slow_region) {
4589 ciMethod* method = callee();
4590 int vtable_index = method->vtable_index();
4591 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4592 "bad index %d", vtable_index);
4593 // Get the Method* out of the appropriate vtable entry.
4594 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4595 vtable_index*vtableEntry::size_in_bytes() +
4596 in_bytes(vtableEntry::method_offset());
4597 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4598 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4599
4600 // Compare the target method with the expected method (e.g., Object.hashCode).
4601 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4602
4603 Node* native_call = makecon(native_call_addr);
4604 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4605 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4606
4607 return generate_slow_guard(test_native, slow_region);
4608 }
4609
4610 //-----------------------generate_method_call----------------------------
4611 // Use generate_method_call to make a slow-call to the real
4612 // method if the fast path fails. An alternative would be to
4613 // use a stub like OptoRuntime::slow_arraycopy_Java.
4614 // This only works for expanding the current library call,
4615 // not another intrinsic. (E.g., don't use this for making an
4616 // arraycopy call inside of the copyOf intrinsic.)
4617 CallJavaNode*
4618 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4619 // When compiling the intrinsic method itself, do not use this technique.
4620 guarantee(callee() != C->method(), "cannot make slow-call to self");
4621
4622 ciMethod* method = callee();
4623 // ensure the JVMS we have will be correct for this call
4624 guarantee(method_id == method->intrinsic_id(), "must match");
4625
4626 const TypeFunc* tf = TypeFunc::make(method);
4627 if (res_not_null) {
4628 assert(tf->return_type() == T_OBJECT, "");
4629 const TypeTuple* range = tf->range();
4630 const Type** fields = TypeTuple::fields(range->cnt());
4631 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4632 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4633 tf = TypeFunc::make(tf->domain(), new_range);
4634 }
4635 CallJavaNode* slow_call;
4636 if (is_static) {
4637 assert(!is_virtual, "");
4638 slow_call = new CallStaticJavaNode(C, tf,
4639 SharedRuntime::get_resolve_static_call_stub(), method);
4640 } else if (is_virtual) {
4641 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4642 int vtable_index = Method::invalid_vtable_index;
4643 if (UseInlineCaches) {
4644 // Suppress the vtable call
4645 } else {
4646 // hashCode and clone are not a miranda methods,
4647 // so the vtable index is fixed.
4648 // No need to use the linkResolver to get it.
4649 vtable_index = method->vtable_index();
4650 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4651 "bad index %d", vtable_index);
4652 }
4653 slow_call = new CallDynamicJavaNode(tf,
4654 SharedRuntime::get_resolve_virtual_call_stub(),
4655 method, vtable_index);
4656 } else { // neither virtual nor static: opt_virtual
4657 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4658 slow_call = new CallStaticJavaNode(C, tf,
4659 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4660 slow_call->set_optimized_virtual(true);
4661 }
4662 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4663 // To be able to issue a direct call (optimized virtual or virtual)
4664 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4665 // about the method being invoked should be attached to the call site to
4666 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4667 slow_call->set_override_symbolic_info(true);
4668 }
4669 set_arguments_for_java_call(slow_call);
4670 set_edges_for_java_call(slow_call);
4671 return slow_call;
4672 }
4673
4674
4675 /**
4676 * Build special case code for calls to hashCode on an object. This call may
4677 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4678 * slightly different code.
4679 */
4680 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4681 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4682 assert(!(is_virtual && is_static), "either virtual, special, or static");
4683
4684 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4685
4686 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4687 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4688 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4689 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4690 Node* obj = nullptr;
4691 if (!is_static) {
4692 // Check for hashing null object
4693 obj = null_check_receiver();
4694 if (stopped()) return true; // unconditionally null
4695 result_reg->init_req(_null_path, top());
4696 result_val->init_req(_null_path, top());
4697 } else {
4698 // Do a null check, and return zero if null.
4699 // System.identityHashCode(null) == 0
4700 obj = argument(0);
4701 Node* null_ctl = top();
4702 obj = null_check_oop(obj, &null_ctl);
4703 result_reg->init_req(_null_path, null_ctl);
4704 result_val->init_req(_null_path, _gvn.intcon(0));
4705 }
4706
4707 // Unconditionally null? Then return right away.
4708 if (stopped()) {
4709 set_control( result_reg->in(_null_path));
4710 if (!stopped())
4711 set_result(result_val->in(_null_path));
4712 return true;
4713 }
4714
4715 // We only go to the fast case code if we pass a number of guards. The
4716 // paths which do not pass are accumulated in the slow_region.
4717 RegionNode* slow_region = new RegionNode(1);
4718 record_for_igvn(slow_region);
4719
4720 // If this is a virtual call, we generate a funny guard. We pull out
4721 // the vtable entry corresponding to hashCode() from the target object.
4722 // If the target method which we are calling happens to be the native
4723 // Object hashCode() method, we pass the guard. We do not need this
4724 // guard for non-virtual calls -- the caller is known to be the native
4725 // Object hashCode().
4726 if (is_virtual) {
4727 // After null check, get the object's klass.
4728 Node* obj_klass = load_object_klass(obj);
4729 generate_virtual_guard(obj_klass, slow_region);
4730 }
4731
4732 // Get the header out of the object, use LoadMarkNode when available
4733 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4734 // The control of the load must be null. Otherwise, the load can move before
4735 // the null check after castPP removal.
4736 Node* no_ctrl = nullptr;
4737 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4738
4739 if (!UseObjectMonitorTable) {
4740 // Test the header to see if it is safe to read w.r.t. locking.
4741 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4742 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4743 if (LockingMode == LM_LIGHTWEIGHT) {
4744 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4745 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4746 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4747
4748 generate_slow_guard(test_monitor, slow_region);
4749 } else {
4750 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4751 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4752 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4753
4754 generate_slow_guard(test_not_unlocked, slow_region);
4755 }
4756 }
4757
4758 // Get the hash value and check to see that it has been properly assigned.
4759 // We depend on hash_mask being at most 32 bits and avoid the use of
4760 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4761 // vm: see markWord.hpp.
4762 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4763 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4764 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4765 // This hack lets the hash bits live anywhere in the mark object now, as long
4766 // as the shift drops the relevant bits into the low 32 bits. Note that
4767 // Java spec says that HashCode is an int so there's no point in capturing
4768 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4769 hshifted_header = ConvX2I(hshifted_header);
4770 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4771
4772 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4773 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4774 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4775
4776 generate_slow_guard(test_assigned, slow_region);
4777
4778 Node* init_mem = reset_memory();
4779 // fill in the rest of the null path:
4780 result_io ->init_req(_null_path, i_o());
4781 result_mem->init_req(_null_path, init_mem);
4782
4783 result_val->init_req(_fast_path, hash_val);
4784 result_reg->init_req(_fast_path, control());
4785 result_io ->init_req(_fast_path, i_o());
4786 result_mem->init_req(_fast_path, init_mem);
4787
4788 // Generate code for the slow case. We make a call to hashCode().
4789 set_control(_gvn.transform(slow_region));
4790 if (!stopped()) {
4791 // No need for PreserveJVMState, because we're using up the present state.
4792 set_all_memory(init_mem);
4793 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4794 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4795 Node* slow_result = set_results_for_java_call(slow_call);
4796 // this->control() comes from set_results_for_java_call
4797 result_reg->init_req(_slow_path, control());
4798 result_val->init_req(_slow_path, slow_result);
4799 result_io ->set_req(_slow_path, i_o());
4800 result_mem ->set_req(_slow_path, reset_memory());
4801 }
4802
4803 // Return the combined state.
4804 set_i_o( _gvn.transform(result_io) );
4805 set_all_memory( _gvn.transform(result_mem));
4806
4807 set_result(result_reg, result_val);
4808 return true;
4809 }
4810
4811 //---------------------------inline_native_getClass----------------------------
4812 // public final native Class<?> java.lang.Object.getClass();
4813 //
4814 // Build special case code for calls to getClass on an object.
4815 bool LibraryCallKit::inline_native_getClass() {
4816 Node* obj = null_check_receiver();
4817 if (stopped()) return true;
4818 set_result(load_mirror_from_klass(load_object_klass(obj)));
4819 return true;
4820 }
4821
4822 //-----------------inline_native_Reflection_getCallerClass---------------------
4823 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4824 //
4825 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4826 //
4827 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4828 // in that it must skip particular security frames and checks for
4829 // caller sensitive methods.
4830 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4831 #ifndef PRODUCT
4832 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4833 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4834 }
4835 #endif
4836
4837 if (!jvms()->has_method()) {
4838 #ifndef PRODUCT
4839 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4840 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4841 }
4842 #endif
4843 return false;
4844 }
4845
4846 // Walk back up the JVM state to find the caller at the required
4847 // depth.
4848 JVMState* caller_jvms = jvms();
4849
4850 // Cf. JVM_GetCallerClass
4851 // NOTE: Start the loop at depth 1 because the current JVM state does
4852 // not include the Reflection.getCallerClass() frame.
4853 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4854 ciMethod* m = caller_jvms->method();
4855 switch (n) {
4856 case 0:
4857 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4858 break;
4859 case 1:
4860 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4861 if (!m->caller_sensitive()) {
4862 #ifndef PRODUCT
4863 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4864 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4865 }
4866 #endif
4867 return false; // bail-out; let JVM_GetCallerClass do the work
4868 }
4869 break;
4870 default:
4871 if (!m->is_ignored_by_security_stack_walk()) {
4872 // We have reached the desired frame; return the holder class.
4873 // Acquire method holder as java.lang.Class and push as constant.
4874 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4875 ciInstance* caller_mirror = caller_klass->java_mirror();
4876 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4877
4878 #ifndef PRODUCT
4879 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4880 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());
4881 tty->print_cr(" JVM state at this point:");
4882 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4883 ciMethod* m = jvms()->of_depth(i)->method();
4884 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4885 }
4886 }
4887 #endif
4888 return true;
4889 }
4890 break;
4891 }
4892 }
4893
4894 #ifndef PRODUCT
4895 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4896 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4897 tty->print_cr(" JVM state at this point:");
4898 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4899 ciMethod* m = jvms()->of_depth(i)->method();
4900 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4901 }
4902 }
4903 #endif
4904
4905 return false; // bail-out; let JVM_GetCallerClass do the work
4906 }
4907
4908 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4909 Node* arg = argument(0);
4910 Node* result = nullptr;
4911
4912 switch (id) {
4913 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4914 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4915 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4916 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4917 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4918 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4919
4920 case vmIntrinsics::_doubleToLongBits: {
4921 // two paths (plus control) merge in a wood
4922 RegionNode *r = new RegionNode(3);
4923 Node *phi = new PhiNode(r, TypeLong::LONG);
4924
4925 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4926 // Build the boolean node
4927 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4928
4929 // Branch either way.
4930 // NaN case is less traveled, which makes all the difference.
4931 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4932 Node *opt_isnan = _gvn.transform(ifisnan);
4933 assert( opt_isnan->is_If(), "Expect an IfNode");
4934 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4935 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4936
4937 set_control(iftrue);
4938
4939 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4940 Node *slow_result = longcon(nan_bits); // return NaN
4941 phi->init_req(1, _gvn.transform( slow_result ));
4942 r->init_req(1, iftrue);
4943
4944 // Else fall through
4945 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4946 set_control(iffalse);
4947
4948 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4949 r->init_req(2, iffalse);
4950
4951 // Post merge
4952 set_control(_gvn.transform(r));
4953 record_for_igvn(r);
4954
4955 C->set_has_split_ifs(true); // Has chance for split-if optimization
4956 result = phi;
4957 assert(result->bottom_type()->isa_long(), "must be");
4958 break;
4959 }
4960
4961 case vmIntrinsics::_floatToIntBits: {
4962 // two paths (plus control) merge in a wood
4963 RegionNode *r = new RegionNode(3);
4964 Node *phi = new PhiNode(r, TypeInt::INT);
4965
4966 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4967 // Build the boolean node
4968 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4969
4970 // Branch either way.
4971 // NaN case is less traveled, which makes all the difference.
4972 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4973 Node *opt_isnan = _gvn.transform(ifisnan);
4974 assert( opt_isnan->is_If(), "Expect an IfNode");
4975 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4976 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4977
4978 set_control(iftrue);
4979
4980 static const jint nan_bits = 0x7fc00000;
4981 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4982 phi->init_req(1, _gvn.transform( slow_result ));
4983 r->init_req(1, iftrue);
4984
4985 // Else fall through
4986 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4987 set_control(iffalse);
4988
4989 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4990 r->init_req(2, iffalse);
4991
4992 // Post merge
4993 set_control(_gvn.transform(r));
4994 record_for_igvn(r);
4995
4996 C->set_has_split_ifs(true); // Has chance for split-if optimization
4997 result = phi;
4998 assert(result->bottom_type()->isa_int(), "must be");
4999 break;
5000 }
5001
5002 default:
5003 fatal_unexpected_iid(id);
5004 break;
5005 }
5006 set_result(_gvn.transform(result));
5007 return true;
5008 }
5009
5010 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5011 Node* arg = argument(0);
5012 Node* result = nullptr;
5013
5014 switch (id) {
5015 case vmIntrinsics::_floatIsInfinite:
5016 result = new IsInfiniteFNode(arg);
5017 break;
5018 case vmIntrinsics::_floatIsFinite:
5019 result = new IsFiniteFNode(arg);
5020 break;
5021 case vmIntrinsics::_doubleIsInfinite:
5022 result = new IsInfiniteDNode(arg);
5023 break;
5024 case vmIntrinsics::_doubleIsFinite:
5025 result = new IsFiniteDNode(arg);
5026 break;
5027 default:
5028 fatal_unexpected_iid(id);
5029 break;
5030 }
5031 set_result(_gvn.transform(result));
5032 return true;
5033 }
5034
5035 //----------------------inline_unsafe_copyMemory-------------------------
5036 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5037
5038 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5039 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5040 const Type* base_t = gvn.type(base);
5041
5042 bool in_native = (base_t == TypePtr::NULL_PTR);
5043 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5044 bool is_mixed = !in_heap && !in_native;
5045
5046 if (is_mixed) {
5047 return true; // mixed accesses can touch both on-heap and off-heap memory
5048 }
5049 if (in_heap) {
5050 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5051 if (!is_prim_array) {
5052 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5053 // there's not enough type information available to determine proper memory slice for it.
5054 return true;
5055 }
5056 }
5057 return false;
5058 }
5059
5060 bool LibraryCallKit::inline_unsafe_copyMemory() {
5061 if (callee()->is_static()) return false; // caller must have the capability!
5062 null_check_receiver(); // null-check receiver
5063 if (stopped()) return true;
5064
5065 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5066
5067 Node* src_base = argument(1); // type: oop
5068 Node* src_off = ConvL2X(argument(2)); // type: long
5069 Node* dst_base = argument(4); // type: oop
5070 Node* dst_off = ConvL2X(argument(5)); // type: long
5071 Node* size = ConvL2X(argument(7)); // type: long
5072
5073 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5074 "fieldOffset must be byte-scaled");
5075
5076 Node* src_addr = make_unsafe_address(src_base, src_off);
5077 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5078
5079 Node* thread = _gvn.transform(new ThreadLocalNode());
5080 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5081 BasicType doing_unsafe_access_bt = T_BYTE;
5082 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5083
5084 // update volatile field
5085 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5086
5087 int flags = RC_LEAF | RC_NO_FP;
5088
5089 const TypePtr* dst_type = TypePtr::BOTTOM;
5090
5091 // Adjust memory effects of the runtime call based on input values.
5092 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5093 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5094 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5095
5096 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5097 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5098 flags |= RC_NARROW_MEM; // narrow in memory
5099 }
5100 }
5101
5102 // Call it. Note that the length argument is not scaled.
5103 make_runtime_call(flags,
5104 OptoRuntime::fast_arraycopy_Type(),
5105 StubRoutines::unsafe_arraycopy(),
5106 "unsafe_arraycopy",
5107 dst_type,
5108 src_addr, dst_addr, size XTOP);
5109
5110 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5111
5112 return true;
5113 }
5114
5115 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5116 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5117 bool LibraryCallKit::inline_unsafe_setMemory() {
5118 if (callee()->is_static()) return false; // caller must have the capability!
5119 null_check_receiver(); // null-check receiver
5120 if (stopped()) return true;
5121
5122 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5123
5124 Node* dst_base = argument(1); // type: oop
5125 Node* dst_off = ConvL2X(argument(2)); // type: long
5126 Node* size = ConvL2X(argument(4)); // type: long
5127 Node* byte = argument(6); // type: byte
5128
5129 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5130 "fieldOffset must be byte-scaled");
5131
5132 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5133
5134 Node* thread = _gvn.transform(new ThreadLocalNode());
5135 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5136 BasicType doing_unsafe_access_bt = T_BYTE;
5137 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5138
5139 // update volatile field
5140 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5141
5142 int flags = RC_LEAF | RC_NO_FP;
5143
5144 const TypePtr* dst_type = TypePtr::BOTTOM;
5145
5146 // Adjust memory effects of the runtime call based on input values.
5147 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5148 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5149
5150 flags |= RC_NARROW_MEM; // narrow in memory
5151 }
5152
5153 // Call it. Note that the length argument is not scaled.
5154 make_runtime_call(flags,
5155 OptoRuntime::make_setmemory_Type(),
5156 StubRoutines::unsafe_setmemory(),
5157 "unsafe_setmemory",
5158 dst_type,
5159 dst_addr, size XTOP, byte);
5160
5161 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5162
5163 return true;
5164 }
5165
5166 #undef XTOP
5167
5168 //------------------------clone_coping-----------------------------------
5169 // Helper function for inline_native_clone.
5170 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5171 assert(obj_size != nullptr, "");
5172 Node* raw_obj = alloc_obj->in(1);
5173 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5174
5175 AllocateNode* alloc = nullptr;
5176 if (ReduceBulkZeroing &&
5177 // If we are implementing an array clone without knowing its source type
5178 // (can happen when compiling the array-guarded branch of a reflective
5179 // Object.clone() invocation), initialize the array within the allocation.
5180 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5181 // to a runtime clone call that assumes fully initialized source arrays.
5182 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5183 // We will be completely responsible for initializing this object -
5184 // mark Initialize node as complete.
5185 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5186 // The object was just allocated - there should be no any stores!
5187 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5188 // Mark as complete_with_arraycopy so that on AllocateNode
5189 // expansion, we know this AllocateNode is initialized by an array
5190 // copy and a StoreStore barrier exists after the array copy.
5191 alloc->initialization()->set_complete_with_arraycopy();
5192 }
5193
5194 Node* size = _gvn.transform(obj_size);
5195 access_clone(obj, alloc_obj, size, is_array);
5196
5197 // Do not let reads from the cloned object float above the arraycopy.
5198 if (alloc != nullptr) {
5199 // Do not let stores that initialize this object be reordered with
5200 // a subsequent store that would make this object accessible by
5201 // other threads.
5202 // Record what AllocateNode this StoreStore protects so that
5203 // escape analysis can go from the MemBarStoreStoreNode to the
5204 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5205 // based on the escape status of the AllocateNode.
5206 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5207 } else {
5208 insert_mem_bar(Op_MemBarCPUOrder);
5209 }
5210 }
5211
5212 //------------------------inline_native_clone----------------------------
5213 // protected native Object java.lang.Object.clone();
5214 //
5215 // Here are the simple edge cases:
5216 // null receiver => normal trap
5217 // virtual and clone was overridden => slow path to out-of-line clone
5218 // not cloneable or finalizer => slow path to out-of-line Object.clone
5219 //
5220 // The general case has two steps, allocation and copying.
5221 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5222 //
5223 // Copying also has two cases, oop arrays and everything else.
5224 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5225 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5226 //
5227 // These steps fold up nicely if and when the cloned object's klass
5228 // can be sharply typed as an object array, a type array, or an instance.
5229 //
5230 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5231 PhiNode* result_val;
5232
5233 // Set the reexecute bit for the interpreter to reexecute
5234 // the bytecode that invokes Object.clone if deoptimization happens.
5235 { PreserveReexecuteState preexecs(this);
5236 jvms()->set_should_reexecute(true);
5237
5238 Node* obj = null_check_receiver();
5239 if (stopped()) return true;
5240
5241 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5242
5243 // If we are going to clone an instance, we need its exact type to
5244 // know the number and types of fields to convert the clone to
5245 // loads/stores. Maybe a speculative type can help us.
5246 if (!obj_type->klass_is_exact() &&
5247 obj_type->speculative_type() != nullptr &&
5248 obj_type->speculative_type()->is_instance_klass()) {
5249 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5250 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5251 !spec_ik->has_injected_fields()) {
5252 if (!obj_type->isa_instptr() ||
5253 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5254 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5255 }
5256 }
5257 }
5258
5259 // Conservatively insert a memory barrier on all memory slices.
5260 // Do not let writes into the original float below the clone.
5261 insert_mem_bar(Op_MemBarCPUOrder);
5262
5263 // paths into result_reg:
5264 enum {
5265 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5266 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5267 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5268 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5269 PATH_LIMIT
5270 };
5271 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5272 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5273 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5274 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5275 record_for_igvn(result_reg);
5276
5277 Node* obj_klass = load_object_klass(obj);
5278 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5279 if (array_ctl != nullptr) {
5280 // It's an array.
5281 PreserveJVMState pjvms(this);
5282 set_control(array_ctl);
5283 Node* obj_length = load_array_length(obj);
5284 Node* array_size = nullptr; // Size of the array without object alignment padding.
5285 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5286
5287 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5288 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5289 // If it is an oop array, it requires very special treatment,
5290 // because gc barriers are required when accessing the array.
5291 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5292 if (is_obja != nullptr) {
5293 PreserveJVMState pjvms2(this);
5294 set_control(is_obja);
5295 // Generate a direct call to the right arraycopy function(s).
5296 // Clones are always tightly coupled.
5297 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5298 ac->set_clone_oop_array();
5299 Node* n = _gvn.transform(ac);
5300 assert(n == ac, "cannot disappear");
5301 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5302
5303 result_reg->init_req(_objArray_path, control());
5304 result_val->init_req(_objArray_path, alloc_obj);
5305 result_i_o ->set_req(_objArray_path, i_o());
5306 result_mem ->set_req(_objArray_path, reset_memory());
5307 }
5308 }
5309 // Otherwise, there are no barriers to worry about.
5310 // (We can dispense with card marks if we know the allocation
5311 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5312 // causes the non-eden paths to take compensating steps to
5313 // simulate a fresh allocation, so that no further
5314 // card marks are required in compiled code to initialize
5315 // the object.)
5316
5317 if (!stopped()) {
5318 copy_to_clone(obj, alloc_obj, array_size, true);
5319
5320 // Present the results of the copy.
5321 result_reg->init_req(_array_path, control());
5322 result_val->init_req(_array_path, alloc_obj);
5323 result_i_o ->set_req(_array_path, i_o());
5324 result_mem ->set_req(_array_path, reset_memory());
5325 }
5326 }
5327
5328 // We only go to the instance fast case code if we pass a number of guards.
5329 // The paths which do not pass are accumulated in the slow_region.
5330 RegionNode* slow_region = new RegionNode(1);
5331 record_for_igvn(slow_region);
5332 if (!stopped()) {
5333 // It's an instance (we did array above). Make the slow-path tests.
5334 // If this is a virtual call, we generate a funny guard. We grab
5335 // the vtable entry corresponding to clone() from the target object.
5336 // If the target method which we are calling happens to be the
5337 // Object clone() method, we pass the guard. We do not need this
5338 // guard for non-virtual calls; the caller is known to be the native
5339 // Object clone().
5340 if (is_virtual) {
5341 generate_virtual_guard(obj_klass, slow_region);
5342 }
5343
5344 // The object must be easily cloneable and must not have a finalizer.
5345 // Both of these conditions may be checked in a single test.
5346 // We could optimize the test further, but we don't care.
5347 generate_misc_flags_guard(obj_klass,
5348 // Test both conditions:
5349 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5350 // Must be cloneable but not finalizer:
5351 KlassFlags::_misc_is_cloneable_fast,
5352 slow_region);
5353 }
5354
5355 if (!stopped()) {
5356 // It's an instance, and it passed the slow-path tests.
5357 PreserveJVMState pjvms(this);
5358 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5359 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5360 // is reexecuted if deoptimization occurs and there could be problems when merging
5361 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5362 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5363
5364 copy_to_clone(obj, alloc_obj, obj_size, false);
5365
5366 // Present the results of the slow call.
5367 result_reg->init_req(_instance_path, control());
5368 result_val->init_req(_instance_path, alloc_obj);
5369 result_i_o ->set_req(_instance_path, i_o());
5370 result_mem ->set_req(_instance_path, reset_memory());
5371 }
5372
5373 // Generate code for the slow case. We make a call to clone().
5374 set_control(_gvn.transform(slow_region));
5375 if (!stopped()) {
5376 PreserveJVMState pjvms(this);
5377 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5378 // We need to deoptimize on exception (see comment above)
5379 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5380 // this->control() comes from set_results_for_java_call
5381 result_reg->init_req(_slow_path, control());
5382 result_val->init_req(_slow_path, slow_result);
5383 result_i_o ->set_req(_slow_path, i_o());
5384 result_mem ->set_req(_slow_path, reset_memory());
5385 }
5386
5387 // Return the combined state.
5388 set_control( _gvn.transform(result_reg));
5389 set_i_o( _gvn.transform(result_i_o));
5390 set_all_memory( _gvn.transform(result_mem));
5391 } // original reexecute is set back here
5392
5393 set_result(_gvn.transform(result_val));
5394 return true;
5395 }
5396
5397 // If we have a tightly coupled allocation, the arraycopy may take care
5398 // of the array initialization. If one of the guards we insert between
5399 // the allocation and the arraycopy causes a deoptimization, an
5400 // uninitialized array will escape the compiled method. To prevent that
5401 // we set the JVM state for uncommon traps between the allocation and
5402 // the arraycopy to the state before the allocation so, in case of
5403 // deoptimization, we'll reexecute the allocation and the
5404 // initialization.
5405 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5406 if (alloc != nullptr) {
5407 ciMethod* trap_method = alloc->jvms()->method();
5408 int trap_bci = alloc->jvms()->bci();
5409
5410 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5411 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5412 // Make sure there's no store between the allocation and the
5413 // arraycopy otherwise visible side effects could be rexecuted
5414 // in case of deoptimization and cause incorrect execution.
5415 bool no_interfering_store = true;
5416 Node* mem = alloc->in(TypeFunc::Memory);
5417 if (mem->is_MergeMem()) {
5418 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5419 Node* n = mms.memory();
5420 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5421 assert(n->is_Store(), "what else?");
5422 no_interfering_store = false;
5423 break;
5424 }
5425 }
5426 } else {
5427 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5428 Node* n = mms.memory();
5429 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5430 assert(n->is_Store(), "what else?");
5431 no_interfering_store = false;
5432 break;
5433 }
5434 }
5435 }
5436
5437 if (no_interfering_store) {
5438 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5439
5440 JVMState* saved_jvms = jvms();
5441 saved_reexecute_sp = _reexecute_sp;
5442
5443 set_jvms(sfpt->jvms());
5444 _reexecute_sp = jvms()->sp();
5445
5446 return saved_jvms;
5447 }
5448 }
5449 }
5450 return nullptr;
5451 }
5452
5453 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5454 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5455 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5456 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5457 uint size = alloc->req();
5458 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5459 old_jvms->set_map(sfpt);
5460 for (uint i = 0; i < size; i++) {
5461 sfpt->init_req(i, alloc->in(i));
5462 }
5463 // re-push array length for deoptimization
5464 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5465 old_jvms->set_sp(old_jvms->sp()+1);
5466 old_jvms->set_monoff(old_jvms->monoff()+1);
5467 old_jvms->set_scloff(old_jvms->scloff()+1);
5468 old_jvms->set_endoff(old_jvms->endoff()+1);
5469 old_jvms->set_should_reexecute(true);
5470
5471 sfpt->set_i_o(map()->i_o());
5472 sfpt->set_memory(map()->memory());
5473 sfpt->set_control(map()->control());
5474 return sfpt;
5475 }
5476
5477 // In case of a deoptimization, we restart execution at the
5478 // allocation, allocating a new array. We would leave an uninitialized
5479 // array in the heap that GCs wouldn't expect. Move the allocation
5480 // after the traps so we don't allocate the array if we
5481 // deoptimize. This is possible because tightly_coupled_allocation()
5482 // guarantees there's no observer of the allocated array at this point
5483 // and the control flow is simple enough.
5484 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5485 int saved_reexecute_sp, uint new_idx) {
5486 if (saved_jvms_before_guards != nullptr && !stopped()) {
5487 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5488
5489 assert(alloc != nullptr, "only with a tightly coupled allocation");
5490 // restore JVM state to the state at the arraycopy
5491 saved_jvms_before_guards->map()->set_control(map()->control());
5492 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5493 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5494 // If we've improved the types of some nodes (null check) while
5495 // emitting the guards, propagate them to the current state
5496 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5497 set_jvms(saved_jvms_before_guards);
5498 _reexecute_sp = saved_reexecute_sp;
5499
5500 // Remove the allocation from above the guards
5501 CallProjections callprojs;
5502 alloc->extract_projections(&callprojs, true);
5503 InitializeNode* init = alloc->initialization();
5504 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5505 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5506 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5507
5508 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5509 // the allocation (i.e. is only valid if the allocation succeeds):
5510 // 1) replace CastIINode with AllocateArrayNode's length here
5511 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5512 //
5513 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5514 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5515 Node* init_control = init->proj_out(TypeFunc::Control);
5516 Node* alloc_length = alloc->Ideal_length();
5517 #ifdef ASSERT
5518 Node* prev_cast = nullptr;
5519 #endif
5520 for (uint i = 0; i < init_control->outcnt(); i++) {
5521 Node* init_out = init_control->raw_out(i);
5522 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5523 #ifdef ASSERT
5524 if (prev_cast == nullptr) {
5525 prev_cast = init_out;
5526 } else {
5527 if (prev_cast->cmp(*init_out) == false) {
5528 prev_cast->dump();
5529 init_out->dump();
5530 assert(false, "not equal CastIINode");
5531 }
5532 }
5533 #endif
5534 C->gvn_replace_by(init_out, alloc_length);
5535 }
5536 }
5537 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5538
5539 // move the allocation here (after the guards)
5540 _gvn.hash_delete(alloc);
5541 alloc->set_req(TypeFunc::Control, control());
5542 alloc->set_req(TypeFunc::I_O, i_o());
5543 Node *mem = reset_memory();
5544 set_all_memory(mem);
5545 alloc->set_req(TypeFunc::Memory, mem);
5546 set_control(init->proj_out_or_null(TypeFunc::Control));
5547 set_i_o(callprojs.fallthrough_ioproj);
5548
5549 // Update memory as done in GraphKit::set_output_for_allocation()
5550 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5551 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5552 if (ary_type->isa_aryptr() && length_type != nullptr) {
5553 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5554 }
5555 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5556 int elemidx = C->get_alias_index(telemref);
5557 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5558 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5559
5560 Node* allocx = _gvn.transform(alloc);
5561 assert(allocx == alloc, "where has the allocation gone?");
5562 assert(dest->is_CheckCastPP(), "not an allocation result?");
5563
5564 _gvn.hash_delete(dest);
5565 dest->set_req(0, control());
5566 Node* destx = _gvn.transform(dest);
5567 assert(destx == dest, "where has the allocation result gone?");
5568
5569 array_ideal_length(alloc, ary_type, true);
5570 }
5571 }
5572
5573 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5574 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5575 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5576 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5577 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5578 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5579 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5580 JVMState* saved_jvms_before_guards) {
5581 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5582 // There is at least one unrelated uncommon trap which needs to be replaced.
5583 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5584
5585 JVMState* saved_jvms = jvms();
5586 const int saved_reexecute_sp = _reexecute_sp;
5587 set_jvms(sfpt->jvms());
5588 _reexecute_sp = jvms()->sp();
5589
5590 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5591
5592 // Restore state
5593 set_jvms(saved_jvms);
5594 _reexecute_sp = saved_reexecute_sp;
5595 }
5596 }
5597
5598 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5599 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5600 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5601 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5602 while (if_proj->is_IfProj()) {
5603 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5604 if (uncommon_trap != nullptr) {
5605 create_new_uncommon_trap(uncommon_trap);
5606 }
5607 assert(if_proj->in(0)->is_If(), "must be If");
5608 if_proj = if_proj->in(0)->in(0);
5609 }
5610 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5611 "must have reached control projection of init node");
5612 }
5613
5614 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5615 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5616 assert(trap_request != 0, "no valid UCT trap request");
5617 PreserveJVMState pjvms(this);
5618 set_control(uncommon_trap_call->in(0));
5619 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5620 Deoptimization::trap_request_action(trap_request));
5621 assert(stopped(), "Should be stopped");
5622 _gvn.hash_delete(uncommon_trap_call);
5623 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5624 }
5625
5626 // Common checks for array sorting intrinsics arguments.
5627 // Returns `true` if checks passed.
5628 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5629 // check address of the class
5630 if (elementType == nullptr || elementType->is_top()) {
5631 return false; // dead path
5632 }
5633 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5634 if (elem_klass == nullptr) {
5635 return false; // dead path
5636 }
5637 // java_mirror_type() returns non-null for compile-time Class constants only
5638 ciType* elem_type = elem_klass->java_mirror_type();
5639 if (elem_type == nullptr) {
5640 return false;
5641 }
5642 bt = elem_type->basic_type();
5643 // Disable the intrinsic if the CPU does not support SIMD sort
5644 if (!Matcher::supports_simd_sort(bt)) {
5645 return false;
5646 }
5647 // check address of the array
5648 if (obj == nullptr || obj->is_top()) {
5649 return false; // dead path
5650 }
5651 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5652 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5653 return false; // failed input validation
5654 }
5655 return true;
5656 }
5657
5658 //------------------------------inline_array_partition-----------------------
5659 bool LibraryCallKit::inline_array_partition() {
5660 address stubAddr = StubRoutines::select_array_partition_function();
5661 if (stubAddr == nullptr) {
5662 return false; // Intrinsic's stub is not implemented on this platform
5663 }
5664 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5665
5666 // no receiver because it is a static method
5667 Node* elementType = argument(0);
5668 Node* obj = argument(1);
5669 Node* offset = argument(2); // long
5670 Node* fromIndex = argument(4);
5671 Node* toIndex = argument(5);
5672 Node* indexPivot1 = argument(6);
5673 Node* indexPivot2 = argument(7);
5674 // PartitionOperation: argument(8) is ignored
5675
5676 Node* pivotIndices = nullptr;
5677 BasicType bt = T_ILLEGAL;
5678
5679 if (!check_array_sort_arguments(elementType, obj, bt)) {
5680 return false;
5681 }
5682 null_check(obj);
5683 // If obj is dead, only null-path is taken.
5684 if (stopped()) {
5685 return true;
5686 }
5687 // Set the original stack and the reexecute bit for the interpreter to reexecute
5688 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5689 { PreserveReexecuteState preexecs(this);
5690 jvms()->set_should_reexecute(true);
5691
5692 Node* obj_adr = make_unsafe_address(obj, offset);
5693
5694 // create the pivotIndices array of type int and size = 2
5695 Node* size = intcon(2);
5696 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5697 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5698 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5699 guarantee(alloc != nullptr, "created above");
5700 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5701
5702 // pass the basic type enum to the stub
5703 Node* elemType = intcon(bt);
5704
5705 // Call the stub
5706 const char *stubName = "array_partition_stub";
5707 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5708 stubAddr, stubName, TypePtr::BOTTOM,
5709 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5710 indexPivot1, indexPivot2);
5711
5712 } // original reexecute is set back here
5713
5714 if (!stopped()) {
5715 set_result(pivotIndices);
5716 }
5717
5718 return true;
5719 }
5720
5721
5722 //------------------------------inline_array_sort-----------------------
5723 bool LibraryCallKit::inline_array_sort() {
5724 address stubAddr = StubRoutines::select_arraysort_function();
5725 if (stubAddr == nullptr) {
5726 return false; // Intrinsic's stub is not implemented on this platform
5727 }
5728 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5729
5730 // no receiver because it is a static method
5731 Node* elementType = argument(0);
5732 Node* obj = argument(1);
5733 Node* offset = argument(2); // long
5734 Node* fromIndex = argument(4);
5735 Node* toIndex = argument(5);
5736 // SortOperation: argument(6) is ignored
5737
5738 BasicType bt = T_ILLEGAL;
5739
5740 if (!check_array_sort_arguments(elementType, obj, bt)) {
5741 return false;
5742 }
5743 null_check(obj);
5744 // If obj is dead, only null-path is taken.
5745 if (stopped()) {
5746 return true;
5747 }
5748 Node* obj_adr = make_unsafe_address(obj, offset);
5749
5750 // pass the basic type enum to the stub
5751 Node* elemType = intcon(bt);
5752
5753 // Call the stub.
5754 const char *stubName = "arraysort_stub";
5755 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5756 stubAddr, stubName, TypePtr::BOTTOM,
5757 obj_adr, elemType, fromIndex, toIndex);
5758
5759 return true;
5760 }
5761
5762
5763 //------------------------------inline_arraycopy-----------------------
5764 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5765 // Object dest, int destPos,
5766 // int length);
5767 bool LibraryCallKit::inline_arraycopy() {
5768 // Get the arguments.
5769 Node* src = argument(0); // type: oop
5770 Node* src_offset = argument(1); // type: int
5771 Node* dest = argument(2); // type: oop
5772 Node* dest_offset = argument(3); // type: int
5773 Node* length = argument(4); // type: int
5774
5775 uint new_idx = C->unique();
5776
5777 // Check for allocation before we add nodes that would confuse
5778 // tightly_coupled_allocation()
5779 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5780
5781 int saved_reexecute_sp = -1;
5782 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5783 // See arraycopy_restore_alloc_state() comment
5784 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5785 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5786 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5787 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5788
5789 // The following tests must be performed
5790 // (1) src and dest are arrays.
5791 // (2) src and dest arrays must have elements of the same BasicType
5792 // (3) src and dest must not be null.
5793 // (4) src_offset must not be negative.
5794 // (5) dest_offset must not be negative.
5795 // (6) length must not be negative.
5796 // (7) src_offset + length must not exceed length of src.
5797 // (8) dest_offset + length must not exceed length of dest.
5798 // (9) each element of an oop array must be assignable
5799
5800 // (3) src and dest must not be null.
5801 // always do this here because we need the JVM state for uncommon traps
5802 Node* null_ctl = top();
5803 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5804 assert(null_ctl->is_top(), "no null control here");
5805 dest = null_check(dest, T_ARRAY);
5806
5807 if (!can_emit_guards) {
5808 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5809 // guards but the arraycopy node could still take advantage of a
5810 // tightly allocated allocation. tightly_coupled_allocation() is
5811 // called again to make sure it takes the null check above into
5812 // account: the null check is mandatory and if it caused an
5813 // uncommon trap to be emitted then the allocation can't be
5814 // considered tightly coupled in this context.
5815 alloc = tightly_coupled_allocation(dest);
5816 }
5817
5818 bool validated = false;
5819
5820 const Type* src_type = _gvn.type(src);
5821 const Type* dest_type = _gvn.type(dest);
5822 const TypeAryPtr* top_src = src_type->isa_aryptr();
5823 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5824
5825 // Do we have the type of src?
5826 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5827 // Do we have the type of dest?
5828 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5829 // Is the type for src from speculation?
5830 bool src_spec = false;
5831 // Is the type for dest from speculation?
5832 bool dest_spec = false;
5833
5834 if ((!has_src || !has_dest) && can_emit_guards) {
5835 // We don't have sufficient type information, let's see if
5836 // speculative types can help. We need to have types for both src
5837 // and dest so that it pays off.
5838
5839 // Do we already have or could we have type information for src
5840 bool could_have_src = has_src;
5841 // Do we already have or could we have type information for dest
5842 bool could_have_dest = has_dest;
5843
5844 ciKlass* src_k = nullptr;
5845 if (!has_src) {
5846 src_k = src_type->speculative_type_not_null();
5847 if (src_k != nullptr && src_k->is_array_klass()) {
5848 could_have_src = true;
5849 }
5850 }
5851
5852 ciKlass* dest_k = nullptr;
5853 if (!has_dest) {
5854 dest_k = dest_type->speculative_type_not_null();
5855 if (dest_k != nullptr && dest_k->is_array_klass()) {
5856 could_have_dest = true;
5857 }
5858 }
5859
5860 if (could_have_src && could_have_dest) {
5861 // This is going to pay off so emit the required guards
5862 if (!has_src) {
5863 src = maybe_cast_profiled_obj(src, src_k, true);
5864 src_type = _gvn.type(src);
5865 top_src = src_type->isa_aryptr();
5866 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5867 src_spec = true;
5868 }
5869 if (!has_dest) {
5870 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5871 dest_type = _gvn.type(dest);
5872 top_dest = dest_type->isa_aryptr();
5873 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5874 dest_spec = true;
5875 }
5876 }
5877 }
5878
5879 if (has_src && has_dest && can_emit_guards) {
5880 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5881 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5882 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5883 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5884
5885 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5886 // If both arrays are object arrays then having the exact types
5887 // for both will remove the need for a subtype check at runtime
5888 // before the call and may make it possible to pick a faster copy
5889 // routine (without a subtype check on every element)
5890 // Do we have the exact type of src?
5891 bool could_have_src = src_spec;
5892 // Do we have the exact type of dest?
5893 bool could_have_dest = dest_spec;
5894 ciKlass* src_k = nullptr;
5895 ciKlass* dest_k = nullptr;
5896 if (!src_spec) {
5897 src_k = src_type->speculative_type_not_null();
5898 if (src_k != nullptr && src_k->is_array_klass()) {
5899 could_have_src = true;
5900 }
5901 }
5902 if (!dest_spec) {
5903 dest_k = dest_type->speculative_type_not_null();
5904 if (dest_k != nullptr && dest_k->is_array_klass()) {
5905 could_have_dest = true;
5906 }
5907 }
5908 if (could_have_src && could_have_dest) {
5909 // If we can have both exact types, emit the missing guards
5910 if (could_have_src && !src_spec) {
5911 src = maybe_cast_profiled_obj(src, src_k, true);
5912 }
5913 if (could_have_dest && !dest_spec) {
5914 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5915 }
5916 }
5917 }
5918 }
5919
5920 ciMethod* trap_method = method();
5921 int trap_bci = bci();
5922 if (saved_jvms_before_guards != nullptr) {
5923 trap_method = alloc->jvms()->method();
5924 trap_bci = alloc->jvms()->bci();
5925 }
5926
5927 bool negative_length_guard_generated = false;
5928
5929 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5930 can_emit_guards &&
5931 !src->is_top() && !dest->is_top()) {
5932 // validate arguments: enables transformation the ArrayCopyNode
5933 validated = true;
5934
5935 RegionNode* slow_region = new RegionNode(1);
5936 record_for_igvn(slow_region);
5937
5938 // (1) src and dest are arrays.
5939 generate_non_array_guard(load_object_klass(src), slow_region);
5940 generate_non_array_guard(load_object_klass(dest), slow_region);
5941
5942 // (2) src and dest arrays must have elements of the same BasicType
5943 // done at macro expansion or at Ideal transformation time
5944
5945 // (4) src_offset must not be negative.
5946 generate_negative_guard(src_offset, slow_region);
5947
5948 // (5) dest_offset must not be negative.
5949 generate_negative_guard(dest_offset, slow_region);
5950
5951 // (7) src_offset + length must not exceed length of src.
5952 generate_limit_guard(src_offset, length,
5953 load_array_length(src),
5954 slow_region);
5955
5956 // (8) dest_offset + length must not exceed length of dest.
5957 generate_limit_guard(dest_offset, length,
5958 load_array_length(dest),
5959 slow_region);
5960
5961 // (6) length must not be negative.
5962 // This is also checked in generate_arraycopy() during macro expansion, but
5963 // we also have to check it here for the case where the ArrayCopyNode will
5964 // be eliminated by Escape Analysis.
5965 if (EliminateAllocations) {
5966 generate_negative_guard(length, slow_region);
5967 negative_length_guard_generated = true;
5968 }
5969
5970 // (9) each element of an oop array must be assignable
5971 Node* dest_klass = load_object_klass(dest);
5972 if (src != dest) {
5973 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5974
5975 if (not_subtype_ctrl != top()) {
5976 PreserveJVMState pjvms(this);
5977 set_control(not_subtype_ctrl);
5978 uncommon_trap(Deoptimization::Reason_intrinsic,
5979 Deoptimization::Action_make_not_entrant);
5980 assert(stopped(), "Should be stopped");
5981 }
5982 }
5983 {
5984 PreserveJVMState pjvms(this);
5985 set_control(_gvn.transform(slow_region));
5986 uncommon_trap(Deoptimization::Reason_intrinsic,
5987 Deoptimization::Action_make_not_entrant);
5988 assert(stopped(), "Should be stopped");
5989 }
5990
5991 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5992 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5993 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5994 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5995 }
5996
5997 if (stopped()) {
5998 return true;
5999 }
6000
6001 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6002 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6003 // so the compiler has a chance to eliminate them: during macro expansion,
6004 // we have to set their control (CastPP nodes are eliminated).
6005 load_object_klass(src), load_object_klass(dest),
6006 load_array_length(src), load_array_length(dest));
6007
6008 ac->set_arraycopy(validated);
6009
6010 Node* n = _gvn.transform(ac);
6011 if (n == ac) {
6012 ac->connect_outputs(this);
6013 } else {
6014 assert(validated, "shouldn't transform if all arguments not validated");
6015 set_all_memory(n);
6016 }
6017 clear_upper_avx();
6018
6019
6020 return true;
6021 }
6022
6023
6024 // Helper function which determines if an arraycopy immediately follows
6025 // an allocation, with no intervening tests or other escapes for the object.
6026 AllocateArrayNode*
6027 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6028 if (stopped()) return nullptr; // no fast path
6029 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6030
6031 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6032 if (alloc == nullptr) return nullptr;
6033
6034 Node* rawmem = memory(Compile::AliasIdxRaw);
6035 // Is the allocation's memory state untouched?
6036 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6037 // Bail out if there have been raw-memory effects since the allocation.
6038 // (Example: There might have been a call or safepoint.)
6039 return nullptr;
6040 }
6041 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6042 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6043 return nullptr;
6044 }
6045
6046 // There must be no unexpected observers of this allocation.
6047 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6048 Node* obs = ptr->fast_out(i);
6049 if (obs != this->map()) {
6050 return nullptr;
6051 }
6052 }
6053
6054 // This arraycopy must unconditionally follow the allocation of the ptr.
6055 Node* alloc_ctl = ptr->in(0);
6056 Node* ctl = control();
6057 while (ctl != alloc_ctl) {
6058 // There may be guards which feed into the slow_region.
6059 // Any other control flow means that we might not get a chance
6060 // to finish initializing the allocated object.
6061 // Various low-level checks bottom out in uncommon traps. These
6062 // are considered safe since we've already checked above that
6063 // there is no unexpected observer of this allocation.
6064 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6065 assert(ctl->in(0)->is_If(), "must be If");
6066 ctl = ctl->in(0)->in(0);
6067 } else {
6068 return nullptr;
6069 }
6070 }
6071
6072 // If we get this far, we have an allocation which immediately
6073 // precedes the arraycopy, and we can take over zeroing the new object.
6074 // The arraycopy will finish the initialization, and provide
6075 // a new control state to which we will anchor the destination pointer.
6076
6077 return alloc;
6078 }
6079
6080 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6081 if (node->is_IfProj()) {
6082 Node* other_proj = node->as_IfProj()->other_if_proj();
6083 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6084 Node* obs = other_proj->fast_out(j);
6085 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6086 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6087 return obs->as_CallStaticJava();
6088 }
6089 }
6090 }
6091 return nullptr;
6092 }
6093
6094 //-------------inline_encodeISOArray-----------------------------------
6095 // encode char[] to byte[] in ISO_8859_1 or ASCII
6096 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6097 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6098 // no receiver since it is static method
6099 Node *src = argument(0);
6100 Node *src_offset = argument(1);
6101 Node *dst = argument(2);
6102 Node *dst_offset = argument(3);
6103 Node *length = argument(4);
6104
6105 src = must_be_not_null(src, true);
6106 dst = must_be_not_null(dst, true);
6107
6108 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6109 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6110 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6111 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6112 // failed array check
6113 return false;
6114 }
6115
6116 // Figure out the size and type of the elements we will be copying.
6117 BasicType src_elem = src_type->elem()->array_element_basic_type();
6118 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6119 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6120 return false;
6121 }
6122
6123 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6124 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6125 // 'src_start' points to src array + scaled offset
6126 // 'dst_start' points to dst array + scaled offset
6127
6128 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6129 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6130 enc = _gvn.transform(enc);
6131 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6132 set_memory(res_mem, mtype);
6133 set_result(enc);
6134 clear_upper_avx();
6135
6136 return true;
6137 }
6138
6139 //-------------inline_multiplyToLen-----------------------------------
6140 bool LibraryCallKit::inline_multiplyToLen() {
6141 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6142
6143 address stubAddr = StubRoutines::multiplyToLen();
6144 if (stubAddr == nullptr) {
6145 return false; // Intrinsic's stub is not implemented on this platform
6146 }
6147 const char* stubName = "multiplyToLen";
6148
6149 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6150
6151 // no receiver because it is a static method
6152 Node* x = argument(0);
6153 Node* xlen = argument(1);
6154 Node* y = argument(2);
6155 Node* ylen = argument(3);
6156 Node* z = argument(4);
6157
6158 x = must_be_not_null(x, true);
6159 y = must_be_not_null(y, true);
6160
6161 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6162 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6163 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6164 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6165 // failed array check
6166 return false;
6167 }
6168
6169 BasicType x_elem = x_type->elem()->array_element_basic_type();
6170 BasicType y_elem = y_type->elem()->array_element_basic_type();
6171 if (x_elem != T_INT || y_elem != T_INT) {
6172 return false;
6173 }
6174
6175 Node* x_start = array_element_address(x, intcon(0), x_elem);
6176 Node* y_start = array_element_address(y, intcon(0), y_elem);
6177 // 'x_start' points to x array + scaled xlen
6178 // 'y_start' points to y array + scaled ylen
6179
6180 Node* z_start = array_element_address(z, intcon(0), T_INT);
6181
6182 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6183 OptoRuntime::multiplyToLen_Type(),
6184 stubAddr, stubName, TypePtr::BOTTOM,
6185 x_start, xlen, y_start, ylen, z_start);
6186
6187 C->set_has_split_ifs(true); // Has chance for split-if optimization
6188 set_result(z);
6189 return true;
6190 }
6191
6192 //-------------inline_squareToLen------------------------------------
6193 bool LibraryCallKit::inline_squareToLen() {
6194 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6195
6196 address stubAddr = StubRoutines::squareToLen();
6197 if (stubAddr == nullptr) {
6198 return false; // Intrinsic's stub is not implemented on this platform
6199 }
6200 const char* stubName = "squareToLen";
6201
6202 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6203
6204 Node* x = argument(0);
6205 Node* len = argument(1);
6206 Node* z = argument(2);
6207 Node* zlen = argument(3);
6208
6209 x = must_be_not_null(x, true);
6210 z = must_be_not_null(z, true);
6211
6212 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6213 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6214 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6215 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6216 // failed array check
6217 return false;
6218 }
6219
6220 BasicType x_elem = x_type->elem()->array_element_basic_type();
6221 BasicType z_elem = z_type->elem()->array_element_basic_type();
6222 if (x_elem != T_INT || z_elem != T_INT) {
6223 return false;
6224 }
6225
6226
6227 Node* x_start = array_element_address(x, intcon(0), x_elem);
6228 Node* z_start = array_element_address(z, intcon(0), z_elem);
6229
6230 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6231 OptoRuntime::squareToLen_Type(),
6232 stubAddr, stubName, TypePtr::BOTTOM,
6233 x_start, len, z_start, zlen);
6234
6235 set_result(z);
6236 return true;
6237 }
6238
6239 //-------------inline_mulAdd------------------------------------------
6240 bool LibraryCallKit::inline_mulAdd() {
6241 assert(UseMulAddIntrinsic, "not implemented on this platform");
6242
6243 address stubAddr = StubRoutines::mulAdd();
6244 if (stubAddr == nullptr) {
6245 return false; // Intrinsic's stub is not implemented on this platform
6246 }
6247 const char* stubName = "mulAdd";
6248
6249 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6250
6251 Node* out = argument(0);
6252 Node* in = argument(1);
6253 Node* offset = argument(2);
6254 Node* len = argument(3);
6255 Node* k = argument(4);
6256
6257 in = must_be_not_null(in, true);
6258 out = must_be_not_null(out, true);
6259
6260 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6261 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6262 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6263 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6264 // failed array check
6265 return false;
6266 }
6267
6268 BasicType out_elem = out_type->elem()->array_element_basic_type();
6269 BasicType in_elem = in_type->elem()->array_element_basic_type();
6270 if (out_elem != T_INT || in_elem != T_INT) {
6271 return false;
6272 }
6273
6274 Node* outlen = load_array_length(out);
6275 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6276 Node* out_start = array_element_address(out, intcon(0), out_elem);
6277 Node* in_start = array_element_address(in, intcon(0), in_elem);
6278
6279 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6280 OptoRuntime::mulAdd_Type(),
6281 stubAddr, stubName, TypePtr::BOTTOM,
6282 out_start,in_start, new_offset, len, k);
6283 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6284 set_result(result);
6285 return true;
6286 }
6287
6288 //-------------inline_montgomeryMultiply-----------------------------------
6289 bool LibraryCallKit::inline_montgomeryMultiply() {
6290 address stubAddr = StubRoutines::montgomeryMultiply();
6291 if (stubAddr == nullptr) {
6292 return false; // Intrinsic's stub is not implemented on this platform
6293 }
6294
6295 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6296 const char* stubName = "montgomery_multiply";
6297
6298 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6299
6300 Node* a = argument(0);
6301 Node* b = argument(1);
6302 Node* n = argument(2);
6303 Node* len = argument(3);
6304 Node* inv = argument(4);
6305 Node* m = argument(6);
6306
6307 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6308 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6309 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6310 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6311 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6312 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6313 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6314 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6315 // failed array check
6316 return false;
6317 }
6318
6319 BasicType a_elem = a_type->elem()->array_element_basic_type();
6320 BasicType b_elem = b_type->elem()->array_element_basic_type();
6321 BasicType n_elem = n_type->elem()->array_element_basic_type();
6322 BasicType m_elem = m_type->elem()->array_element_basic_type();
6323 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6324 return false;
6325 }
6326
6327 // Make the call
6328 {
6329 Node* a_start = array_element_address(a, intcon(0), a_elem);
6330 Node* b_start = array_element_address(b, intcon(0), b_elem);
6331 Node* n_start = array_element_address(n, intcon(0), n_elem);
6332 Node* m_start = array_element_address(m, intcon(0), m_elem);
6333
6334 Node* call = make_runtime_call(RC_LEAF,
6335 OptoRuntime::montgomeryMultiply_Type(),
6336 stubAddr, stubName, TypePtr::BOTTOM,
6337 a_start, b_start, n_start, len, inv, top(),
6338 m_start);
6339 set_result(m);
6340 }
6341
6342 return true;
6343 }
6344
6345 bool LibraryCallKit::inline_montgomerySquare() {
6346 address stubAddr = StubRoutines::montgomerySquare();
6347 if (stubAddr == nullptr) {
6348 return false; // Intrinsic's stub is not implemented on this platform
6349 }
6350
6351 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6352 const char* stubName = "montgomery_square";
6353
6354 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6355
6356 Node* a = argument(0);
6357 Node* n = argument(1);
6358 Node* len = argument(2);
6359 Node* inv = argument(3);
6360 Node* m = argument(5);
6361
6362 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6363 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6364 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6365 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6366 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6367 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6368 // failed array check
6369 return false;
6370 }
6371
6372 BasicType a_elem = a_type->elem()->array_element_basic_type();
6373 BasicType n_elem = n_type->elem()->array_element_basic_type();
6374 BasicType m_elem = m_type->elem()->array_element_basic_type();
6375 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6376 return false;
6377 }
6378
6379 // Make the call
6380 {
6381 Node* a_start = array_element_address(a, intcon(0), a_elem);
6382 Node* n_start = array_element_address(n, intcon(0), n_elem);
6383 Node* m_start = array_element_address(m, intcon(0), m_elem);
6384
6385 Node* call = make_runtime_call(RC_LEAF,
6386 OptoRuntime::montgomerySquare_Type(),
6387 stubAddr, stubName, TypePtr::BOTTOM,
6388 a_start, n_start, len, inv, top(),
6389 m_start);
6390 set_result(m);
6391 }
6392
6393 return true;
6394 }
6395
6396 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6397 address stubAddr = nullptr;
6398 const char* stubName = nullptr;
6399
6400 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6401 if (stubAddr == nullptr) {
6402 return false; // Intrinsic's stub is not implemented on this platform
6403 }
6404
6405 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6406
6407 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6408
6409 Node* newArr = argument(0);
6410 Node* oldArr = argument(1);
6411 Node* newIdx = argument(2);
6412 Node* shiftCount = argument(3);
6413 Node* numIter = argument(4);
6414
6415 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6416 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6417 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6418 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6419 return false;
6420 }
6421
6422 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6423 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6424 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6425 return false;
6426 }
6427
6428 // Make the call
6429 {
6430 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6431 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6432
6433 Node* call = make_runtime_call(RC_LEAF,
6434 OptoRuntime::bigIntegerShift_Type(),
6435 stubAddr,
6436 stubName,
6437 TypePtr::BOTTOM,
6438 newArr_start,
6439 oldArr_start,
6440 newIdx,
6441 shiftCount,
6442 numIter);
6443 }
6444
6445 return true;
6446 }
6447
6448 //-------------inline_vectorizedMismatch------------------------------
6449 bool LibraryCallKit::inline_vectorizedMismatch() {
6450 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6451
6452 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6453 Node* obja = argument(0); // Object
6454 Node* aoffset = argument(1); // long
6455 Node* objb = argument(3); // Object
6456 Node* boffset = argument(4); // long
6457 Node* length = argument(6); // int
6458 Node* scale = argument(7); // int
6459
6460 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6461 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6462 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6463 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6464 scale == top()) {
6465 return false; // failed input validation
6466 }
6467
6468 Node* obja_adr = make_unsafe_address(obja, aoffset);
6469 Node* objb_adr = make_unsafe_address(objb, boffset);
6470
6471 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6472 //
6473 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6474 // if (length <= inline_limit) {
6475 // inline_path:
6476 // vmask = VectorMaskGen length
6477 // vload1 = LoadVectorMasked obja, vmask
6478 // vload2 = LoadVectorMasked objb, vmask
6479 // result1 = VectorCmpMasked vload1, vload2, vmask
6480 // } else {
6481 // call_stub_path:
6482 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6483 // }
6484 // exit_block:
6485 // return Phi(result1, result2);
6486 //
6487 enum { inline_path = 1, // input is small enough to process it all at once
6488 stub_path = 2, // input is too large; call into the VM
6489 PATH_LIMIT = 3
6490 };
6491
6492 Node* exit_block = new RegionNode(PATH_LIMIT);
6493 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6494 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6495
6496 Node* call_stub_path = control();
6497
6498 BasicType elem_bt = T_ILLEGAL;
6499
6500 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6501 if (scale_t->is_con()) {
6502 switch (scale_t->get_con()) {
6503 case 0: elem_bt = T_BYTE; break;
6504 case 1: elem_bt = T_SHORT; break;
6505 case 2: elem_bt = T_INT; break;
6506 case 3: elem_bt = T_LONG; break;
6507
6508 default: elem_bt = T_ILLEGAL; break; // not supported
6509 }
6510 }
6511
6512 int inline_limit = 0;
6513 bool do_partial_inline = false;
6514
6515 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6516 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6517 do_partial_inline = inline_limit >= 16;
6518 }
6519
6520 if (do_partial_inline) {
6521 assert(elem_bt != T_ILLEGAL, "sanity");
6522
6523 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6524 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6525 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6526
6527 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6528 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6529 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6530
6531 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6532
6533 if (!stopped()) {
6534 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6535
6536 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6537 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6538 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6539 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6540
6541 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6542 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6543 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6544 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6545
6546 exit_block->init_req(inline_path, control());
6547 memory_phi->init_req(inline_path, map()->memory());
6548 result_phi->init_req(inline_path, result);
6549
6550 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6551 clear_upper_avx();
6552 }
6553 }
6554 }
6555
6556 if (call_stub_path != nullptr) {
6557 set_control(call_stub_path);
6558
6559 Node* call = make_runtime_call(RC_LEAF,
6560 OptoRuntime::vectorizedMismatch_Type(),
6561 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6562 obja_adr, objb_adr, length, scale);
6563
6564 exit_block->init_req(stub_path, control());
6565 memory_phi->init_req(stub_path, map()->memory());
6566 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6567 }
6568
6569 exit_block = _gvn.transform(exit_block);
6570 memory_phi = _gvn.transform(memory_phi);
6571 result_phi = _gvn.transform(result_phi);
6572
6573 set_control(exit_block);
6574 set_all_memory(memory_phi);
6575 set_result(result_phi);
6576
6577 return true;
6578 }
6579
6580 //------------------------------inline_vectorizedHashcode----------------------------
6581 bool LibraryCallKit::inline_vectorizedHashCode() {
6582 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6583
6584 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6585 Node* array = argument(0);
6586 Node* offset = argument(1);
6587 Node* length = argument(2);
6588 Node* initialValue = argument(3);
6589 Node* basic_type = argument(4);
6590
6591 if (basic_type == top()) {
6592 return false; // failed input validation
6593 }
6594
6595 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6596 if (!basic_type_t->is_con()) {
6597 return false; // Only intrinsify if mode argument is constant
6598 }
6599
6600 array = must_be_not_null(array, true);
6601
6602 BasicType bt = (BasicType)basic_type_t->get_con();
6603
6604 // Resolve address of first element
6605 Node* array_start = array_element_address(array, offset, bt);
6606
6607 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6608 array_start, length, initialValue, basic_type)));
6609 clear_upper_avx();
6610
6611 return true;
6612 }
6613
6614 /**
6615 * Calculate CRC32 for byte.
6616 * int java.util.zip.CRC32.update(int crc, int b)
6617 */
6618 bool LibraryCallKit::inline_updateCRC32() {
6619 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6620 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6621 // no receiver since it is static method
6622 Node* crc = argument(0); // type: int
6623 Node* b = argument(1); // type: int
6624
6625 /*
6626 * int c = ~ crc;
6627 * b = timesXtoThe32[(b ^ c) & 0xFF];
6628 * b = b ^ (c >>> 8);
6629 * crc = ~b;
6630 */
6631
6632 Node* M1 = intcon(-1);
6633 crc = _gvn.transform(new XorINode(crc, M1));
6634 Node* result = _gvn.transform(new XorINode(crc, b));
6635 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6636
6637 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6638 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6639 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6640 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6641
6642 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6643 result = _gvn.transform(new XorINode(crc, result));
6644 result = _gvn.transform(new XorINode(result, M1));
6645 set_result(result);
6646 return true;
6647 }
6648
6649 /**
6650 * Calculate CRC32 for byte[] array.
6651 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6652 */
6653 bool LibraryCallKit::inline_updateBytesCRC32() {
6654 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6655 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6656 // no receiver since it is static method
6657 Node* crc = argument(0); // type: int
6658 Node* src = argument(1); // type: oop
6659 Node* offset = argument(2); // type: int
6660 Node* length = argument(3); // type: int
6661
6662 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6663 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6664 // failed array check
6665 return false;
6666 }
6667
6668 // Figure out the size and type of the elements we will be copying.
6669 BasicType src_elem = src_type->elem()->array_element_basic_type();
6670 if (src_elem != T_BYTE) {
6671 return false;
6672 }
6673
6674 // 'src_start' points to src array + scaled offset
6675 src = must_be_not_null(src, true);
6676 Node* src_start = array_element_address(src, offset, src_elem);
6677
6678 // We assume that range check is done by caller.
6679 // TODO: generate range check (offset+length < src.length) in debug VM.
6680
6681 // Call the stub.
6682 address stubAddr = StubRoutines::updateBytesCRC32();
6683 const char *stubName = "updateBytesCRC32";
6684
6685 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6686 stubAddr, stubName, TypePtr::BOTTOM,
6687 crc, src_start, length);
6688 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6689 set_result(result);
6690 return true;
6691 }
6692
6693 /**
6694 * Calculate CRC32 for ByteBuffer.
6695 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6696 */
6697 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6698 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6699 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6700 // no receiver since it is static method
6701 Node* crc = argument(0); // type: int
6702 Node* src = argument(1); // type: long
6703 Node* offset = argument(3); // type: int
6704 Node* length = argument(4); // type: int
6705
6706 src = ConvL2X(src); // adjust Java long to machine word
6707 Node* base = _gvn.transform(new CastX2PNode(src));
6708 offset = ConvI2X(offset);
6709
6710 // 'src_start' points to src array + scaled offset
6711 Node* src_start = basic_plus_adr(top(), base, offset);
6712
6713 // Call the stub.
6714 address stubAddr = StubRoutines::updateBytesCRC32();
6715 const char *stubName = "updateBytesCRC32";
6716
6717 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6718 stubAddr, stubName, TypePtr::BOTTOM,
6719 crc, src_start, length);
6720 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6721 set_result(result);
6722 return true;
6723 }
6724
6725 //------------------------------get_table_from_crc32c_class-----------------------
6726 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6727 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6728 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6729
6730 return table;
6731 }
6732
6733 //------------------------------inline_updateBytesCRC32C-----------------------
6734 //
6735 // Calculate CRC32C for byte[] array.
6736 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6737 //
6738 bool LibraryCallKit::inline_updateBytesCRC32C() {
6739 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6740 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6741 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6742 // no receiver since it is a static method
6743 Node* crc = argument(0); // type: int
6744 Node* src = argument(1); // type: oop
6745 Node* offset = argument(2); // type: int
6746 Node* end = argument(3); // type: int
6747
6748 Node* length = _gvn.transform(new SubINode(end, offset));
6749
6750 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6751 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6752 // failed array check
6753 return false;
6754 }
6755
6756 // Figure out the size and type of the elements we will be copying.
6757 BasicType src_elem = src_type->elem()->array_element_basic_type();
6758 if (src_elem != T_BYTE) {
6759 return false;
6760 }
6761
6762 // 'src_start' points to src array + scaled offset
6763 src = must_be_not_null(src, true);
6764 Node* src_start = array_element_address(src, offset, src_elem);
6765
6766 // static final int[] byteTable in class CRC32C
6767 Node* table = get_table_from_crc32c_class(callee()->holder());
6768 table = must_be_not_null(table, true);
6769 Node* table_start = array_element_address(table, intcon(0), T_INT);
6770
6771 // We assume that range check is done by caller.
6772 // TODO: generate range check (offset+length < src.length) in debug VM.
6773
6774 // Call the stub.
6775 address stubAddr = StubRoutines::updateBytesCRC32C();
6776 const char *stubName = "updateBytesCRC32C";
6777
6778 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6779 stubAddr, stubName, TypePtr::BOTTOM,
6780 crc, src_start, length, table_start);
6781 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6782 set_result(result);
6783 return true;
6784 }
6785
6786 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6787 //
6788 // Calculate CRC32C for DirectByteBuffer.
6789 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6790 //
6791 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6792 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6793 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6794 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6795 // no receiver since it is a static method
6796 Node* crc = argument(0); // type: int
6797 Node* src = argument(1); // type: long
6798 Node* offset = argument(3); // type: int
6799 Node* end = argument(4); // type: int
6800
6801 Node* length = _gvn.transform(new SubINode(end, offset));
6802
6803 src = ConvL2X(src); // adjust Java long to machine word
6804 Node* base = _gvn.transform(new CastX2PNode(src));
6805 offset = ConvI2X(offset);
6806
6807 // 'src_start' points to src array + scaled offset
6808 Node* src_start = basic_plus_adr(top(), base, offset);
6809
6810 // static final int[] byteTable in class CRC32C
6811 Node* table = get_table_from_crc32c_class(callee()->holder());
6812 table = must_be_not_null(table, true);
6813 Node* table_start = array_element_address(table, intcon(0), T_INT);
6814
6815 // Call the stub.
6816 address stubAddr = StubRoutines::updateBytesCRC32C();
6817 const char *stubName = "updateBytesCRC32C";
6818
6819 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6820 stubAddr, stubName, TypePtr::BOTTOM,
6821 crc, src_start, length, table_start);
6822 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6823 set_result(result);
6824 return true;
6825 }
6826
6827 //------------------------------inline_updateBytesAdler32----------------------
6828 //
6829 // Calculate Adler32 checksum for byte[] array.
6830 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6831 //
6832 bool LibraryCallKit::inline_updateBytesAdler32() {
6833 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6834 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6835 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6836 // no receiver since it is static method
6837 Node* crc = argument(0); // type: int
6838 Node* src = argument(1); // type: oop
6839 Node* offset = argument(2); // type: int
6840 Node* length = argument(3); // type: int
6841
6842 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6843 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6844 // failed array check
6845 return false;
6846 }
6847
6848 // Figure out the size and type of the elements we will be copying.
6849 BasicType src_elem = src_type->elem()->array_element_basic_type();
6850 if (src_elem != T_BYTE) {
6851 return false;
6852 }
6853
6854 // 'src_start' points to src array + scaled offset
6855 Node* src_start = array_element_address(src, offset, src_elem);
6856
6857 // We assume that range check is done by caller.
6858 // TODO: generate range check (offset+length < src.length) in debug VM.
6859
6860 // Call the stub.
6861 address stubAddr = StubRoutines::updateBytesAdler32();
6862 const char *stubName = "updateBytesAdler32";
6863
6864 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6865 stubAddr, stubName, TypePtr::BOTTOM,
6866 crc, src_start, length);
6867 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6868 set_result(result);
6869 return true;
6870 }
6871
6872 //------------------------------inline_updateByteBufferAdler32---------------
6873 //
6874 // Calculate Adler32 checksum for DirectByteBuffer.
6875 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6876 //
6877 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6878 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6879 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6880 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6881 // no receiver since it is static method
6882 Node* crc = argument(0); // type: int
6883 Node* src = argument(1); // type: long
6884 Node* offset = argument(3); // type: int
6885 Node* length = argument(4); // type: int
6886
6887 src = ConvL2X(src); // adjust Java long to machine word
6888 Node* base = _gvn.transform(new CastX2PNode(src));
6889 offset = ConvI2X(offset);
6890
6891 // 'src_start' points to src array + scaled offset
6892 Node* src_start = basic_plus_adr(top(), base, offset);
6893
6894 // Call the stub.
6895 address stubAddr = StubRoutines::updateBytesAdler32();
6896 const char *stubName = "updateBytesAdler32";
6897
6898 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6899 stubAddr, stubName, TypePtr::BOTTOM,
6900 crc, src_start, length);
6901
6902 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6903 set_result(result);
6904 return true;
6905 }
6906
6907 //----------------------------inline_reference_get----------------------------
6908 // public T java.lang.ref.Reference.get();
6909 bool LibraryCallKit::inline_reference_get() {
6910 const int referent_offset = java_lang_ref_Reference::referent_offset();
6911
6912 // Get the argument:
6913 Node* reference_obj = null_check_receiver();
6914 if (stopped()) return true;
6915
6916 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6917 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6918 decorators, /*is_static*/ false, nullptr);
6919 if (result == nullptr) return false;
6920
6921 // Add memory barrier to prevent commoning reads from this field
6922 // across safepoint since GC can change its value.
6923 insert_mem_bar(Op_MemBarCPUOrder);
6924
6925 set_result(result);
6926 return true;
6927 }
6928
6929 //----------------------------inline_reference_refersTo0----------------------------
6930 // bool java.lang.ref.Reference.refersTo0();
6931 // bool java.lang.ref.PhantomReference.refersTo0();
6932 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6933 // Get arguments:
6934 Node* reference_obj = null_check_receiver();
6935 Node* other_obj = argument(1);
6936 if (stopped()) return true;
6937
6938 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6939 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6940 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6941 decorators, /*is_static*/ false, nullptr);
6942 if (referent == nullptr) return false;
6943
6944 // Add memory barrier to prevent commoning reads from this field
6945 // across safepoint since GC can change its value.
6946 insert_mem_bar(Op_MemBarCPUOrder);
6947
6948 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6949 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6950 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6951
6952 RegionNode* region = new RegionNode(3);
6953 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6954
6955 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6956 region->init_req(1, if_true);
6957 phi->init_req(1, intcon(1));
6958
6959 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6960 region->init_req(2, if_false);
6961 phi->init_req(2, intcon(0));
6962
6963 set_control(_gvn.transform(region));
6964 record_for_igvn(region);
6965 set_result(_gvn.transform(phi));
6966 return true;
6967 }
6968
6969 //----------------------------inline_reference_clear0----------------------------
6970 // void java.lang.ref.Reference.clear0();
6971 // void java.lang.ref.PhantomReference.clear0();
6972 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
6973 // This matches the implementation in JVM_ReferenceClear, see the comments there.
6974
6975 // Get arguments
6976 Node* reference_obj = null_check_receiver();
6977 if (stopped()) return true;
6978
6979 // Common access parameters
6980 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6981 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6982 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
6983 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
6984 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
6985
6986 Node* referent = access_load_at(reference_obj,
6987 referent_field_addr,
6988 referent_field_addr_type,
6989 val_type,
6990 T_OBJECT,
6991 decorators);
6992
6993 IdealKit ideal(this);
6994 #define __ ideal.
6995 __ if_then(referent, BoolTest::ne, null());
6996 sync_kit(ideal);
6997 access_store_at(reference_obj,
6998 referent_field_addr,
6999 referent_field_addr_type,
7000 null(),
7001 val_type,
7002 T_OBJECT,
7003 decorators);
7004 __ sync_kit(this);
7005 __ end_if();
7006 final_sync(ideal);
7007 #undef __
7008
7009 return true;
7010 }
7011
7012 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7013 DecoratorSet decorators, bool is_static,
7014 ciInstanceKlass* fromKls) {
7015 if (fromKls == nullptr) {
7016 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7017 assert(tinst != nullptr, "obj is null");
7018 assert(tinst->is_loaded(), "obj is not loaded");
7019 fromKls = tinst->instance_klass();
7020 } else {
7021 assert(is_static, "only for static field access");
7022 }
7023 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7024 ciSymbol::make(fieldTypeString),
7025 is_static);
7026
7027 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7028 if (field == nullptr) return (Node *) nullptr;
7029
7030 if (is_static) {
7031 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7032 fromObj = makecon(tip);
7033 }
7034
7035 // Next code copied from Parse::do_get_xxx():
7036
7037 // Compute address and memory type.
7038 int offset = field->offset_in_bytes();
7039 bool is_vol = field->is_volatile();
7040 ciType* field_klass = field->type();
7041 assert(field_klass->is_loaded(), "should be loaded");
7042 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7043 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7044 BasicType bt = field->layout_type();
7045
7046 // Build the resultant type of the load
7047 const Type *type;
7048 if (bt == T_OBJECT) {
7049 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7050 } else {
7051 type = Type::get_const_basic_type(bt);
7052 }
7053
7054 if (is_vol) {
7055 decorators |= MO_SEQ_CST;
7056 }
7057
7058 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7059 }
7060
7061 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7062 bool is_exact /* true */, bool is_static /* false */,
7063 ciInstanceKlass * fromKls /* nullptr */) {
7064 if (fromKls == nullptr) {
7065 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7066 assert(tinst != nullptr, "obj is null");
7067 assert(tinst->is_loaded(), "obj is not loaded");
7068 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7069 fromKls = tinst->instance_klass();
7070 }
7071 else {
7072 assert(is_static, "only for static field access");
7073 }
7074 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7075 ciSymbol::make(fieldTypeString),
7076 is_static);
7077
7078 assert(field != nullptr, "undefined field");
7079 assert(!field->is_volatile(), "not defined for volatile fields");
7080
7081 if (is_static) {
7082 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7083 fromObj = makecon(tip);
7084 }
7085
7086 // Next code copied from Parse::do_get_xxx():
7087
7088 // Compute address and memory type.
7089 int offset = field->offset_in_bytes();
7090 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7091
7092 return adr;
7093 }
7094
7095 //------------------------------inline_aescrypt_Block-----------------------
7096 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7097 address stubAddr = nullptr;
7098 const char *stubName;
7099 assert(UseAES, "need AES instruction support");
7100
7101 switch(id) {
7102 case vmIntrinsics::_aescrypt_encryptBlock:
7103 stubAddr = StubRoutines::aescrypt_encryptBlock();
7104 stubName = "aescrypt_encryptBlock";
7105 break;
7106 case vmIntrinsics::_aescrypt_decryptBlock:
7107 stubAddr = StubRoutines::aescrypt_decryptBlock();
7108 stubName = "aescrypt_decryptBlock";
7109 break;
7110 default:
7111 break;
7112 }
7113 if (stubAddr == nullptr) return false;
7114
7115 Node* aescrypt_object = argument(0);
7116 Node* src = argument(1);
7117 Node* src_offset = argument(2);
7118 Node* dest = argument(3);
7119 Node* dest_offset = argument(4);
7120
7121 src = must_be_not_null(src, true);
7122 dest = must_be_not_null(dest, true);
7123
7124 // (1) src and dest are arrays.
7125 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7126 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7127 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7128 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7129
7130 // for the quick and dirty code we will skip all the checks.
7131 // we are just trying to get the call to be generated.
7132 Node* src_start = src;
7133 Node* dest_start = dest;
7134 if (src_offset != nullptr || dest_offset != nullptr) {
7135 assert(src_offset != nullptr && dest_offset != nullptr, "");
7136 src_start = array_element_address(src, src_offset, T_BYTE);
7137 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7138 }
7139
7140 // now need to get the start of its expanded key array
7141 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7142 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7143 if (k_start == nullptr) return false;
7144
7145 // Call the stub.
7146 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7147 stubAddr, stubName, TypePtr::BOTTOM,
7148 src_start, dest_start, k_start);
7149
7150 return true;
7151 }
7152
7153 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7154 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7155 address stubAddr = nullptr;
7156 const char *stubName = nullptr;
7157
7158 assert(UseAES, "need AES instruction support");
7159
7160 switch(id) {
7161 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7162 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7163 stubName = "cipherBlockChaining_encryptAESCrypt";
7164 break;
7165 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7166 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7167 stubName = "cipherBlockChaining_decryptAESCrypt";
7168 break;
7169 default:
7170 break;
7171 }
7172 if (stubAddr == nullptr) return false;
7173
7174 Node* cipherBlockChaining_object = argument(0);
7175 Node* src = argument(1);
7176 Node* src_offset = argument(2);
7177 Node* len = argument(3);
7178 Node* dest = argument(4);
7179 Node* dest_offset = argument(5);
7180
7181 src = must_be_not_null(src, false);
7182 dest = must_be_not_null(dest, false);
7183
7184 // (1) src and dest are arrays.
7185 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7186 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7187 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7188 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7189
7190 // checks are the responsibility of the caller
7191 Node* src_start = src;
7192 Node* dest_start = dest;
7193 if (src_offset != nullptr || dest_offset != nullptr) {
7194 assert(src_offset != nullptr && dest_offset != nullptr, "");
7195 src_start = array_element_address(src, src_offset, T_BYTE);
7196 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7197 }
7198
7199 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7200 // (because of the predicated logic executed earlier).
7201 // so we cast it here safely.
7202 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7203
7204 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7205 if (embeddedCipherObj == nullptr) return false;
7206
7207 // cast it to what we know it will be at runtime
7208 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7209 assert(tinst != nullptr, "CBC obj is null");
7210 assert(tinst->is_loaded(), "CBC obj is not loaded");
7211 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7212 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7213
7214 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7215 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7216 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7217 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7218 aescrypt_object = _gvn.transform(aescrypt_object);
7219
7220 // we need to get the start of the aescrypt_object's expanded key array
7221 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7222 if (k_start == nullptr) return false;
7223
7224 // similarly, get the start address of the r vector
7225 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7226 if (objRvec == nullptr) return false;
7227 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7228
7229 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7230 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7231 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7232 stubAddr, stubName, TypePtr::BOTTOM,
7233 src_start, dest_start, k_start, r_start, len);
7234
7235 // return cipher length (int)
7236 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7237 set_result(retvalue);
7238 return true;
7239 }
7240
7241 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7242 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7243 address stubAddr = nullptr;
7244 const char *stubName = nullptr;
7245
7246 assert(UseAES, "need AES instruction support");
7247
7248 switch (id) {
7249 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7250 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7251 stubName = "electronicCodeBook_encryptAESCrypt";
7252 break;
7253 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7254 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7255 stubName = "electronicCodeBook_decryptAESCrypt";
7256 break;
7257 default:
7258 break;
7259 }
7260
7261 if (stubAddr == nullptr) return false;
7262
7263 Node* electronicCodeBook_object = argument(0);
7264 Node* src = argument(1);
7265 Node* src_offset = argument(2);
7266 Node* len = argument(3);
7267 Node* dest = argument(4);
7268 Node* dest_offset = argument(5);
7269
7270 // (1) src and dest are arrays.
7271 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7272 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7273 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7274 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7275
7276 // checks are the responsibility of the caller
7277 Node* src_start = src;
7278 Node* dest_start = dest;
7279 if (src_offset != nullptr || dest_offset != nullptr) {
7280 assert(src_offset != nullptr && dest_offset != nullptr, "");
7281 src_start = array_element_address(src, src_offset, T_BYTE);
7282 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7283 }
7284
7285 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7286 // (because of the predicated logic executed earlier).
7287 // so we cast it here safely.
7288 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7289
7290 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7291 if (embeddedCipherObj == nullptr) return false;
7292
7293 // cast it to what we know it will be at runtime
7294 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7295 assert(tinst != nullptr, "ECB obj is null");
7296 assert(tinst->is_loaded(), "ECB obj is not loaded");
7297 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7298 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7299
7300 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7301 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7302 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7303 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7304 aescrypt_object = _gvn.transform(aescrypt_object);
7305
7306 // we need to get the start of the aescrypt_object's expanded key array
7307 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7308 if (k_start == nullptr) return false;
7309
7310 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7311 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7312 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7313 stubAddr, stubName, TypePtr::BOTTOM,
7314 src_start, dest_start, k_start, len);
7315
7316 // return cipher length (int)
7317 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7318 set_result(retvalue);
7319 return true;
7320 }
7321
7322 //------------------------------inline_counterMode_AESCrypt-----------------------
7323 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7324 assert(UseAES, "need AES instruction support");
7325 if (!UseAESCTRIntrinsics) return false;
7326
7327 address stubAddr = nullptr;
7328 const char *stubName = nullptr;
7329 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7330 stubAddr = StubRoutines::counterMode_AESCrypt();
7331 stubName = "counterMode_AESCrypt";
7332 }
7333 if (stubAddr == nullptr) return false;
7334
7335 Node* counterMode_object = argument(0);
7336 Node* src = argument(1);
7337 Node* src_offset = argument(2);
7338 Node* len = argument(3);
7339 Node* dest = argument(4);
7340 Node* dest_offset = argument(5);
7341
7342 // (1) src and dest are arrays.
7343 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7344 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7345 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7346 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7347
7348 // checks are the responsibility of the caller
7349 Node* src_start = src;
7350 Node* dest_start = dest;
7351 if (src_offset != nullptr || dest_offset != nullptr) {
7352 assert(src_offset != nullptr && dest_offset != nullptr, "");
7353 src_start = array_element_address(src, src_offset, T_BYTE);
7354 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7355 }
7356
7357 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7358 // (because of the predicated logic executed earlier).
7359 // so we cast it here safely.
7360 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7361 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7362 if (embeddedCipherObj == nullptr) return false;
7363 // cast it to what we know it will be at runtime
7364 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7365 assert(tinst != nullptr, "CTR obj is null");
7366 assert(tinst->is_loaded(), "CTR obj is not loaded");
7367 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7368 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7369 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7370 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7371 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7372 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7373 aescrypt_object = _gvn.transform(aescrypt_object);
7374 // we need to get the start of the aescrypt_object's expanded key array
7375 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7376 if (k_start == nullptr) return false;
7377 // similarly, get the start address of the r vector
7378 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7379 if (obj_counter == nullptr) return false;
7380 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7381
7382 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7383 if (saved_encCounter == nullptr) return false;
7384 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7385 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7386
7387 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7388 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7389 OptoRuntime::counterMode_aescrypt_Type(),
7390 stubAddr, stubName, TypePtr::BOTTOM,
7391 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7392
7393 // return cipher length (int)
7394 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7395 set_result(retvalue);
7396 return true;
7397 }
7398
7399 //------------------------------get_key_start_from_aescrypt_object-----------------------
7400 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7401 #if defined(PPC64) || defined(S390)
7402 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7403 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7404 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7405 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7406 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7407 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7408 if (objSessionK == nullptr) {
7409 return (Node *) nullptr;
7410 }
7411 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7412 #else
7413 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7414 #endif // PPC64
7415 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7416 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7417
7418 // now have the array, need to get the start address of the K array
7419 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7420 return k_start;
7421 }
7422
7423 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7424 // Return node representing slow path of predicate check.
7425 // the pseudo code we want to emulate with this predicate is:
7426 // for encryption:
7427 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7428 // for decryption:
7429 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7430 // note cipher==plain is more conservative than the original java code but that's OK
7431 //
7432 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7433 // The receiver was checked for null already.
7434 Node* objCBC = argument(0);
7435
7436 Node* src = argument(1);
7437 Node* dest = argument(4);
7438
7439 // Load embeddedCipher field of CipherBlockChaining object.
7440 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7441
7442 // get AESCrypt klass for instanceOf check
7443 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7444 // will have same classloader as CipherBlockChaining object
7445 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7446 assert(tinst != nullptr, "CBCobj is null");
7447 assert(tinst->is_loaded(), "CBCobj is not loaded");
7448
7449 // we want to do an instanceof comparison against the AESCrypt class
7450 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7451 if (!klass_AESCrypt->is_loaded()) {
7452 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7453 Node* ctrl = control();
7454 set_control(top()); // no regular fast path
7455 return ctrl;
7456 }
7457
7458 src = must_be_not_null(src, true);
7459 dest = must_be_not_null(dest, true);
7460
7461 // Resolve oops to stable for CmpP below.
7462 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7463
7464 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7465 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7466 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7467
7468 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7469
7470 // for encryption, we are done
7471 if (!decrypting)
7472 return instof_false; // even if it is null
7473
7474 // for decryption, we need to add a further check to avoid
7475 // taking the intrinsic path when cipher and plain are the same
7476 // see the original java code for why.
7477 RegionNode* region = new RegionNode(3);
7478 region->init_req(1, instof_false);
7479
7480 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7481 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7482 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7483 region->init_req(2, src_dest_conjoint);
7484
7485 record_for_igvn(region);
7486 return _gvn.transform(region);
7487 }
7488
7489 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7490 // Return node representing slow path of predicate check.
7491 // the pseudo code we want to emulate with this predicate is:
7492 // for encryption:
7493 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7494 // for decryption:
7495 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7496 // note cipher==plain is more conservative than the original java code but that's OK
7497 //
7498 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7499 // The receiver was checked for null already.
7500 Node* objECB = argument(0);
7501
7502 // Load embeddedCipher field of ElectronicCodeBook object.
7503 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7504
7505 // get AESCrypt klass for instanceOf check
7506 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7507 // will have same classloader as ElectronicCodeBook object
7508 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7509 assert(tinst != nullptr, "ECBobj is null");
7510 assert(tinst->is_loaded(), "ECBobj is not loaded");
7511
7512 // we want to do an instanceof comparison against the AESCrypt class
7513 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7514 if (!klass_AESCrypt->is_loaded()) {
7515 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7516 Node* ctrl = control();
7517 set_control(top()); // no regular fast path
7518 return ctrl;
7519 }
7520 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7521
7522 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7523 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7524 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7525
7526 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7527
7528 // for encryption, we are done
7529 if (!decrypting)
7530 return instof_false; // even if it is null
7531
7532 // for decryption, we need to add a further check to avoid
7533 // taking the intrinsic path when cipher and plain are the same
7534 // see the original java code for why.
7535 RegionNode* region = new RegionNode(3);
7536 region->init_req(1, instof_false);
7537 Node* src = argument(1);
7538 Node* dest = argument(4);
7539 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7540 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7541 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7542 region->init_req(2, src_dest_conjoint);
7543
7544 record_for_igvn(region);
7545 return _gvn.transform(region);
7546 }
7547
7548 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7549 // Return node representing slow path of predicate check.
7550 // the pseudo code we want to emulate with this predicate is:
7551 // for encryption:
7552 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7553 // for decryption:
7554 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7555 // note cipher==plain is more conservative than the original java code but that's OK
7556 //
7557
7558 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7559 // The receiver was checked for null already.
7560 Node* objCTR = argument(0);
7561
7562 // Load embeddedCipher field of CipherBlockChaining object.
7563 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7564
7565 // get AESCrypt klass for instanceOf check
7566 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7567 // will have same classloader as CipherBlockChaining object
7568 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7569 assert(tinst != nullptr, "CTRobj is null");
7570 assert(tinst->is_loaded(), "CTRobj is not loaded");
7571
7572 // we want to do an instanceof comparison against the AESCrypt class
7573 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7574 if (!klass_AESCrypt->is_loaded()) {
7575 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7576 Node* ctrl = control();
7577 set_control(top()); // no regular fast path
7578 return ctrl;
7579 }
7580
7581 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7582 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7583 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7584 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7585 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7586
7587 return instof_false; // even if it is null
7588 }
7589
7590 //------------------------------inline_ghash_processBlocks
7591 bool LibraryCallKit::inline_ghash_processBlocks() {
7592 address stubAddr;
7593 const char *stubName;
7594 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7595
7596 stubAddr = StubRoutines::ghash_processBlocks();
7597 stubName = "ghash_processBlocks";
7598
7599 Node* data = argument(0);
7600 Node* offset = argument(1);
7601 Node* len = argument(2);
7602 Node* state = argument(3);
7603 Node* subkeyH = argument(4);
7604
7605 state = must_be_not_null(state, true);
7606 subkeyH = must_be_not_null(subkeyH, true);
7607 data = must_be_not_null(data, true);
7608
7609 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7610 assert(state_start, "state is null");
7611 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7612 assert(subkeyH_start, "subkeyH is null");
7613 Node* data_start = array_element_address(data, offset, T_BYTE);
7614 assert(data_start, "data is null");
7615
7616 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7617 OptoRuntime::ghash_processBlocks_Type(),
7618 stubAddr, stubName, TypePtr::BOTTOM,
7619 state_start, subkeyH_start, data_start, len);
7620 return true;
7621 }
7622
7623 //------------------------------inline_chacha20Block
7624 bool LibraryCallKit::inline_chacha20Block() {
7625 address stubAddr;
7626 const char *stubName;
7627 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7628
7629 stubAddr = StubRoutines::chacha20Block();
7630 stubName = "chacha20Block";
7631
7632 Node* state = argument(0);
7633 Node* result = argument(1);
7634
7635 state = must_be_not_null(state, true);
7636 result = must_be_not_null(result, true);
7637
7638 Node* state_start = array_element_address(state, intcon(0), T_INT);
7639 assert(state_start, "state is null");
7640 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7641 assert(result_start, "result is null");
7642
7643 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7644 OptoRuntime::chacha20Block_Type(),
7645 stubAddr, stubName, TypePtr::BOTTOM,
7646 state_start, result_start);
7647 // return key stream length (int)
7648 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7649 set_result(retvalue);
7650 return true;
7651 }
7652
7653 bool LibraryCallKit::inline_base64_encodeBlock() {
7654 address stubAddr;
7655 const char *stubName;
7656 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7657 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7658 stubAddr = StubRoutines::base64_encodeBlock();
7659 stubName = "encodeBlock";
7660
7661 if (!stubAddr) return false;
7662 Node* base64obj = argument(0);
7663 Node* src = argument(1);
7664 Node* offset = argument(2);
7665 Node* len = argument(3);
7666 Node* dest = argument(4);
7667 Node* dp = argument(5);
7668 Node* isURL = argument(6);
7669
7670 src = must_be_not_null(src, true);
7671 dest = must_be_not_null(dest, true);
7672
7673 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7674 assert(src_start, "source array is null");
7675 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7676 assert(dest_start, "destination array is null");
7677
7678 Node* base64 = make_runtime_call(RC_LEAF,
7679 OptoRuntime::base64_encodeBlock_Type(),
7680 stubAddr, stubName, TypePtr::BOTTOM,
7681 src_start, offset, len, dest_start, dp, isURL);
7682 return true;
7683 }
7684
7685 bool LibraryCallKit::inline_base64_decodeBlock() {
7686 address stubAddr;
7687 const char *stubName;
7688 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7689 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7690 stubAddr = StubRoutines::base64_decodeBlock();
7691 stubName = "decodeBlock";
7692
7693 if (!stubAddr) return false;
7694 Node* base64obj = argument(0);
7695 Node* src = argument(1);
7696 Node* src_offset = argument(2);
7697 Node* len = argument(3);
7698 Node* dest = argument(4);
7699 Node* dest_offset = argument(5);
7700 Node* isURL = argument(6);
7701 Node* isMIME = argument(7);
7702
7703 src = must_be_not_null(src, true);
7704 dest = must_be_not_null(dest, true);
7705
7706 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7707 assert(src_start, "source array is null");
7708 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7709 assert(dest_start, "destination array is null");
7710
7711 Node* call = make_runtime_call(RC_LEAF,
7712 OptoRuntime::base64_decodeBlock_Type(),
7713 stubAddr, stubName, TypePtr::BOTTOM,
7714 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7715 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7716 set_result(result);
7717 return true;
7718 }
7719
7720 bool LibraryCallKit::inline_poly1305_processBlocks() {
7721 address stubAddr;
7722 const char *stubName;
7723 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7724 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7725 stubAddr = StubRoutines::poly1305_processBlocks();
7726 stubName = "poly1305_processBlocks";
7727
7728 if (!stubAddr) return false;
7729 null_check_receiver(); // null-check receiver
7730 if (stopped()) return true;
7731
7732 Node* input = argument(1);
7733 Node* input_offset = argument(2);
7734 Node* len = argument(3);
7735 Node* alimbs = argument(4);
7736 Node* rlimbs = argument(5);
7737
7738 input = must_be_not_null(input, true);
7739 alimbs = must_be_not_null(alimbs, true);
7740 rlimbs = must_be_not_null(rlimbs, true);
7741
7742 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7743 assert(input_start, "input array is null");
7744 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7745 assert(acc_start, "acc array is null");
7746 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7747 assert(r_start, "r array is null");
7748
7749 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7750 OptoRuntime::poly1305_processBlocks_Type(),
7751 stubAddr, stubName, TypePtr::BOTTOM,
7752 input_start, len, acc_start, r_start);
7753 return true;
7754 }
7755
7756 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7757 address stubAddr;
7758 const char *stubName;
7759 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7760 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7761 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7762 stubName = "intpoly_montgomeryMult_P256";
7763
7764 if (!stubAddr) return false;
7765 null_check_receiver(); // null-check receiver
7766 if (stopped()) return true;
7767
7768 Node* a = argument(1);
7769 Node* b = argument(2);
7770 Node* r = argument(3);
7771
7772 a = must_be_not_null(a, true);
7773 b = must_be_not_null(b, true);
7774 r = must_be_not_null(r, true);
7775
7776 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7777 assert(a_start, "a array is NULL");
7778 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7779 assert(b_start, "b array is NULL");
7780 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7781 assert(r_start, "r array is NULL");
7782
7783 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7784 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7785 stubAddr, stubName, TypePtr::BOTTOM,
7786 a_start, b_start, r_start);
7787 return true;
7788 }
7789
7790 bool LibraryCallKit::inline_intpoly_assign() {
7791 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7792 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7793 const char *stubName = "intpoly_assign";
7794 address stubAddr = StubRoutines::intpoly_assign();
7795 if (!stubAddr) return false;
7796
7797 Node* set = argument(0);
7798 Node* a = argument(1);
7799 Node* b = argument(2);
7800 Node* arr_length = load_array_length(a);
7801
7802 a = must_be_not_null(a, true);
7803 b = must_be_not_null(b, true);
7804
7805 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7806 assert(a_start, "a array is NULL");
7807 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7808 assert(b_start, "b array is NULL");
7809
7810 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7811 OptoRuntime::intpoly_assign_Type(),
7812 stubAddr, stubName, TypePtr::BOTTOM,
7813 set, a_start, b_start, arr_length);
7814 return true;
7815 }
7816
7817 //------------------------------inline_digestBase_implCompress-----------------------
7818 //
7819 // Calculate MD5 for single-block byte[] array.
7820 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7821 //
7822 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7823 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7824 //
7825 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7826 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7827 //
7828 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7829 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7830 //
7831 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7832 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7833 //
7834 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7835 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7836
7837 Node* digestBase_obj = argument(0);
7838 Node* src = argument(1); // type oop
7839 Node* ofs = argument(2); // type int
7840
7841 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7842 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7843 // failed array check
7844 return false;
7845 }
7846 // Figure out the size and type of the elements we will be copying.
7847 BasicType src_elem = src_type->elem()->array_element_basic_type();
7848 if (src_elem != T_BYTE) {
7849 return false;
7850 }
7851 // 'src_start' points to src array + offset
7852 src = must_be_not_null(src, true);
7853 Node* src_start = array_element_address(src, ofs, src_elem);
7854 Node* state = nullptr;
7855 Node* block_size = nullptr;
7856 address stubAddr;
7857 const char *stubName;
7858
7859 switch(id) {
7860 case vmIntrinsics::_md5_implCompress:
7861 assert(UseMD5Intrinsics, "need MD5 instruction support");
7862 state = get_state_from_digest_object(digestBase_obj, T_INT);
7863 stubAddr = StubRoutines::md5_implCompress();
7864 stubName = "md5_implCompress";
7865 break;
7866 case vmIntrinsics::_sha_implCompress:
7867 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7868 state = get_state_from_digest_object(digestBase_obj, T_INT);
7869 stubAddr = StubRoutines::sha1_implCompress();
7870 stubName = "sha1_implCompress";
7871 break;
7872 case vmIntrinsics::_sha2_implCompress:
7873 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7874 state = get_state_from_digest_object(digestBase_obj, T_INT);
7875 stubAddr = StubRoutines::sha256_implCompress();
7876 stubName = "sha256_implCompress";
7877 break;
7878 case vmIntrinsics::_sha5_implCompress:
7879 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7880 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7881 stubAddr = StubRoutines::sha512_implCompress();
7882 stubName = "sha512_implCompress";
7883 break;
7884 case vmIntrinsics::_sha3_implCompress:
7885 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7886 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7887 stubAddr = StubRoutines::sha3_implCompress();
7888 stubName = "sha3_implCompress";
7889 block_size = get_block_size_from_digest_object(digestBase_obj);
7890 if (block_size == nullptr) return false;
7891 break;
7892 default:
7893 fatal_unexpected_iid(id);
7894 return false;
7895 }
7896 if (state == nullptr) return false;
7897
7898 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7899 if (stubAddr == nullptr) return false;
7900
7901 // Call the stub.
7902 Node* call;
7903 if (block_size == nullptr) {
7904 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7905 stubAddr, stubName, TypePtr::BOTTOM,
7906 src_start, state);
7907 } else {
7908 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7909 stubAddr, stubName, TypePtr::BOTTOM,
7910 src_start, state, block_size);
7911 }
7912
7913 return true;
7914 }
7915
7916 //------------------------------inline_digestBase_implCompressMB-----------------------
7917 //
7918 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7919 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7920 //
7921 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7922 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7923 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7924 assert((uint)predicate < 5, "sanity");
7925 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7926
7927 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7928 Node* src = argument(1); // byte[] array
7929 Node* ofs = argument(2); // type int
7930 Node* limit = argument(3); // type int
7931
7932 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7933 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7934 // failed array check
7935 return false;
7936 }
7937 // Figure out the size and type of the elements we will be copying.
7938 BasicType src_elem = src_type->elem()->array_element_basic_type();
7939 if (src_elem != T_BYTE) {
7940 return false;
7941 }
7942 // 'src_start' points to src array + offset
7943 src = must_be_not_null(src, false);
7944 Node* src_start = array_element_address(src, ofs, src_elem);
7945
7946 const char* klass_digestBase_name = nullptr;
7947 const char* stub_name = nullptr;
7948 address stub_addr = nullptr;
7949 BasicType elem_type = T_INT;
7950
7951 switch (predicate) {
7952 case 0:
7953 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7954 klass_digestBase_name = "sun/security/provider/MD5";
7955 stub_name = "md5_implCompressMB";
7956 stub_addr = StubRoutines::md5_implCompressMB();
7957 }
7958 break;
7959 case 1:
7960 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7961 klass_digestBase_name = "sun/security/provider/SHA";
7962 stub_name = "sha1_implCompressMB";
7963 stub_addr = StubRoutines::sha1_implCompressMB();
7964 }
7965 break;
7966 case 2:
7967 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7968 klass_digestBase_name = "sun/security/provider/SHA2";
7969 stub_name = "sha256_implCompressMB";
7970 stub_addr = StubRoutines::sha256_implCompressMB();
7971 }
7972 break;
7973 case 3:
7974 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7975 klass_digestBase_name = "sun/security/provider/SHA5";
7976 stub_name = "sha512_implCompressMB";
7977 stub_addr = StubRoutines::sha512_implCompressMB();
7978 elem_type = T_LONG;
7979 }
7980 break;
7981 case 4:
7982 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7983 klass_digestBase_name = "sun/security/provider/SHA3";
7984 stub_name = "sha3_implCompressMB";
7985 stub_addr = StubRoutines::sha3_implCompressMB();
7986 elem_type = T_LONG;
7987 }
7988 break;
7989 default:
7990 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7991 }
7992 if (klass_digestBase_name != nullptr) {
7993 assert(stub_addr != nullptr, "Stub is generated");
7994 if (stub_addr == nullptr) return false;
7995
7996 // get DigestBase klass to lookup for SHA klass
7997 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7998 assert(tinst != nullptr, "digestBase_obj is not instance???");
7999 assert(tinst->is_loaded(), "DigestBase is not loaded");
8000
8001 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8002 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8003 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8004 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8005 }
8006 return false;
8007 }
8008
8009 //------------------------------inline_digestBase_implCompressMB-----------------------
8010 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8011 BasicType elem_type, address stubAddr, const char *stubName,
8012 Node* src_start, Node* ofs, Node* limit) {
8013 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8014 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8015 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8016 digest_obj = _gvn.transform(digest_obj);
8017
8018 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8019 if (state == nullptr) return false;
8020
8021 Node* block_size = nullptr;
8022 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8023 block_size = get_block_size_from_digest_object(digest_obj);
8024 if (block_size == nullptr) return false;
8025 }
8026
8027 // Call the stub.
8028 Node* call;
8029 if (block_size == nullptr) {
8030 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8031 OptoRuntime::digestBase_implCompressMB_Type(false),
8032 stubAddr, stubName, TypePtr::BOTTOM,
8033 src_start, state, ofs, limit);
8034 } else {
8035 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8036 OptoRuntime::digestBase_implCompressMB_Type(true),
8037 stubAddr, stubName, TypePtr::BOTTOM,
8038 src_start, state, block_size, ofs, limit);
8039 }
8040
8041 // return ofs (int)
8042 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8043 set_result(result);
8044
8045 return true;
8046 }
8047
8048 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8049 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8050 assert(UseAES, "need AES instruction support");
8051 address stubAddr = nullptr;
8052 const char *stubName = nullptr;
8053 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8054 stubName = "galoisCounterMode_AESCrypt";
8055
8056 if (stubAddr == nullptr) return false;
8057
8058 Node* in = argument(0);
8059 Node* inOfs = argument(1);
8060 Node* len = argument(2);
8061 Node* ct = argument(3);
8062 Node* ctOfs = argument(4);
8063 Node* out = argument(5);
8064 Node* outOfs = argument(6);
8065 Node* gctr_object = argument(7);
8066 Node* ghash_object = argument(8);
8067
8068 // (1) in, ct and out are arrays.
8069 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8070 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8071 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8072 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8073 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8074 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8075
8076 // checks are the responsibility of the caller
8077 Node* in_start = in;
8078 Node* ct_start = ct;
8079 Node* out_start = out;
8080 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8081 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8082 in_start = array_element_address(in, inOfs, T_BYTE);
8083 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8084 out_start = array_element_address(out, outOfs, T_BYTE);
8085 }
8086
8087 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8088 // (because of the predicated logic executed earlier).
8089 // so we cast it here safely.
8090 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8091 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8092 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8093 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8094 Node* state = load_field_from_object(ghash_object, "state", "[J");
8095
8096 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8097 return false;
8098 }
8099 // cast it to what we know it will be at runtime
8100 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8101 assert(tinst != nullptr, "GCTR obj is null");
8102 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8103 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8104 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8105 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8106 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8107 const TypeOopPtr* xtype = aklass->as_instance_type();
8108 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8109 aescrypt_object = _gvn.transform(aescrypt_object);
8110 // we need to get the start of the aescrypt_object's expanded key array
8111 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8112 if (k_start == nullptr) return false;
8113 // similarly, get the start address of the r vector
8114 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8115 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8116 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8117
8118
8119 // Call the stub, passing params
8120 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8121 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8122 stubAddr, stubName, TypePtr::BOTTOM,
8123 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8124
8125 // return cipher length (int)
8126 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8127 set_result(retvalue);
8128
8129 return true;
8130 }
8131
8132 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8133 // Return node representing slow path of predicate check.
8134 // the pseudo code we want to emulate with this predicate is:
8135 // for encryption:
8136 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8137 // for decryption:
8138 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8139 // note cipher==plain is more conservative than the original java code but that's OK
8140 //
8141
8142 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8143 // The receiver was checked for null already.
8144 Node* objGCTR = argument(7);
8145 // Load embeddedCipher field of GCTR object.
8146 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8147 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8148
8149 // get AESCrypt klass for instanceOf check
8150 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8151 // will have same classloader as CipherBlockChaining object
8152 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8153 assert(tinst != nullptr, "GCTR obj is null");
8154 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8155
8156 // we want to do an instanceof comparison against the AESCrypt class
8157 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8158 if (!klass_AESCrypt->is_loaded()) {
8159 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8160 Node* ctrl = control();
8161 set_control(top()); // no regular fast path
8162 return ctrl;
8163 }
8164
8165 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8166 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8167 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8168 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8169 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8170
8171 return instof_false; // even if it is null
8172 }
8173
8174 //------------------------------get_state_from_digest_object-----------------------
8175 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8176 const char* state_type;
8177 switch (elem_type) {
8178 case T_BYTE: state_type = "[B"; break;
8179 case T_INT: state_type = "[I"; break;
8180 case T_LONG: state_type = "[J"; break;
8181 default: ShouldNotReachHere();
8182 }
8183 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8184 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8185 if (digest_state == nullptr) return (Node *) nullptr;
8186
8187 // now have the array, need to get the start address of the state array
8188 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8189 return state;
8190 }
8191
8192 //------------------------------get_block_size_from_sha3_object----------------------------------
8193 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8194 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8195 assert (block_size != nullptr, "sanity");
8196 return block_size;
8197 }
8198
8199 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8200 // Return node representing slow path of predicate check.
8201 // the pseudo code we want to emulate with this predicate is:
8202 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8203 //
8204 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8205 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8206 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8207 assert((uint)predicate < 5, "sanity");
8208
8209 // The receiver was checked for null already.
8210 Node* digestBaseObj = argument(0);
8211
8212 // get DigestBase klass for instanceOf check
8213 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8214 assert(tinst != nullptr, "digestBaseObj is null");
8215 assert(tinst->is_loaded(), "DigestBase is not loaded");
8216
8217 const char* klass_name = nullptr;
8218 switch (predicate) {
8219 case 0:
8220 if (UseMD5Intrinsics) {
8221 // we want to do an instanceof comparison against the MD5 class
8222 klass_name = "sun/security/provider/MD5";
8223 }
8224 break;
8225 case 1:
8226 if (UseSHA1Intrinsics) {
8227 // we want to do an instanceof comparison against the SHA class
8228 klass_name = "sun/security/provider/SHA";
8229 }
8230 break;
8231 case 2:
8232 if (UseSHA256Intrinsics) {
8233 // we want to do an instanceof comparison against the SHA2 class
8234 klass_name = "sun/security/provider/SHA2";
8235 }
8236 break;
8237 case 3:
8238 if (UseSHA512Intrinsics) {
8239 // we want to do an instanceof comparison against the SHA5 class
8240 klass_name = "sun/security/provider/SHA5";
8241 }
8242 break;
8243 case 4:
8244 if (UseSHA3Intrinsics) {
8245 // we want to do an instanceof comparison against the SHA3 class
8246 klass_name = "sun/security/provider/SHA3";
8247 }
8248 break;
8249 default:
8250 fatal("unknown SHA intrinsic predicate: %d", predicate);
8251 }
8252
8253 ciKlass* klass = nullptr;
8254 if (klass_name != nullptr) {
8255 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8256 }
8257 if ((klass == nullptr) || !klass->is_loaded()) {
8258 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8259 Node* ctrl = control();
8260 set_control(top()); // no intrinsic path
8261 return ctrl;
8262 }
8263 ciInstanceKlass* instklass = klass->as_instance_klass();
8264
8265 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8266 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8267 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8268 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8269
8270 return instof_false; // even if it is null
8271 }
8272
8273 //-------------inline_fma-----------------------------------
8274 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8275 Node *a = nullptr;
8276 Node *b = nullptr;
8277 Node *c = nullptr;
8278 Node* result = nullptr;
8279 switch (id) {
8280 case vmIntrinsics::_fmaD:
8281 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8282 // no receiver since it is static method
8283 a = round_double_node(argument(0));
8284 b = round_double_node(argument(2));
8285 c = round_double_node(argument(4));
8286 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8287 break;
8288 case vmIntrinsics::_fmaF:
8289 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8290 a = argument(0);
8291 b = argument(1);
8292 c = argument(2);
8293 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8294 break;
8295 default:
8296 fatal_unexpected_iid(id); break;
8297 }
8298 set_result(result);
8299 return true;
8300 }
8301
8302 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8303 // argument(0) is receiver
8304 Node* codePoint = argument(1);
8305 Node* n = nullptr;
8306
8307 switch (id) {
8308 case vmIntrinsics::_isDigit :
8309 n = new DigitNode(control(), codePoint);
8310 break;
8311 case vmIntrinsics::_isLowerCase :
8312 n = new LowerCaseNode(control(), codePoint);
8313 break;
8314 case vmIntrinsics::_isUpperCase :
8315 n = new UpperCaseNode(control(), codePoint);
8316 break;
8317 case vmIntrinsics::_isWhitespace :
8318 n = new WhitespaceNode(control(), codePoint);
8319 break;
8320 default:
8321 fatal_unexpected_iid(id);
8322 }
8323
8324 set_result(_gvn.transform(n));
8325 return true;
8326 }
8327
8328 //------------------------------inline_fp_min_max------------------------------
8329 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8330 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8331
8332 // The intrinsic should be used only when the API branches aren't predictable,
8333 // the last one performing the most important comparison. The following heuristic
8334 // uses the branch statistics to eventually bail out if necessary.
8335
8336 ciMethodData *md = callee()->method_data();
8337
8338 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8339 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8340
8341 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8342 // Bail out if the call-site didn't contribute enough to the statistics.
8343 return false;
8344 }
8345
8346 uint taken = 0, not_taken = 0;
8347
8348 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8349 if (p->is_BranchData()) {
8350 taken = ((ciBranchData*)p)->taken();
8351 not_taken = ((ciBranchData*)p)->not_taken();
8352 }
8353 }
8354
8355 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8356 balance = balance < 0 ? -balance : balance;
8357 if ( balance > 0.2 ) {
8358 // Bail out if the most important branch is predictable enough.
8359 return false;
8360 }
8361 }
8362 */
8363
8364 Node *a = nullptr;
8365 Node *b = nullptr;
8366 Node *n = nullptr;
8367 switch (id) {
8368 case vmIntrinsics::_maxF:
8369 case vmIntrinsics::_minF:
8370 case vmIntrinsics::_maxF_strict:
8371 case vmIntrinsics::_minF_strict:
8372 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8373 a = argument(0);
8374 b = argument(1);
8375 break;
8376 case vmIntrinsics::_maxD:
8377 case vmIntrinsics::_minD:
8378 case vmIntrinsics::_maxD_strict:
8379 case vmIntrinsics::_minD_strict:
8380 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8381 a = round_double_node(argument(0));
8382 b = round_double_node(argument(2));
8383 break;
8384 default:
8385 fatal_unexpected_iid(id);
8386 break;
8387 }
8388 switch (id) {
8389 case vmIntrinsics::_maxF:
8390 case vmIntrinsics::_maxF_strict:
8391 n = new MaxFNode(a, b);
8392 break;
8393 case vmIntrinsics::_minF:
8394 case vmIntrinsics::_minF_strict:
8395 n = new MinFNode(a, b);
8396 break;
8397 case vmIntrinsics::_maxD:
8398 case vmIntrinsics::_maxD_strict:
8399 n = new MaxDNode(a, b);
8400 break;
8401 case vmIntrinsics::_minD:
8402 case vmIntrinsics::_minD_strict:
8403 n = new MinDNode(a, b);
8404 break;
8405 default:
8406 fatal_unexpected_iid(id);
8407 break;
8408 }
8409 set_result(_gvn.transform(n));
8410 return true;
8411 }
8412
8413 bool LibraryCallKit::inline_profileBoolean() {
8414 Node* counts = argument(1);
8415 const TypeAryPtr* ary = nullptr;
8416 ciArray* aobj = nullptr;
8417 if (counts->is_Con()
8418 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8419 && (aobj = ary->const_oop()->as_array()) != nullptr
8420 && (aobj->length() == 2)) {
8421 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8422 jint false_cnt = aobj->element_value(0).as_int();
8423 jint true_cnt = aobj->element_value(1).as_int();
8424
8425 if (C->log() != nullptr) {
8426 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8427 false_cnt, true_cnt);
8428 }
8429
8430 if (false_cnt + true_cnt == 0) {
8431 // According to profile, never executed.
8432 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8433 Deoptimization::Action_reinterpret);
8434 return true;
8435 }
8436
8437 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8438 // is a number of each value occurrences.
8439 Node* result = argument(0);
8440 if (false_cnt == 0 || true_cnt == 0) {
8441 // According to profile, one value has been never seen.
8442 int expected_val = (false_cnt == 0) ? 1 : 0;
8443
8444 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8445 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8446
8447 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8448 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8449 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8450
8451 { // Slow path: uncommon trap for never seen value and then reexecute
8452 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8453 // the value has been seen at least once.
8454 PreserveJVMState pjvms(this);
8455 PreserveReexecuteState preexecs(this);
8456 jvms()->set_should_reexecute(true);
8457
8458 set_control(slow_path);
8459 set_i_o(i_o());
8460
8461 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8462 Deoptimization::Action_reinterpret);
8463 }
8464 // The guard for never seen value enables sharpening of the result and
8465 // returning a constant. It allows to eliminate branches on the same value
8466 // later on.
8467 set_control(fast_path);
8468 result = intcon(expected_val);
8469 }
8470 // Stop profiling.
8471 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8472 // By replacing method body with profile data (represented as ProfileBooleanNode
8473 // on IR level) we effectively disable profiling.
8474 // It enables full speed execution once optimized code is generated.
8475 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8476 C->record_for_igvn(profile);
8477 set_result(profile);
8478 return true;
8479 } else {
8480 // Continue profiling.
8481 // Profile data isn't available at the moment. So, execute method's bytecode version.
8482 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8483 // is compiled and counters aren't available since corresponding MethodHandle
8484 // isn't a compile-time constant.
8485 return false;
8486 }
8487 }
8488
8489 bool LibraryCallKit::inline_isCompileConstant() {
8490 Node* n = argument(0);
8491 set_result(n->is_Con() ? intcon(1) : intcon(0));
8492 return true;
8493 }
8494
8495 //------------------------------- inline_getObjectSize --------------------------------------
8496 //
8497 // Calculate the runtime size of the object/array.
8498 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8499 //
8500 bool LibraryCallKit::inline_getObjectSize() {
8501 Node* obj = argument(3);
8502 Node* klass_node = load_object_klass(obj);
8503
8504 jint layout_con = Klass::_lh_neutral_value;
8505 Node* layout_val = get_layout_helper(klass_node, layout_con);
8506 int layout_is_con = (layout_val == nullptr);
8507
8508 if (layout_is_con) {
8509 // Layout helper is constant, can figure out things at compile time.
8510
8511 if (Klass::layout_helper_is_instance(layout_con)) {
8512 // Instance case: layout_con contains the size itself.
8513 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8514 set_result(size);
8515 } else {
8516 // Array case: size is round(header + element_size*arraylength).
8517 // Since arraylength is different for every array instance, we have to
8518 // compute the whole thing at runtime.
8519
8520 Node* arr_length = load_array_length(obj);
8521
8522 int round_mask = MinObjAlignmentInBytes - 1;
8523 int hsize = Klass::layout_helper_header_size(layout_con);
8524 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8525
8526 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8527 round_mask = 0; // strength-reduce it if it goes away completely
8528 }
8529 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8530 Node* header_size = intcon(hsize + round_mask);
8531
8532 Node* lengthx = ConvI2X(arr_length);
8533 Node* headerx = ConvI2X(header_size);
8534
8535 Node* abody = lengthx;
8536 if (eshift != 0) {
8537 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8538 }
8539 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8540 if (round_mask != 0) {
8541 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8542 }
8543 size = ConvX2L(size);
8544 set_result(size);
8545 }
8546 } else {
8547 // Layout helper is not constant, need to test for array-ness at runtime.
8548
8549 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8550 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8551 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8552 record_for_igvn(result_reg);
8553
8554 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8555 if (array_ctl != nullptr) {
8556 // Array case: size is round(header + element_size*arraylength).
8557 // Since arraylength is different for every array instance, we have to
8558 // compute the whole thing at runtime.
8559
8560 PreserveJVMState pjvms(this);
8561 set_control(array_ctl);
8562 Node* arr_length = load_array_length(obj);
8563
8564 int round_mask = MinObjAlignmentInBytes - 1;
8565 Node* mask = intcon(round_mask);
8566
8567 Node* hss = intcon(Klass::_lh_header_size_shift);
8568 Node* hsm = intcon(Klass::_lh_header_size_mask);
8569 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8570 header_size = _gvn.transform(new AndINode(header_size, hsm));
8571 header_size = _gvn.transform(new AddINode(header_size, mask));
8572
8573 // There is no need to mask or shift this value.
8574 // The semantics of LShiftINode include an implicit mask to 0x1F.
8575 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8576 Node* elem_shift = layout_val;
8577
8578 Node* lengthx = ConvI2X(arr_length);
8579 Node* headerx = ConvI2X(header_size);
8580
8581 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8582 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8583 if (round_mask != 0) {
8584 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8585 }
8586 size = ConvX2L(size);
8587
8588 result_reg->init_req(_array_path, control());
8589 result_val->init_req(_array_path, size);
8590 }
8591
8592 if (!stopped()) {
8593 // Instance case: the layout helper gives us instance size almost directly,
8594 // but we need to mask out the _lh_instance_slow_path_bit.
8595 Node* size = ConvI2X(layout_val);
8596 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8597 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8598 size = _gvn.transform(new AndXNode(size, mask));
8599 size = ConvX2L(size);
8600
8601 result_reg->init_req(_instance_path, control());
8602 result_val->init_req(_instance_path, size);
8603 }
8604
8605 set_result(result_reg, result_val);
8606 }
8607
8608 return true;
8609 }
8610
8611 //------------------------------- inline_blackhole --------------------------------------
8612 //
8613 // Make sure all arguments to this node are alive.
8614 // This matches methods that were requested to be blackholed through compile commands.
8615 //
8616 bool LibraryCallKit::inline_blackhole() {
8617 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8618 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8619 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8620
8621 // Blackhole node pinches only the control, not memory. This allows
8622 // the blackhole to be pinned in the loop that computes blackholed
8623 // values, but have no other side effects, like breaking the optimizations
8624 // across the blackhole.
8625
8626 Node* bh = _gvn.transform(new BlackholeNode(control()));
8627 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8628
8629 // Bind call arguments as blackhole arguments to keep them alive
8630 uint nargs = callee()->arg_size();
8631 for (uint i = 0; i < nargs; i++) {
8632 bh->add_req(argument(i));
8633 }
8634
8635 return true;
8636 }