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