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