1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dabs:
258 case vmIntrinsics::_fabs:
259 case vmIntrinsics::_iabs:
260 case vmIntrinsics::_labs:
261 case vmIntrinsics::_datan2:
262 case vmIntrinsics::_dsqrt:
263 case vmIntrinsics::_dsqrt_strict:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_roundF:
272 case vmIntrinsics::_roundD:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_arraySort: return inline_array_sort();
297 case vmIntrinsics::_arrayPartition: return inline_array_partition();
298
299 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
300 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
301 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
302 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
303
304 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
308 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
309 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
310 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
311 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
312
313 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
314
315 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
316
317 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
318 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
319 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
320 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
321
322 case vmIntrinsics::_compressStringC:
323 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
324 case vmIntrinsics::_inflateStringC:
325 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
326
327 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
328 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
329 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
330 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
331 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
332 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
333 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
334 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
335 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
336
337 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
338 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
339 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
340 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
341 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
342 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
343 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
344 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
345 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
346
347 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
348 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
349 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
350 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
351 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
352 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
353 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
354 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
355 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
356
357 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
358 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
359 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
360 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
361 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
362 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
363 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
364 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
365 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
366
367 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
373 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
374 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
375 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
376
377 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
378 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
379 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
380 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
381 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
382 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
383 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
384 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
385 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
386
387 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
388 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
389 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
390 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
391 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
392 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
393 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
394 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
395 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
396
397 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
398 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
399 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
400 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
401 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
402 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
403 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
404 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
405 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
406
407 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
408 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
409 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
410 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
411 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
412 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
413 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
414 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
415 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
416
417 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
422
423 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
443
444 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
459
460 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
464
465 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
470
471 case vmIntrinsics::_loadFence:
472 case vmIntrinsics::_storeFence:
473 case vmIntrinsics::_storeStoreFence:
474 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
475
476 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
477
478 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
479 case vmIntrinsics::_currentThread: return inline_native_currentThread();
480 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
481
482 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
483 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
484
485 #if INCLUDE_JVMTI
486 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
487 "notifyJvmtiStart", true, false);
488 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
489 "notifyJvmtiEnd", false, true);
490 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
491 "notifyJvmtiMount", false, false);
492 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
493 "notifyJvmtiUnmount", false, false);
494 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
495 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
496 #endif
497
498 #ifdef JFR_HAVE_INTRINSICS
499 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
500 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
501 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
502 #endif
503 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
504 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
505 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
506 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
507 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
508 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
509 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
510 case vmIntrinsics::_getLength: return inline_native_getLength();
511 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
512 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
513 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
514 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
515 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
516 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
517 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
518
519 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
520 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
521
522 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
523
524 case vmIntrinsics::_isInstance:
525 case vmIntrinsics::_getModifiers:
526 case vmIntrinsics::_isInterface:
527 case vmIntrinsics::_isArray:
528 case vmIntrinsics::_isPrimitive:
529 case vmIntrinsics::_isHidden:
530 case vmIntrinsics::_getSuperclass:
531 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
532
533 case vmIntrinsics::_floatToRawIntBits:
534 case vmIntrinsics::_floatToIntBits:
535 case vmIntrinsics::_intBitsToFloat:
536 case vmIntrinsics::_doubleToRawLongBits:
537 case vmIntrinsics::_doubleToLongBits:
538 case vmIntrinsics::_longBitsToDouble:
539 case vmIntrinsics::_floatToFloat16:
540 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
541
542 case vmIntrinsics::_floatIsFinite:
543 case vmIntrinsics::_floatIsInfinite:
544 case vmIntrinsics::_doubleIsFinite:
545 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
546
547 case vmIntrinsics::_numberOfLeadingZeros_i:
548 case vmIntrinsics::_numberOfLeadingZeros_l:
549 case vmIntrinsics::_numberOfTrailingZeros_i:
550 case vmIntrinsics::_numberOfTrailingZeros_l:
551 case vmIntrinsics::_bitCount_i:
552 case vmIntrinsics::_bitCount_l:
553 case vmIntrinsics::_reverse_i:
554 case vmIntrinsics::_reverse_l:
555 case vmIntrinsics::_reverseBytes_i:
556 case vmIntrinsics::_reverseBytes_l:
557 case vmIntrinsics::_reverseBytes_s:
558 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
559
560 case vmIntrinsics::_compress_i:
561 case vmIntrinsics::_compress_l:
562 case vmIntrinsics::_expand_i:
563 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
564
565 case vmIntrinsics::_compareUnsigned_i:
566 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
567
568 case vmIntrinsics::_divideUnsigned_i:
569 case vmIntrinsics::_divideUnsigned_l:
570 case vmIntrinsics::_remainderUnsigned_i:
571 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
572
573 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
574
575 case vmIntrinsics::_Reference_get: return inline_reference_get();
576 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
577 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
578
579 case vmIntrinsics::_Class_cast: return inline_Class_cast();
580
581 case vmIntrinsics::_aescrypt_encryptBlock:
582 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
583
584 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
585 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
586 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
589 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
590 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
591
592 case vmIntrinsics::_counterMode_AESCrypt:
593 return inline_counterMode_AESCrypt(intrinsic_id());
594
595 case vmIntrinsics::_galoisCounterMode_AESCrypt:
596 return inline_galoisCounterMode_AESCrypt();
597
598 case vmIntrinsics::_md5_implCompress:
599 case vmIntrinsics::_sha_implCompress:
600 case vmIntrinsics::_sha2_implCompress:
601 case vmIntrinsics::_sha5_implCompress:
602 case vmIntrinsics::_sha3_implCompress:
603 return inline_digestBase_implCompress(intrinsic_id());
604
605 case vmIntrinsics::_digestBase_implCompressMB:
606 return inline_digestBase_implCompressMB(predicate);
607
608 case vmIntrinsics::_multiplyToLen:
609 return inline_multiplyToLen();
610
611 case vmIntrinsics::_squareToLen:
612 return inline_squareToLen();
613
614 case vmIntrinsics::_mulAdd:
615 return inline_mulAdd();
616
617 case vmIntrinsics::_montgomeryMultiply:
618 return inline_montgomeryMultiply();
619 case vmIntrinsics::_montgomerySquare:
620 return inline_montgomerySquare();
621
622 case vmIntrinsics::_bigIntegerRightShiftWorker:
623 return inline_bigIntegerShift(true);
624 case vmIntrinsics::_bigIntegerLeftShiftWorker:
625 return inline_bigIntegerShift(false);
626
627 case vmIntrinsics::_vectorizedMismatch:
628 return inline_vectorizedMismatch();
629
630 case vmIntrinsics::_ghash_processBlocks:
631 return inline_ghash_processBlocks();
632 case vmIntrinsics::_chacha20Block:
633 return inline_chacha20Block();
634 case vmIntrinsics::_base64_encodeBlock:
635 return inline_base64_encodeBlock();
636 case vmIntrinsics::_base64_decodeBlock:
637 return inline_base64_decodeBlock();
638 case vmIntrinsics::_poly1305_processBlocks:
639 return inline_poly1305_processBlocks();
640
641 case vmIntrinsics::_encodeISOArray:
642 case vmIntrinsics::_encodeByteISOArray:
643 return inline_encodeISOArray(false);
644 case vmIntrinsics::_encodeAsciiArray:
645 return inline_encodeISOArray(true);
646
647 case vmIntrinsics::_updateCRC32:
648 return inline_updateCRC32();
649 case vmIntrinsics::_updateBytesCRC32:
650 return inline_updateBytesCRC32();
651 case vmIntrinsics::_updateByteBufferCRC32:
652 return inline_updateByteBufferCRC32();
653
654 case vmIntrinsics::_updateBytesCRC32C:
655 return inline_updateBytesCRC32C();
656 case vmIntrinsics::_updateDirectByteBufferCRC32C:
657 return inline_updateDirectByteBufferCRC32C();
658
659 case vmIntrinsics::_updateBytesAdler32:
660 return inline_updateBytesAdler32();
661 case vmIntrinsics::_updateByteBufferAdler32:
662 return inline_updateByteBufferAdler32();
663
664 case vmIntrinsics::_profileBoolean:
665 return inline_profileBoolean();
666 case vmIntrinsics::_isCompileConstant:
667 return inline_isCompileConstant();
668
669 case vmIntrinsics::_countPositives:
670 return inline_countPositives();
671
672 case vmIntrinsics::_fmaD:
673 case vmIntrinsics::_fmaF:
674 return inline_fma(intrinsic_id());
675
676 case vmIntrinsics::_isDigit:
677 case vmIntrinsics::_isLowerCase:
678 case vmIntrinsics::_isUpperCase:
679 case vmIntrinsics::_isWhitespace:
680 return inline_character_compare(intrinsic_id());
681
682 case vmIntrinsics::_min:
683 case vmIntrinsics::_max:
684 case vmIntrinsics::_min_strict:
685 case vmIntrinsics::_max_strict:
686 return inline_min_max(intrinsic_id());
687
688 case vmIntrinsics::_maxF:
689 case vmIntrinsics::_minF:
690 case vmIntrinsics::_maxD:
691 case vmIntrinsics::_minD:
692 case vmIntrinsics::_maxF_strict:
693 case vmIntrinsics::_minF_strict:
694 case vmIntrinsics::_maxD_strict:
695 case vmIntrinsics::_minD_strict:
696 return inline_fp_min_max(intrinsic_id());
697
698 case vmIntrinsics::_VectorUnaryOp:
699 return inline_vector_nary_operation(1);
700 case vmIntrinsics::_VectorBinaryOp:
701 return inline_vector_nary_operation(2);
702 case vmIntrinsics::_VectorTernaryOp:
703 return inline_vector_nary_operation(3);
704 case vmIntrinsics::_VectorFromBitsCoerced:
705 return inline_vector_frombits_coerced();
706 case vmIntrinsics::_VectorShuffleIota:
707 return inline_vector_shuffle_iota();
708 case vmIntrinsics::_VectorMaskOp:
709 return inline_vector_mask_operation();
710 case vmIntrinsics::_VectorShuffleToVector:
711 return inline_vector_shuffle_to_vector();
712 case vmIntrinsics::_VectorLoadOp:
713 return inline_vector_mem_operation(/*is_store=*/false);
714 case vmIntrinsics::_VectorLoadMaskedOp:
715 return inline_vector_mem_masked_operation(/*is_store*/false);
716 case vmIntrinsics::_VectorStoreOp:
717 return inline_vector_mem_operation(/*is_store=*/true);
718 case vmIntrinsics::_VectorStoreMaskedOp:
719 return inline_vector_mem_masked_operation(/*is_store=*/true);
720 case vmIntrinsics::_VectorGatherOp:
721 return inline_vector_gather_scatter(/*is_scatter*/ false);
722 case vmIntrinsics::_VectorScatterOp:
723 return inline_vector_gather_scatter(/*is_scatter*/ true);
724 case vmIntrinsics::_VectorReductionCoerced:
725 return inline_vector_reduction();
726 case vmIntrinsics::_VectorTest:
727 return inline_vector_test();
728 case vmIntrinsics::_VectorBlend:
729 return inline_vector_blend();
730 case vmIntrinsics::_VectorRearrange:
731 return inline_vector_rearrange();
732 case vmIntrinsics::_VectorCompare:
733 return inline_vector_compare();
734 case vmIntrinsics::_VectorBroadcastInt:
735 return inline_vector_broadcast_int();
736 case vmIntrinsics::_VectorConvert:
737 return inline_vector_convert();
738 case vmIntrinsics::_VectorInsert:
739 return inline_vector_insert();
740 case vmIntrinsics::_VectorExtract:
741 return inline_vector_extract();
742 case vmIntrinsics::_VectorCompressExpand:
743 return inline_vector_compress_expand();
744 case vmIntrinsics::_IndexVector:
745 return inline_index_vector();
746 case vmIntrinsics::_IndexPartiallyInUpperRange:
747 return inline_index_partially_in_upper_range();
748
749 case vmIntrinsics::_getObjectSize:
750 return inline_getObjectSize();
751
752 case vmIntrinsics::_blackhole:
753 return inline_blackhole();
754
755 default:
756 // If you get here, it may be that someone has added a new intrinsic
757 // to the list in vmIntrinsics.hpp without implementing it here.
758 #ifndef PRODUCT
759 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
760 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
761 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
762 }
763 #endif
764 return false;
765 }
766 }
767
768 Node* LibraryCallKit::try_to_predicate(int predicate) {
769 if (!jvms()->has_method()) {
770 // Root JVMState has a null method.
771 assert(map()->memory()->Opcode() == Op_Parm, "");
772 // Insert the memory aliasing node
773 set_all_memory(reset_memory());
774 }
775 assert(merged_memory(), "");
776
777 switch (intrinsic_id()) {
778 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
779 return inline_cipherBlockChaining_AESCrypt_predicate(false);
780 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
781 return inline_cipherBlockChaining_AESCrypt_predicate(true);
782 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
783 return inline_electronicCodeBook_AESCrypt_predicate(false);
784 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
785 return inline_electronicCodeBook_AESCrypt_predicate(true);
786 case vmIntrinsics::_counterMode_AESCrypt:
787 return inline_counterMode_AESCrypt_predicate();
788 case vmIntrinsics::_digestBase_implCompressMB:
789 return inline_digestBase_implCompressMB_predicate(predicate);
790 case vmIntrinsics::_galoisCounterMode_AESCrypt:
791 return inline_galoisCounterMode_AESCrypt_predicate();
792
793 default:
794 // If you get here, it may be that someone has added a new intrinsic
795 // to the list in vmIntrinsics.hpp without implementing it here.
796 #ifndef PRODUCT
797 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
798 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
799 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
800 }
801 #endif
802 Node* slow_ctl = control();
803 set_control(top()); // No fast path intrinsic
804 return slow_ctl;
805 }
806 }
807
808 //------------------------------set_result-------------------------------
809 // Helper function for finishing intrinsics.
810 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
811 record_for_igvn(region);
812 set_control(_gvn.transform(region));
813 set_result( _gvn.transform(value));
814 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
815 }
816
817 //------------------------------generate_guard---------------------------
818 // Helper function for generating guarded fast-slow graph structures.
819 // The given 'test', if true, guards a slow path. If the test fails
820 // then a fast path can be taken. (We generally hope it fails.)
821 // In all cases, GraphKit::control() is updated to the fast path.
822 // The returned value represents the control for the slow path.
823 // The return value is never 'top'; it is either a valid control
824 // or null if it is obvious that the slow path can never be taken.
825 // Also, if region and the slow control are not null, the slow edge
826 // is appended to the region.
827 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
828 if (stopped()) {
829 // Already short circuited.
830 return nullptr;
831 }
832
833 // Build an if node and its projections.
834 // If test is true we take the slow path, which we assume is uncommon.
835 if (_gvn.type(test) == TypeInt::ZERO) {
836 // The slow branch is never taken. No need to build this guard.
837 return nullptr;
838 }
839
840 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
841
842 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
843 if (if_slow == top()) {
844 // The slow branch is never taken. No need to build this guard.
845 return nullptr;
846 }
847
848 if (region != nullptr)
849 region->add_req(if_slow);
850
851 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
852 set_control(if_fast);
853
854 return if_slow;
855 }
856
857 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
858 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
859 }
860 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
861 return generate_guard(test, region, PROB_FAIR);
862 }
863
864 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
865 Node* *pos_index) {
866 if (stopped())
867 return nullptr; // already stopped
868 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
869 return nullptr; // index is already adequately typed
870 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
871 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
872 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
873 if (is_neg != nullptr && pos_index != nullptr) {
874 // Emulate effect of Parse::adjust_map_after_if.
875 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
876 (*pos_index) = _gvn.transform(ccast);
877 }
878 return is_neg;
879 }
880
881 // Make sure that 'position' is a valid limit index, in [0..length].
882 // There are two equivalent plans for checking this:
883 // A. (offset + copyLength) unsigned<= arrayLength
884 // B. offset <= (arrayLength - copyLength)
885 // We require that all of the values above, except for the sum and
886 // difference, are already known to be non-negative.
887 // Plan A is robust in the face of overflow, if offset and copyLength
888 // are both hugely positive.
889 //
890 // Plan B is less direct and intuitive, but it does not overflow at
891 // all, since the difference of two non-negatives is always
892 // representable. Whenever Java methods must perform the equivalent
893 // check they generally use Plan B instead of Plan A.
894 // For the moment we use Plan A.
895 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
896 Node* subseq_length,
897 Node* array_length,
898 RegionNode* region) {
899 if (stopped())
900 return nullptr; // already stopped
901 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
902 if (zero_offset && subseq_length->eqv_uncast(array_length))
903 return nullptr; // common case of whole-array copy
904 Node* last = subseq_length;
905 if (!zero_offset) // last += offset
906 last = _gvn.transform(new AddINode(last, offset));
907 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
908 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
909 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
910 return is_over;
911 }
912
913 // Emit range checks for the given String.value byte array
914 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
915 if (stopped()) {
916 return; // already stopped
917 }
918 RegionNode* bailout = new RegionNode(1);
919 record_for_igvn(bailout);
920 if (char_count) {
921 // Convert char count to byte count
922 count = _gvn.transform(new LShiftINode(count, intcon(1)));
923 }
924
925 // Offset and count must not be negative
926 generate_negative_guard(offset, bailout);
927 generate_negative_guard(count, bailout);
928 // Offset + count must not exceed length of array
929 generate_limit_guard(offset, count, load_array_length(array), bailout);
930
931 if (bailout->req() > 1) {
932 PreserveJVMState pjvms(this);
933 set_control(_gvn.transform(bailout));
934 uncommon_trap(Deoptimization::Reason_intrinsic,
935 Deoptimization::Action_maybe_recompile);
936 }
937 }
938
939 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
940 bool is_immutable) {
941 ciKlass* thread_klass = env()->Thread_klass();
942 const Type* thread_type
943 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
944
945 Node* thread = _gvn.transform(new ThreadLocalNode());
946 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
947 tls_output = thread;
948
949 Node* thread_obj_handle
950 = (is_immutable
951 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
952 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
953 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
954 thread_obj_handle = _gvn.transform(thread_obj_handle);
955
956 DecoratorSet decorators = IN_NATIVE;
957 if (is_immutable) {
958 decorators |= C2_IMMUTABLE_MEMORY;
959 }
960 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
961 }
962
963 //--------------------------generate_current_thread--------------------
964 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
965 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
966 /*is_immutable*/false);
967 }
968
969 //--------------------------generate_virtual_thread--------------------
970 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
971 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
972 !C->method()->changes_current_thread());
973 }
974
975 //------------------------------make_string_method_node------------------------
976 // Helper method for String intrinsic functions. This version is called with
977 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
978 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
979 // containing the lengths of str1 and str2.
980 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
981 Node* result = nullptr;
982 switch (opcode) {
983 case Op_StrIndexOf:
984 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
985 str1_start, cnt1, str2_start, cnt2, ae);
986 break;
987 case Op_StrComp:
988 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
989 str1_start, cnt1, str2_start, cnt2, ae);
990 break;
991 case Op_StrEquals:
992 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
993 // Use the constant length if there is one because optimized match rule may exist.
994 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
995 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
996 break;
997 default:
998 ShouldNotReachHere();
999 return nullptr;
1000 }
1001
1002 // All these intrinsics have checks.
1003 C->set_has_split_ifs(true); // Has chance for split-if optimization
1004 clear_upper_avx();
1005
1006 return _gvn.transform(result);
1007 }
1008
1009 //------------------------------inline_string_compareTo------------------------
1010 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1011 Node* arg1 = argument(0);
1012 Node* arg2 = argument(1);
1013
1014 arg1 = must_be_not_null(arg1, true);
1015 arg2 = must_be_not_null(arg2, true);
1016
1017 // Get start addr and length of first argument
1018 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1019 Node* arg1_cnt = load_array_length(arg1);
1020
1021 // Get start addr and length of second argument
1022 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1023 Node* arg2_cnt = load_array_length(arg2);
1024
1025 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1026 set_result(result);
1027 return true;
1028 }
1029
1030 //------------------------------inline_string_equals------------------------
1031 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1032 Node* arg1 = argument(0);
1033 Node* arg2 = argument(1);
1034
1035 // paths (plus control) merge
1036 RegionNode* region = new RegionNode(3);
1037 Node* phi = new PhiNode(region, TypeInt::BOOL);
1038
1039 if (!stopped()) {
1040
1041 arg1 = must_be_not_null(arg1, true);
1042 arg2 = must_be_not_null(arg2, true);
1043
1044 // Get start addr and length of first argument
1045 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1046 Node* arg1_cnt = load_array_length(arg1);
1047
1048 // Get start addr and length of second argument
1049 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1050 Node* arg2_cnt = load_array_length(arg2);
1051
1052 // Check for arg1_cnt != arg2_cnt
1053 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1054 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1055 Node* if_ne = generate_slow_guard(bol, nullptr);
1056 if (if_ne != nullptr) {
1057 phi->init_req(2, intcon(0));
1058 region->init_req(2, if_ne);
1059 }
1060
1061 // Check for count == 0 is done by assembler code for StrEquals.
1062
1063 if (!stopped()) {
1064 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1065 phi->init_req(1, equals);
1066 region->init_req(1, control());
1067 }
1068 }
1069
1070 // post merge
1071 set_control(_gvn.transform(region));
1072 record_for_igvn(region);
1073
1074 set_result(_gvn.transform(phi));
1075 return true;
1076 }
1077
1078 //------------------------------inline_array_equals----------------------------
1079 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1080 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1081 Node* arg1 = argument(0);
1082 Node* arg2 = argument(1);
1083
1084 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1085 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1086 clear_upper_avx();
1087
1088 return true;
1089 }
1090
1091
1092 //------------------------------inline_countPositives------------------------------
1093 bool LibraryCallKit::inline_countPositives() {
1094 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1095 return false;
1096 }
1097
1098 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1099 // no receiver since it is static method
1100 Node* ba = argument(0);
1101 Node* offset = argument(1);
1102 Node* len = argument(2);
1103
1104 ba = must_be_not_null(ba, true);
1105
1106 // Range checks
1107 generate_string_range_check(ba, offset, len, false);
1108 if (stopped()) {
1109 return true;
1110 }
1111 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1112 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1113 set_result(_gvn.transform(result));
1114 clear_upper_avx();
1115 return true;
1116 }
1117
1118 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1119 Node* index = argument(0);
1120 Node* length = bt == T_INT ? argument(1) : argument(2);
1121 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1122 return false;
1123 }
1124
1125 // check that length is positive
1126 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1127 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1128
1129 {
1130 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1131 uncommon_trap(Deoptimization::Reason_intrinsic,
1132 Deoptimization::Action_make_not_entrant);
1133 }
1134
1135 if (stopped()) {
1136 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1137 return true;
1138 }
1139
1140 // length is now known positive, add a cast node to make this explicit
1141 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1142 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1143 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1144 ConstraintCastNode::RegularDependency, bt);
1145 casted_length = _gvn.transform(casted_length);
1146 replace_in_map(length, casted_length);
1147 length = casted_length;
1148
1149 // Use an unsigned comparison for the range check itself
1150 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1151 BoolTest::mask btest = BoolTest::lt;
1152 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1153 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1154 _gvn.set_type(rc, rc->Value(&_gvn));
1155 if (!rc_bool->is_Con()) {
1156 record_for_igvn(rc);
1157 }
1158 set_control(_gvn.transform(new IfTrueNode(rc)));
1159 {
1160 PreserveJVMState pjvms(this);
1161 set_control(_gvn.transform(new IfFalseNode(rc)));
1162 uncommon_trap(Deoptimization::Reason_range_check,
1163 Deoptimization::Action_make_not_entrant);
1164 }
1165
1166 if (stopped()) {
1167 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1168 return true;
1169 }
1170
1171 // index is now known to be >= 0 and < length, cast it
1172 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1173 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1174 ConstraintCastNode::RegularDependency, bt);
1175 result = _gvn.transform(result);
1176 set_result(result);
1177 replace_in_map(index, result);
1178 return true;
1179 }
1180
1181 //------------------------------inline_string_indexOf------------------------
1182 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1183 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1184 return false;
1185 }
1186 Node* src = argument(0);
1187 Node* tgt = argument(1);
1188
1189 // Make the merge point
1190 RegionNode* result_rgn = new RegionNode(4);
1191 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1192
1193 src = must_be_not_null(src, true);
1194 tgt = must_be_not_null(tgt, true);
1195
1196 // Get start addr and length of source string
1197 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1198 Node* src_count = load_array_length(src);
1199
1200 // Get start addr and length of substring
1201 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1202 Node* tgt_count = load_array_length(tgt);
1203
1204 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1205 // Divide src size by 2 if String is UTF16 encoded
1206 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1207 }
1208 if (ae == StrIntrinsicNode::UU) {
1209 // Divide substring size by 2 if String is UTF16 encoded
1210 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1211 }
1212
1213 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1214 if (result != nullptr) {
1215 result_phi->init_req(3, result);
1216 result_rgn->init_req(3, control());
1217 }
1218 set_control(_gvn.transform(result_rgn));
1219 record_for_igvn(result_rgn);
1220 set_result(_gvn.transform(result_phi));
1221
1222 return true;
1223 }
1224
1225 //-----------------------------inline_string_indexOf-----------------------
1226 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1227 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1228 return false;
1229 }
1230 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1231 return false;
1232 }
1233 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1234 Node* src = argument(0); // byte[]
1235 Node* src_count = argument(1); // char count
1236 Node* tgt = argument(2); // byte[]
1237 Node* tgt_count = argument(3); // char count
1238 Node* from_index = argument(4); // char index
1239
1240 src = must_be_not_null(src, true);
1241 tgt = must_be_not_null(tgt, true);
1242
1243 // Multiply byte array index by 2 if String is UTF16 encoded
1244 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1245 src_count = _gvn.transform(new SubINode(src_count, from_index));
1246 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1247 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1248
1249 // Range checks
1250 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1251 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1252 if (stopped()) {
1253 return true;
1254 }
1255
1256 RegionNode* region = new RegionNode(5);
1257 Node* phi = new PhiNode(region, TypeInt::INT);
1258
1259 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1260 if (result != nullptr) {
1261 // The result is index relative to from_index if substring was found, -1 otherwise.
1262 // Generate code which will fold into cmove.
1263 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1264 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1265
1266 Node* if_lt = generate_slow_guard(bol, nullptr);
1267 if (if_lt != nullptr) {
1268 // result == -1
1269 phi->init_req(3, result);
1270 region->init_req(3, if_lt);
1271 }
1272 if (!stopped()) {
1273 result = _gvn.transform(new AddINode(result, from_index));
1274 phi->init_req(4, result);
1275 region->init_req(4, control());
1276 }
1277 }
1278
1279 set_control(_gvn.transform(region));
1280 record_for_igvn(region);
1281 set_result(_gvn.transform(phi));
1282 clear_upper_avx();
1283
1284 return true;
1285 }
1286
1287 // Create StrIndexOfNode with fast path checks
1288 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1289 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1290 // Check for substr count > string count
1291 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1292 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1293 Node* if_gt = generate_slow_guard(bol, nullptr);
1294 if (if_gt != nullptr) {
1295 phi->init_req(1, intcon(-1));
1296 region->init_req(1, if_gt);
1297 }
1298 if (!stopped()) {
1299 // Check for substr count == 0
1300 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1301 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1302 Node* if_zero = generate_slow_guard(bol, nullptr);
1303 if (if_zero != nullptr) {
1304 phi->init_req(2, intcon(0));
1305 region->init_req(2, if_zero);
1306 }
1307 }
1308 if (!stopped()) {
1309 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1310 }
1311 return nullptr;
1312 }
1313
1314 //-----------------------------inline_string_indexOfChar-----------------------
1315 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1316 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1317 return false;
1318 }
1319 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1320 return false;
1321 }
1322 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1323 Node* src = argument(0); // byte[]
1324 Node* int_ch = argument(1);
1325 Node* from_index = argument(2);
1326 Node* max = argument(3);
1327
1328 src = must_be_not_null(src, true);
1329
1330 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1331 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1332 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1333
1334 // Range checks
1335 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1336
1337 // Check for int_ch >= 0
1338 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1339 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1340 {
1341 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1342 uncommon_trap(Deoptimization::Reason_intrinsic,
1343 Deoptimization::Action_maybe_recompile);
1344 }
1345 if (stopped()) {
1346 return true;
1347 }
1348
1349 RegionNode* region = new RegionNode(3);
1350 Node* phi = new PhiNode(region, TypeInt::INT);
1351
1352 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1353 C->set_has_split_ifs(true); // Has chance for split-if optimization
1354 _gvn.transform(result);
1355
1356 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1357 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1358
1359 Node* if_lt = generate_slow_guard(bol, nullptr);
1360 if (if_lt != nullptr) {
1361 // result == -1
1362 phi->init_req(2, result);
1363 region->init_req(2, if_lt);
1364 }
1365 if (!stopped()) {
1366 result = _gvn.transform(new AddINode(result, from_index));
1367 phi->init_req(1, result);
1368 region->init_req(1, control());
1369 }
1370 set_control(_gvn.transform(region));
1371 record_for_igvn(region);
1372 set_result(_gvn.transform(phi));
1373 clear_upper_avx();
1374
1375 return true;
1376 }
1377 //---------------------------inline_string_copy---------------------
1378 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1379 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1380 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1381 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1382 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1383 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1384 bool LibraryCallKit::inline_string_copy(bool compress) {
1385 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1386 return false;
1387 }
1388 int nargs = 5; // 2 oops, 3 ints
1389 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1390
1391 Node* src = argument(0);
1392 Node* src_offset = argument(1);
1393 Node* dst = argument(2);
1394 Node* dst_offset = argument(3);
1395 Node* length = argument(4);
1396
1397 // Check for allocation before we add nodes that would confuse
1398 // tightly_coupled_allocation()
1399 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1400
1401 // Figure out the size and type of the elements we will be copying.
1402 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1403 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1404 if (src_type == nullptr || dst_type == nullptr) {
1405 return false;
1406 }
1407 BasicType src_elem = src_type->elem()->array_element_basic_type();
1408 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1409 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1410 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1411 "Unsupported array types for inline_string_copy");
1412
1413 src = must_be_not_null(src, true);
1414 dst = must_be_not_null(dst, true);
1415
1416 // Convert char[] offsets to byte[] offsets
1417 bool convert_src = (compress && src_elem == T_BYTE);
1418 bool convert_dst = (!compress && dst_elem == T_BYTE);
1419 if (convert_src) {
1420 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1421 } else if (convert_dst) {
1422 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1423 }
1424
1425 // Range checks
1426 generate_string_range_check(src, src_offset, length, convert_src);
1427 generate_string_range_check(dst, dst_offset, length, convert_dst);
1428 if (stopped()) {
1429 return true;
1430 }
1431
1432 Node* src_start = array_element_address(src, src_offset, src_elem);
1433 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1434 // 'src_start' points to src array + scaled offset
1435 // 'dst_start' points to dst array + scaled offset
1436 Node* count = nullptr;
1437 if (compress) {
1438 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1439 } else {
1440 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1441 }
1442
1443 if (alloc != nullptr) {
1444 if (alloc->maybe_set_complete(&_gvn)) {
1445 // "You break it, you buy it."
1446 InitializeNode* init = alloc->initialization();
1447 assert(init->is_complete(), "we just did this");
1448 init->set_complete_with_arraycopy();
1449 assert(dst->is_CheckCastPP(), "sanity");
1450 assert(dst->in(0)->in(0) == init, "dest pinned");
1451 }
1452 // Do not let stores that initialize this object be reordered with
1453 // a subsequent store that would make this object accessible by
1454 // other threads.
1455 // Record what AllocateNode this StoreStore protects so that
1456 // escape analysis can go from the MemBarStoreStoreNode to the
1457 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1458 // based on the escape status of the AllocateNode.
1459 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1460 }
1461 if (compress) {
1462 set_result(_gvn.transform(count));
1463 }
1464 clear_upper_avx();
1465
1466 return true;
1467 }
1468
1469 #ifdef _LP64
1470 #define XTOP ,top() /*additional argument*/
1471 #else //_LP64
1472 #define XTOP /*no additional argument*/
1473 #endif //_LP64
1474
1475 //------------------------inline_string_toBytesU--------------------------
1476 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1477 bool LibraryCallKit::inline_string_toBytesU() {
1478 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1479 return false;
1480 }
1481 // Get the arguments.
1482 Node* value = argument(0);
1483 Node* offset = argument(1);
1484 Node* length = argument(2);
1485
1486 Node* newcopy = nullptr;
1487
1488 // Set the original stack and the reexecute bit for the interpreter to reexecute
1489 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1490 { PreserveReexecuteState preexecs(this);
1491 jvms()->set_should_reexecute(true);
1492
1493 // Check if a null path was taken unconditionally.
1494 value = null_check(value);
1495
1496 RegionNode* bailout = new RegionNode(1);
1497 record_for_igvn(bailout);
1498
1499 // Range checks
1500 generate_negative_guard(offset, bailout);
1501 generate_negative_guard(length, bailout);
1502 generate_limit_guard(offset, length, load_array_length(value), bailout);
1503 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1504 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1505
1506 if (bailout->req() > 1) {
1507 PreserveJVMState pjvms(this);
1508 set_control(_gvn.transform(bailout));
1509 uncommon_trap(Deoptimization::Reason_intrinsic,
1510 Deoptimization::Action_maybe_recompile);
1511 }
1512 if (stopped()) {
1513 return true;
1514 }
1515
1516 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1517 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1518 newcopy = new_array(klass_node, size, 0); // no arguments to push
1519 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1520 guarantee(alloc != nullptr, "created above");
1521
1522 // Calculate starting addresses.
1523 Node* src_start = array_element_address(value, offset, T_CHAR);
1524 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1525
1526 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1527 const TypeInt* toffset = gvn().type(offset)->is_int();
1528 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1529
1530 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1531 const char* copyfunc_name = "arraycopy";
1532 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1533 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1534 OptoRuntime::fast_arraycopy_Type(),
1535 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1536 src_start, dst_start, ConvI2X(length) XTOP);
1537 // Do not let reads from the cloned object float above the arraycopy.
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(newcopy->is_CheckCastPP(), "sanity");
1544 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1554 } // original reexecute is set back here
1555
1556 C->set_has_split_ifs(true); // Has chance for split-if optimization
1557 if (!stopped()) {
1558 set_result(newcopy);
1559 }
1560 clear_upper_avx();
1561
1562 return true;
1563 }
1564
1565 //------------------------inline_string_getCharsU--------------------------
1566 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1567 bool LibraryCallKit::inline_string_getCharsU() {
1568 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1569 return false;
1570 }
1571
1572 // Get the arguments.
1573 Node* src = argument(0);
1574 Node* src_begin = argument(1);
1575 Node* src_end = argument(2); // exclusive offset (i < src_end)
1576 Node* dst = argument(3);
1577 Node* dst_begin = argument(4);
1578
1579 // Check for allocation before we add nodes that would confuse
1580 // tightly_coupled_allocation()
1581 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1582
1583 // Check if a null path was taken unconditionally.
1584 src = null_check(src);
1585 dst = null_check(dst);
1586 if (stopped()) {
1587 return true;
1588 }
1589
1590 // Get length and convert char[] offset to byte[] offset
1591 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1592 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1593
1594 // Range checks
1595 generate_string_range_check(src, src_begin, length, true);
1596 generate_string_range_check(dst, dst_begin, length, false);
1597 if (stopped()) {
1598 return true;
1599 }
1600
1601 if (!stopped()) {
1602 // Calculate starting addresses.
1603 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1604 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1605
1606 // Check if array addresses are aligned to HeapWordSize
1607 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1608 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1609 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1610 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1611
1612 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1613 const char* copyfunc_name = "arraycopy";
1614 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1615 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1616 OptoRuntime::fast_arraycopy_Type(),
1617 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1618 src_start, dst_start, ConvI2X(length) XTOP);
1619 // Do not let reads from the cloned object float above the arraycopy.
1620 if (alloc != nullptr) {
1621 if (alloc->maybe_set_complete(&_gvn)) {
1622 // "You break it, you buy it."
1623 InitializeNode* init = alloc->initialization();
1624 assert(init->is_complete(), "we just did this");
1625 init->set_complete_with_arraycopy();
1626 assert(dst->is_CheckCastPP(), "sanity");
1627 assert(dst->in(0)->in(0) == init, "dest pinned");
1628 }
1629 // Do not let stores that initialize this object be reordered with
1630 // a subsequent store that would make this object accessible by
1631 // other threads.
1632 // Record what AllocateNode this StoreStore protects so that
1633 // escape analysis can go from the MemBarStoreStoreNode to the
1634 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1635 // based on the escape status of the AllocateNode.
1636 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1637 } else {
1638 insert_mem_bar(Op_MemBarCPUOrder);
1639 }
1640 }
1641
1642 C->set_has_split_ifs(true); // Has chance for split-if optimization
1643 return true;
1644 }
1645
1646 //----------------------inline_string_char_access----------------------------
1647 // Store/Load char to/from byte[] array.
1648 // static void StringUTF16.putChar(byte[] val, int index, int c)
1649 // static char StringUTF16.getChar(byte[] val, int index)
1650 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1651 Node* value = argument(0);
1652 Node* index = argument(1);
1653 Node* ch = is_store ? argument(2) : nullptr;
1654
1655 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1656 // correctly requires matched array shapes.
1657 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1658 "sanity: byte[] and char[] bases agree");
1659 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1660 "sanity: byte[] and char[] scales agree");
1661
1662 // Bail when getChar over constants is requested: constant folding would
1663 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1664 // Java method would constant fold nicely instead.
1665 if (!is_store && value->is_Con() && index->is_Con()) {
1666 return false;
1667 }
1668
1669 // Save state and restore on bailout
1670 uint old_sp = sp();
1671 SafePointNode* old_map = clone_map();
1672
1673 value = must_be_not_null(value, true);
1674
1675 Node* adr = array_element_address(value, index, T_CHAR);
1676 if (adr->is_top()) {
1677 set_map(old_map);
1678 set_sp(old_sp);
1679 return false;
1680 }
1681 destruct_map_clone(old_map);
1682 if (is_store) {
1683 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1684 } else {
1685 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);
1686 set_result(ch);
1687 }
1688 return true;
1689 }
1690
1691 //--------------------------round_double_node--------------------------------
1692 // Round a double node if necessary.
1693 Node* LibraryCallKit::round_double_node(Node* n) {
1694 if (Matcher::strict_fp_requires_explicit_rounding) {
1695 #ifdef IA32
1696 if (UseSSE < 2) {
1697 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1698 }
1699 #else
1700 Unimplemented();
1701 #endif // IA32
1702 }
1703 return n;
1704 }
1705
1706 //------------------------------inline_math-----------------------------------
1707 // public static double Math.abs(double)
1708 // public static double Math.sqrt(double)
1709 // public static double Math.log(double)
1710 // public static double Math.log10(double)
1711 // public static double Math.round(double)
1712 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1713 Node* arg = round_double_node(argument(0));
1714 Node* n = nullptr;
1715 switch (id) {
1716 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1717 case vmIntrinsics::_dsqrt:
1718 case vmIntrinsics::_dsqrt_strict:
1719 n = new SqrtDNode(C, control(), arg); break;
1720 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1721 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1722 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1723 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1724 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1725 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1726 default: fatal_unexpected_iid(id); break;
1727 }
1728 set_result(_gvn.transform(n));
1729 return true;
1730 }
1731
1732 //------------------------------inline_math-----------------------------------
1733 // public static float Math.abs(float)
1734 // public static int Math.abs(int)
1735 // public static long Math.abs(long)
1736 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1737 Node* arg = argument(0);
1738 Node* n = nullptr;
1739 switch (id) {
1740 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1741 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1742 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1743 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1744 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1745 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1746 default: fatal_unexpected_iid(id); break;
1747 }
1748 set_result(_gvn.transform(n));
1749 return true;
1750 }
1751
1752 //------------------------------runtime_math-----------------------------
1753 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1754 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1755 "must be (DD)D or (D)D type");
1756
1757 // Inputs
1758 Node* a = round_double_node(argument(0));
1759 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1760
1761 const TypePtr* no_memory_effects = nullptr;
1762 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1763 no_memory_effects,
1764 a, top(), b, b ? top() : nullptr);
1765 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1766 #ifdef ASSERT
1767 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1768 assert(value_top == top(), "second value must be top");
1769 #endif
1770
1771 set_result(value);
1772 return true;
1773 }
1774
1775 //------------------------------inline_math_pow-----------------------------
1776 bool LibraryCallKit::inline_math_pow() {
1777 Node* exp = round_double_node(argument(2));
1778 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1779 if (d != nullptr) {
1780 if (d->getd() == 2.0) {
1781 // Special case: pow(x, 2.0) => x * x
1782 Node* base = round_double_node(argument(0));
1783 set_result(_gvn.transform(new MulDNode(base, base)));
1784 return true;
1785 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1786 // Special case: pow(x, 0.5) => sqrt(x)
1787 Node* base = round_double_node(argument(0));
1788 Node* zero = _gvn.zerocon(T_DOUBLE);
1789
1790 RegionNode* region = new RegionNode(3);
1791 Node* phi = new PhiNode(region, Type::DOUBLE);
1792
1793 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1794 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1795 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1796 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1797 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1798
1799 Node* if_pow = generate_slow_guard(test, nullptr);
1800 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1801 phi->init_req(1, value_sqrt);
1802 region->init_req(1, control());
1803
1804 if (if_pow != nullptr) {
1805 set_control(if_pow);
1806 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1807 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1808 const TypePtr* no_memory_effects = nullptr;
1809 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1810 no_memory_effects, base, top(), exp, top());
1811 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1812 #ifdef ASSERT
1813 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1814 assert(value_top == top(), "second value must be top");
1815 #endif
1816 phi->init_req(2, value_pow);
1817 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1818 }
1819
1820 C->set_has_split_ifs(true); // Has chance for split-if optimization
1821 set_control(_gvn.transform(region));
1822 record_for_igvn(region);
1823 set_result(_gvn.transform(phi));
1824
1825 return true;
1826 }
1827 }
1828
1829 return StubRoutines::dpow() != nullptr ?
1830 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1831 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1832 }
1833
1834 //------------------------------inline_math_native-----------------------------
1835 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1836 switch (id) {
1837 case vmIntrinsics::_dsin:
1838 return StubRoutines::dsin() != nullptr ?
1839 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1840 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1841 case vmIntrinsics::_dcos:
1842 return StubRoutines::dcos() != nullptr ?
1843 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1844 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1845 case vmIntrinsics::_dtan:
1846 return StubRoutines::dtan() != nullptr ?
1847 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1848 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1849 case vmIntrinsics::_dexp:
1850 return StubRoutines::dexp() != nullptr ?
1851 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1852 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1853 case vmIntrinsics::_dlog:
1854 return StubRoutines::dlog() != nullptr ?
1855 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1856 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1857 case vmIntrinsics::_dlog10:
1858 return StubRoutines::dlog10() != nullptr ?
1859 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1860 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1861
1862 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1863 case vmIntrinsics::_ceil:
1864 case vmIntrinsics::_floor:
1865 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1866
1867 case vmIntrinsics::_dsqrt:
1868 case vmIntrinsics::_dsqrt_strict:
1869 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1870 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1871 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1872 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1873 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1874
1875 case vmIntrinsics::_dpow: return inline_math_pow();
1876 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1877 case vmIntrinsics::_fcopySign: return inline_math(id);
1878 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1879 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1880 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1881
1882 // These intrinsics are not yet correctly implemented
1883 case vmIntrinsics::_datan2:
1884 return false;
1885
1886 default:
1887 fatal_unexpected_iid(id);
1888 return false;
1889 }
1890 }
1891
1892 //----------------------------inline_notify-----------------------------------*
1893 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1894 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1895 address func;
1896 if (id == vmIntrinsics::_notify) {
1897 func = OptoRuntime::monitor_notify_Java();
1898 } else {
1899 func = OptoRuntime::monitor_notifyAll_Java();
1900 }
1901 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1902 make_slow_call_ex(call, env()->Throwable_klass(), false);
1903 return true;
1904 }
1905
1906
1907 //----------------------------inline_min_max-----------------------------------
1908 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1909 set_result(generate_min_max(id, argument(0), argument(1)));
1910 return true;
1911 }
1912
1913 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1914 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1915 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1916 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1917 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1918
1919 {
1920 PreserveJVMState pjvms(this);
1921 PreserveReexecuteState preexecs(this);
1922 jvms()->set_should_reexecute(true);
1923
1924 set_control(slow_path);
1925 set_i_o(i_o());
1926
1927 uncommon_trap(Deoptimization::Reason_intrinsic,
1928 Deoptimization::Action_none);
1929 }
1930
1931 set_control(fast_path);
1932 set_result(math);
1933 }
1934
1935 template <typename OverflowOp>
1936 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1937 typedef typename OverflowOp::MathOp MathOp;
1938
1939 MathOp* mathOp = new MathOp(arg1, arg2);
1940 Node* operation = _gvn.transform( mathOp );
1941 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1942 inline_math_mathExact(operation, ofcheck);
1943 return true;
1944 }
1945
1946 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1947 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1948 }
1949
1950 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1951 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1952 }
1953
1954 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1955 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1956 }
1957
1958 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1959 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1960 }
1961
1962 bool LibraryCallKit::inline_math_negateExactI() {
1963 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1964 }
1965
1966 bool LibraryCallKit::inline_math_negateExactL() {
1967 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1968 }
1969
1970 bool LibraryCallKit::inline_math_multiplyExactI() {
1971 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1972 }
1973
1974 bool LibraryCallKit::inline_math_multiplyExactL() {
1975 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1976 }
1977
1978 bool LibraryCallKit::inline_math_multiplyHigh() {
1979 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1980 return true;
1981 }
1982
1983 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1984 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1985 return true;
1986 }
1987
1988 Node*
1989 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1990 Node* result_val = nullptr;
1991 switch (id) {
1992 case vmIntrinsics::_min:
1993 case vmIntrinsics::_min_strict:
1994 result_val = _gvn.transform(new MinINode(x0, y0));
1995 break;
1996 case vmIntrinsics::_max:
1997 case vmIntrinsics::_max_strict:
1998 result_val = _gvn.transform(new MaxINode(x0, y0));
1999 break;
2000 default:
2001 fatal_unexpected_iid(id);
2002 break;
2003 }
2004 return result_val;
2005 }
2006
2007 inline int
2008 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2009 const TypePtr* base_type = TypePtr::NULL_PTR;
2010 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2011 if (base_type == nullptr) {
2012 // Unknown type.
2013 return Type::AnyPtr;
2014 } else if (base_type == TypePtr::NULL_PTR) {
2015 // Since this is a null+long form, we have to switch to a rawptr.
2016 base = _gvn.transform(new CastX2PNode(offset));
2017 offset = MakeConX(0);
2018 return Type::RawPtr;
2019 } else if (base_type->base() == Type::RawPtr) {
2020 return Type::RawPtr;
2021 } else if (base_type->isa_oopptr()) {
2022 // Base is never null => always a heap address.
2023 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2024 return Type::OopPtr;
2025 }
2026 // Offset is small => always a heap address.
2027 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2028 if (offset_type != nullptr &&
2029 base_type->offset() == 0 && // (should always be?)
2030 offset_type->_lo >= 0 &&
2031 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2032 return Type::OopPtr;
2033 } else if (type == T_OBJECT) {
2034 // off heap access to an oop doesn't make any sense. Has to be on
2035 // heap.
2036 return Type::OopPtr;
2037 }
2038 // Otherwise, it might either be oop+off or null+addr.
2039 return Type::AnyPtr;
2040 } else {
2041 // No information:
2042 return Type::AnyPtr;
2043 }
2044 }
2045
2046 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2047 Node* uncasted_base = base;
2048 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2049 if (kind == Type::RawPtr) {
2050 return basic_plus_adr(top(), uncasted_base, offset);
2051 } else if (kind == Type::AnyPtr) {
2052 assert(base == uncasted_base, "unexpected base change");
2053 if (can_cast) {
2054 if (!_gvn.type(base)->speculative_maybe_null() &&
2055 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2056 // According to profiling, this access is always on
2057 // heap. Casting the base to not null and thus avoiding membars
2058 // around the access should allow better optimizations
2059 Node* null_ctl = top();
2060 base = null_check_oop(base, &null_ctl, true, true, true);
2061 assert(null_ctl->is_top(), "no null control here");
2062 return basic_plus_adr(base, offset);
2063 } else if (_gvn.type(base)->speculative_always_null() &&
2064 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2065 // According to profiling, this access is always off
2066 // heap.
2067 base = null_assert(base);
2068 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2069 offset = MakeConX(0);
2070 return basic_plus_adr(top(), raw_base, offset);
2071 }
2072 }
2073 // We don't know if it's an on heap or off heap access. Fall back
2074 // to raw memory access.
2075 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2076 return basic_plus_adr(top(), raw, offset);
2077 } else {
2078 assert(base == uncasted_base, "unexpected base change");
2079 // We know it's an on heap access so base can't be null
2080 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2081 base = must_be_not_null(base, true);
2082 }
2083 return basic_plus_adr(base, offset);
2084 }
2085 }
2086
2087 //--------------------------inline_number_methods-----------------------------
2088 // inline int Integer.numberOfLeadingZeros(int)
2089 // inline int Long.numberOfLeadingZeros(long)
2090 //
2091 // inline int Integer.numberOfTrailingZeros(int)
2092 // inline int Long.numberOfTrailingZeros(long)
2093 //
2094 // inline int Integer.bitCount(int)
2095 // inline int Long.bitCount(long)
2096 //
2097 // inline char Character.reverseBytes(char)
2098 // inline short Short.reverseBytes(short)
2099 // inline int Integer.reverseBytes(int)
2100 // inline long Long.reverseBytes(long)
2101 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2102 Node* arg = argument(0);
2103 Node* n = nullptr;
2104 switch (id) {
2105 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2106 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2107 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2108 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2109 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2110 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2111 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2112 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2113 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2114 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2115 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2116 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2117 default: fatal_unexpected_iid(id); break;
2118 }
2119 set_result(_gvn.transform(n));
2120 return true;
2121 }
2122
2123 //--------------------------inline_bitshuffle_methods-----------------------------
2124 // inline int Integer.compress(int, int)
2125 // inline int Integer.expand(int, int)
2126 // inline long Long.compress(long, long)
2127 // inline long Long.expand(long, long)
2128 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2129 Node* n = nullptr;
2130 switch (id) {
2131 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2132 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2133 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2134 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2135 default: fatal_unexpected_iid(id); break;
2136 }
2137 set_result(_gvn.transform(n));
2138 return true;
2139 }
2140
2141 //--------------------------inline_number_methods-----------------------------
2142 // inline int Integer.compareUnsigned(int, int)
2143 // inline int Long.compareUnsigned(long, long)
2144 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2145 Node* arg1 = argument(0);
2146 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2147 Node* n = nullptr;
2148 switch (id) {
2149 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2150 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2151 default: fatal_unexpected_iid(id); break;
2152 }
2153 set_result(_gvn.transform(n));
2154 return true;
2155 }
2156
2157 //--------------------------inline_unsigned_divmod_methods-----------------------------
2158 // inline int Integer.divideUnsigned(int, int)
2159 // inline int Integer.remainderUnsigned(int, int)
2160 // inline long Long.divideUnsigned(long, long)
2161 // inline long Long.remainderUnsigned(long, long)
2162 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2163 Node* n = nullptr;
2164 switch (id) {
2165 case vmIntrinsics::_divideUnsigned_i: {
2166 zero_check_int(argument(1));
2167 // Compile-time detect of null-exception
2168 if (stopped()) {
2169 return true; // keep the graph constructed so far
2170 }
2171 n = new UDivINode(control(), argument(0), argument(1));
2172 break;
2173 }
2174 case vmIntrinsics::_divideUnsigned_l: {
2175 zero_check_long(argument(2));
2176 // Compile-time detect of null-exception
2177 if (stopped()) {
2178 return true; // keep the graph constructed so far
2179 }
2180 n = new UDivLNode(control(), argument(0), argument(2));
2181 break;
2182 }
2183 case vmIntrinsics::_remainderUnsigned_i: {
2184 zero_check_int(argument(1));
2185 // Compile-time detect of null-exception
2186 if (stopped()) {
2187 return true; // keep the graph constructed so far
2188 }
2189 n = new UModINode(control(), argument(0), argument(1));
2190 break;
2191 }
2192 case vmIntrinsics::_remainderUnsigned_l: {
2193 zero_check_long(argument(2));
2194 // Compile-time detect of null-exception
2195 if (stopped()) {
2196 return true; // keep the graph constructed so far
2197 }
2198 n = new UModLNode(control(), argument(0), argument(2));
2199 break;
2200 }
2201 default: fatal_unexpected_iid(id); break;
2202 }
2203 set_result(_gvn.transform(n));
2204 return true;
2205 }
2206
2207 //----------------------------inline_unsafe_access----------------------------
2208
2209 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2210 // Attempt to infer a sharper value type from the offset and base type.
2211 ciKlass* sharpened_klass = nullptr;
2212
2213 // See if it is an instance field, with an object type.
2214 if (alias_type->field() != nullptr) {
2215 if (alias_type->field()->type()->is_klass()) {
2216 sharpened_klass = alias_type->field()->type()->as_klass();
2217 }
2218 }
2219
2220 const TypeOopPtr* result = nullptr;
2221 // See if it is a narrow oop array.
2222 if (adr_type->isa_aryptr()) {
2223 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2224 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2225 if (elem_type != nullptr && elem_type->is_loaded()) {
2226 // Sharpen the value type.
2227 result = elem_type;
2228 }
2229 }
2230 }
2231
2232 // The sharpened class might be unloaded if there is no class loader
2233 // contraint in place.
2234 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2235 // Sharpen the value type.
2236 result = TypeOopPtr::make_from_klass(sharpened_klass);
2237 }
2238 if (result != nullptr) {
2239 #ifndef PRODUCT
2240 if (C->print_intrinsics() || C->print_inlining()) {
2241 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2242 tty->print(" sharpened value: "); result->dump(); tty->cr();
2243 }
2244 #endif
2245 }
2246 return result;
2247 }
2248
2249 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2250 switch (kind) {
2251 case Relaxed:
2252 return MO_UNORDERED;
2253 case Opaque:
2254 return MO_RELAXED;
2255 case Acquire:
2256 return MO_ACQUIRE;
2257 case Release:
2258 return MO_RELEASE;
2259 case Volatile:
2260 return MO_SEQ_CST;
2261 default:
2262 ShouldNotReachHere();
2263 return 0;
2264 }
2265 }
2266
2267 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2268 if (callee()->is_static()) return false; // caller must have the capability!
2269 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2270 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2271 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2272 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2273
2274 if (is_reference_type(type)) {
2275 decorators |= ON_UNKNOWN_OOP_REF;
2276 }
2277
2278 if (unaligned) {
2279 decorators |= C2_UNALIGNED;
2280 }
2281
2282 #ifndef PRODUCT
2283 {
2284 ResourceMark rm;
2285 // Check the signatures.
2286 ciSignature* sig = callee()->signature();
2287 #ifdef ASSERT
2288 if (!is_store) {
2289 // Object getReference(Object base, int/long offset), etc.
2290 BasicType rtype = sig->return_type()->basic_type();
2291 assert(rtype == type, "getter must return the expected value");
2292 assert(sig->count() == 2, "oop getter has 2 arguments");
2293 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2294 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2295 } else {
2296 // void putReference(Object base, int/long offset, Object x), etc.
2297 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2298 assert(sig->count() == 3, "oop putter has 3 arguments");
2299 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2300 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2301 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2302 assert(vtype == type, "putter must accept the expected value");
2303 }
2304 #endif // ASSERT
2305 }
2306 #endif //PRODUCT
2307
2308 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2309
2310 Node* receiver = argument(0); // type: oop
2311
2312 // Build address expression.
2313 Node* heap_base_oop = top();
2314
2315 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2316 Node* base = argument(1); // type: oop
2317 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2318 Node* offset = argument(2); // type: long
2319 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2320 // to be plain byte offsets, which are also the same as those accepted
2321 // by oopDesc::field_addr.
2322 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2323 "fieldOffset must be byte-scaled");
2324 // 32-bit machines ignore the high half!
2325 offset = ConvL2X(offset);
2326
2327 // Save state and restore on bailout
2328 uint old_sp = sp();
2329 SafePointNode* old_map = clone_map();
2330
2331 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2332
2333 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2334 if (type != T_OBJECT) {
2335 decorators |= IN_NATIVE; // off-heap primitive access
2336 } else {
2337 set_map(old_map);
2338 set_sp(old_sp);
2339 return false; // off-heap oop accesses are not supported
2340 }
2341 } else {
2342 heap_base_oop = base; // on-heap or mixed access
2343 }
2344
2345 // Can base be null? Otherwise, always on-heap access.
2346 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2347
2348 if (!can_access_non_heap) {
2349 decorators |= IN_HEAP;
2350 }
2351
2352 Node* val = is_store ? argument(4) : nullptr;
2353
2354 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2355 if (adr_type == TypePtr::NULL_PTR) {
2356 set_map(old_map);
2357 set_sp(old_sp);
2358 return false; // off-heap access with zero address
2359 }
2360
2361 // Try to categorize the address.
2362 Compile::AliasType* alias_type = C->alias_type(adr_type);
2363 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2364
2365 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2366 alias_type->adr_type() == TypeAryPtr::RANGE) {
2367 set_map(old_map);
2368 set_sp(old_sp);
2369 return false; // not supported
2370 }
2371
2372 bool mismatched = false;
2373 BasicType bt = alias_type->basic_type();
2374 if (bt != T_ILLEGAL) {
2375 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2376 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2377 // Alias type doesn't differentiate between byte[] and boolean[]).
2378 // Use address type to get the element type.
2379 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2380 }
2381 if (is_reference_type(bt, true)) {
2382 // accessing an array field with getReference is not a mismatch
2383 bt = T_OBJECT;
2384 }
2385 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2386 // Don't intrinsify mismatched object accesses
2387 set_map(old_map);
2388 set_sp(old_sp);
2389 return false;
2390 }
2391 mismatched = (bt != type);
2392 } else if (alias_type->adr_type()->isa_oopptr()) {
2393 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2394 }
2395
2396 destruct_map_clone(old_map);
2397 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2398
2399 if (mismatched) {
2400 decorators |= C2_MISMATCHED;
2401 }
2402
2403 // First guess at the value type.
2404 const Type *value_type = Type::get_const_basic_type(type);
2405
2406 // Figure out the memory ordering.
2407 decorators |= mo_decorator_for_access_kind(kind);
2408
2409 if (!is_store && type == T_OBJECT) {
2410 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2411 if (tjp != nullptr) {
2412 value_type = tjp;
2413 }
2414 }
2415
2416 receiver = null_check(receiver);
2417 if (stopped()) {
2418 return true;
2419 }
2420 // Heap pointers get a null-check from the interpreter,
2421 // as a courtesy. However, this is not guaranteed by Unsafe,
2422 // and it is not possible to fully distinguish unintended nulls
2423 // from intended ones in this API.
2424
2425 if (!is_store) {
2426 Node* p = nullptr;
2427 // Try to constant fold a load from a constant field
2428 ciField* field = alias_type->field();
2429 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2430 // final or stable field
2431 p = make_constant_from_field(field, heap_base_oop);
2432 }
2433
2434 if (p == nullptr) { // Could not constant fold the load
2435 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2436 // Normalize the value returned by getBoolean in the following cases
2437 if (type == T_BOOLEAN &&
2438 (mismatched ||
2439 heap_base_oop == top() || // - heap_base_oop is null or
2440 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2441 // and the unsafe access is made to large offset
2442 // (i.e., larger than the maximum offset necessary for any
2443 // field access)
2444 ) {
2445 IdealKit ideal = IdealKit(this);
2446 #define __ ideal.
2447 IdealVariable normalized_result(ideal);
2448 __ declarations_done();
2449 __ set(normalized_result, p);
2450 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2451 __ set(normalized_result, ideal.ConI(1));
2452 ideal.end_if();
2453 final_sync(ideal);
2454 p = __ value(normalized_result);
2455 #undef __
2456 }
2457 }
2458 if (type == T_ADDRESS) {
2459 p = gvn().transform(new CastP2XNode(nullptr, p));
2460 p = ConvX2UL(p);
2461 }
2462 // The load node has the control of the preceding MemBarCPUOrder. All
2463 // following nodes will have the control of the MemBarCPUOrder inserted at
2464 // the end of this method. So, pushing the load onto the stack at a later
2465 // point is fine.
2466 set_result(p);
2467 } else {
2468 if (bt == T_ADDRESS) {
2469 // Repackage the long as a pointer.
2470 val = ConvL2X(val);
2471 val = gvn().transform(new CastX2PNode(val));
2472 }
2473 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2474 }
2475
2476 return true;
2477 }
2478
2479 //----------------------------inline_unsafe_load_store----------------------------
2480 // This method serves a couple of different customers (depending on LoadStoreKind):
2481 //
2482 // LS_cmp_swap:
2483 //
2484 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2485 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2486 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2487 //
2488 // LS_cmp_swap_weak:
2489 //
2490 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2491 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2492 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2493 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2494 //
2495 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2496 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2497 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2498 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2499 //
2500 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2501 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2502 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2503 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2504 //
2505 // LS_cmp_exchange:
2506 //
2507 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2508 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2510 //
2511 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2512 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2513 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2514 //
2515 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2516 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2517 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2518 //
2519 // LS_get_add:
2520 //
2521 // int getAndAddInt( Object o, long offset, int delta)
2522 // long getAndAddLong(Object o, long offset, long delta)
2523 //
2524 // LS_get_set:
2525 //
2526 // int getAndSet(Object o, long offset, int newValue)
2527 // long getAndSet(Object o, long offset, long newValue)
2528 // Object getAndSet(Object o, long offset, Object newValue)
2529 //
2530 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2531 // This basic scheme here is the same as inline_unsafe_access, but
2532 // differs in enough details that combining them would make the code
2533 // overly confusing. (This is a true fact! I originally combined
2534 // them, but even I was confused by it!) As much code/comments as
2535 // possible are retained from inline_unsafe_access though to make
2536 // the correspondences clearer. - dl
2537
2538 if (callee()->is_static()) return false; // caller must have the capability!
2539
2540 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2541 decorators |= mo_decorator_for_access_kind(access_kind);
2542
2543 #ifndef PRODUCT
2544 BasicType rtype;
2545 {
2546 ResourceMark rm;
2547 // Check the signatures.
2548 ciSignature* sig = callee()->signature();
2549 rtype = sig->return_type()->basic_type();
2550 switch(kind) {
2551 case LS_get_add:
2552 case LS_get_set: {
2553 // Check the signatures.
2554 #ifdef ASSERT
2555 assert(rtype == type, "get and set must return the expected type");
2556 assert(sig->count() == 3, "get and set has 3 arguments");
2557 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2558 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2559 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2560 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2561 #endif // ASSERT
2562 break;
2563 }
2564 case LS_cmp_swap:
2565 case LS_cmp_swap_weak: {
2566 // Check the signatures.
2567 #ifdef ASSERT
2568 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2569 assert(sig->count() == 4, "CAS has 4 arguments");
2570 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2571 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2572 #endif // ASSERT
2573 break;
2574 }
2575 case LS_cmp_exchange: {
2576 // Check the signatures.
2577 #ifdef ASSERT
2578 assert(rtype == type, "CAS must return the expected type");
2579 assert(sig->count() == 4, "CAS has 4 arguments");
2580 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2581 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2582 #endif // ASSERT
2583 break;
2584 }
2585 default:
2586 ShouldNotReachHere();
2587 }
2588 }
2589 #endif //PRODUCT
2590
2591 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2592
2593 // Get arguments:
2594 Node* receiver = nullptr;
2595 Node* base = nullptr;
2596 Node* offset = nullptr;
2597 Node* oldval = nullptr;
2598 Node* newval = nullptr;
2599 switch(kind) {
2600 case LS_cmp_swap:
2601 case LS_cmp_swap_weak:
2602 case LS_cmp_exchange: {
2603 const bool two_slot_type = type2size[type] == 2;
2604 receiver = argument(0); // type: oop
2605 base = argument(1); // type: oop
2606 offset = argument(2); // type: long
2607 oldval = argument(4); // type: oop, int, or long
2608 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2609 break;
2610 }
2611 case LS_get_add:
2612 case LS_get_set: {
2613 receiver = argument(0); // type: oop
2614 base = argument(1); // type: oop
2615 offset = argument(2); // type: long
2616 oldval = nullptr;
2617 newval = argument(4); // type: oop, int, or long
2618 break;
2619 }
2620 default:
2621 ShouldNotReachHere();
2622 }
2623
2624 // Build field offset expression.
2625 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2626 // to be plain byte offsets, which are also the same as those accepted
2627 // by oopDesc::field_addr.
2628 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2629 // 32-bit machines ignore the high half of long offsets
2630 offset = ConvL2X(offset);
2631 // Save state and restore on bailout
2632 uint old_sp = sp();
2633 SafePointNode* old_map = clone_map();
2634 Node* adr = make_unsafe_address(base, offset,type, false);
2635 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2636
2637 Compile::AliasType* alias_type = C->alias_type(adr_type);
2638 BasicType bt = alias_type->basic_type();
2639 if (bt != T_ILLEGAL &&
2640 (is_reference_type(bt) != (type == T_OBJECT))) {
2641 // Don't intrinsify mismatched object accesses.
2642 set_map(old_map);
2643 set_sp(old_sp);
2644 return false;
2645 }
2646
2647 destruct_map_clone(old_map);
2648
2649 // For CAS, unlike inline_unsafe_access, there seems no point in
2650 // trying to refine types. Just use the coarse types here.
2651 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2652 const Type *value_type = Type::get_const_basic_type(type);
2653
2654 switch (kind) {
2655 case LS_get_set:
2656 case LS_cmp_exchange: {
2657 if (type == T_OBJECT) {
2658 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2659 if (tjp != nullptr) {
2660 value_type = tjp;
2661 }
2662 }
2663 break;
2664 }
2665 case LS_cmp_swap:
2666 case LS_cmp_swap_weak:
2667 case LS_get_add:
2668 break;
2669 default:
2670 ShouldNotReachHere();
2671 }
2672
2673 // Null check receiver.
2674 receiver = null_check(receiver);
2675 if (stopped()) {
2676 return true;
2677 }
2678
2679 int alias_idx = C->get_alias_index(adr_type);
2680
2681 if (is_reference_type(type)) {
2682 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2683
2684 // Transformation of a value which could be null pointer (CastPP #null)
2685 // could be delayed during Parse (for example, in adjust_map_after_if()).
2686 // Execute transformation here to avoid barrier generation in such case.
2687 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2688 newval = _gvn.makecon(TypePtr::NULL_PTR);
2689
2690 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2691 // Refine the value to a null constant, when it is known to be null
2692 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2693 }
2694 }
2695
2696 Node* result = nullptr;
2697 switch (kind) {
2698 case LS_cmp_exchange: {
2699 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2700 oldval, newval, value_type, type, decorators);
2701 break;
2702 }
2703 case LS_cmp_swap_weak:
2704 decorators |= C2_WEAK_CMPXCHG;
2705 case LS_cmp_swap: {
2706 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2707 oldval, newval, value_type, type, decorators);
2708 break;
2709 }
2710 case LS_get_set: {
2711 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2712 newval, value_type, type, decorators);
2713 break;
2714 }
2715 case LS_get_add: {
2716 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2717 newval, value_type, type, decorators);
2718 break;
2719 }
2720 default:
2721 ShouldNotReachHere();
2722 }
2723
2724 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2725 set_result(result);
2726 return true;
2727 }
2728
2729 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2730 // Regardless of form, don't allow previous ld/st to move down,
2731 // then issue acquire, release, or volatile mem_bar.
2732 insert_mem_bar(Op_MemBarCPUOrder);
2733 switch(id) {
2734 case vmIntrinsics::_loadFence:
2735 insert_mem_bar(Op_LoadFence);
2736 return true;
2737 case vmIntrinsics::_storeFence:
2738 insert_mem_bar(Op_StoreFence);
2739 return true;
2740 case vmIntrinsics::_storeStoreFence:
2741 insert_mem_bar(Op_StoreStoreFence);
2742 return true;
2743 case vmIntrinsics::_fullFence:
2744 insert_mem_bar(Op_MemBarVolatile);
2745 return true;
2746 default:
2747 fatal_unexpected_iid(id);
2748 return false;
2749 }
2750 }
2751
2752 bool LibraryCallKit::inline_onspinwait() {
2753 insert_mem_bar(Op_OnSpinWait);
2754 return true;
2755 }
2756
2757 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2758 if (!kls->is_Con()) {
2759 return true;
2760 }
2761 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2762 if (klsptr == nullptr) {
2763 return true;
2764 }
2765 ciInstanceKlass* ik = klsptr->instance_klass();
2766 // don't need a guard for a klass that is already initialized
2767 return !ik->is_initialized();
2768 }
2769
2770 //----------------------------inline_unsafe_writeback0-------------------------
2771 // public native void Unsafe.writeback0(long address)
2772 bool LibraryCallKit::inline_unsafe_writeback0() {
2773 if (!Matcher::has_match_rule(Op_CacheWB)) {
2774 return false;
2775 }
2776 #ifndef PRODUCT
2777 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2778 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2779 ciSignature* sig = callee()->signature();
2780 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2781 #endif
2782 null_check_receiver(); // null-check, then ignore
2783 Node *addr = argument(1);
2784 addr = new CastX2PNode(addr);
2785 addr = _gvn.transform(addr);
2786 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2787 flush = _gvn.transform(flush);
2788 set_memory(flush, TypeRawPtr::BOTTOM);
2789 return true;
2790 }
2791
2792 //----------------------------inline_unsafe_writeback0-------------------------
2793 // public native void Unsafe.writeback0(long address)
2794 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2795 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2796 return false;
2797 }
2798 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2799 return false;
2800 }
2801 #ifndef PRODUCT
2802 assert(Matcher::has_match_rule(Op_CacheWB),
2803 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2804 : "found match rule for CacheWBPostSync but not CacheWB"));
2805
2806 #endif
2807 null_check_receiver(); // null-check, then ignore
2808 Node *sync;
2809 if (is_pre) {
2810 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2811 } else {
2812 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2813 }
2814 sync = _gvn.transform(sync);
2815 set_memory(sync, TypeRawPtr::BOTTOM);
2816 return true;
2817 }
2818
2819 //----------------------------inline_unsafe_allocate---------------------------
2820 // public native Object Unsafe.allocateInstance(Class<?> cls);
2821 bool LibraryCallKit::inline_unsafe_allocate() {
2822
2823 #if INCLUDE_JVMTI
2824 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2825 return false;
2826 }
2827 #endif //INCLUDE_JVMTI
2828
2829 if (callee()->is_static()) return false; // caller must have the capability!
2830
2831 null_check_receiver(); // null-check, then ignore
2832 Node* cls = null_check(argument(1));
2833 if (stopped()) return true;
2834
2835 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2836 kls = null_check(kls);
2837 if (stopped()) return true; // argument was like int.class
2838
2839 #if INCLUDE_JVMTI
2840 // Don't try to access new allocated obj in the intrinsic.
2841 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2842 // Deoptimize and allocate in interpreter instead.
2843 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2844 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2845 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2846 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2847 {
2848 BuildCutout unless(this, tst, PROB_MAX);
2849 uncommon_trap(Deoptimization::Reason_intrinsic,
2850 Deoptimization::Action_make_not_entrant);
2851 }
2852 if (stopped()) {
2853 return true;
2854 }
2855 #endif //INCLUDE_JVMTI
2856
2857 Node* test = nullptr;
2858 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2859 // Note: The argument might still be an illegal value like
2860 // Serializable.class or Object[].class. The runtime will handle it.
2861 // But we must make an explicit check for initialization.
2862 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2863 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2864 // can generate code to load it as unsigned byte.
2865 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2866 Node* bits = intcon(InstanceKlass::fully_initialized);
2867 test = _gvn.transform(new SubINode(inst, bits));
2868 // The 'test' is non-zero if we need to take a slow path.
2869 }
2870
2871 Node* obj = new_instance(kls, test);
2872 set_result(obj);
2873 return true;
2874 }
2875
2876 //------------------------inline_native_time_funcs--------------
2877 // inline code for System.currentTimeMillis() and System.nanoTime()
2878 // these have the same type and signature
2879 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2880 const TypeFunc* tf = OptoRuntime::void_long_Type();
2881 const TypePtr* no_memory_effects = nullptr;
2882 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2883 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2884 #ifdef ASSERT
2885 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2886 assert(value_top == top(), "second value must be top");
2887 #endif
2888 set_result(value);
2889 return true;
2890 }
2891
2892
2893 #if INCLUDE_JVMTI
2894
2895 // When notifications are disabled then just update the VTMS transition bit and return.
2896 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2897 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2898 if (!DoJVMTIVirtualThreadTransitions) {
2899 return true;
2900 }
2901 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2902 IdealKit ideal(this);
2903
2904 Node* ONE = ideal.ConI(1);
2905 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2906 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2907 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2908
2909 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2910 sync_kit(ideal);
2911 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2912 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2913 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2914 ideal.sync_kit(this);
2915 } ideal.else_(); {
2916 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2917 Node* thread = ideal.thread();
2918 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2919 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2920 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2921
2922 sync_kit(ideal);
2923 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2924 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2925
2926 ideal.sync_kit(this);
2927 } ideal.end_if();
2928 final_sync(ideal);
2929
2930 return true;
2931 }
2932
2933 // Always update the temporary VTMS transition bit.
2934 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2935 if (!DoJVMTIVirtualThreadTransitions) {
2936 return true;
2937 }
2938 IdealKit ideal(this);
2939
2940 {
2941 // unconditionally update the temporary VTMS transition bit in current JavaThread
2942 Node* thread = ideal.thread();
2943 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
2944 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2945 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2946
2947 sync_kit(ideal);
2948 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2949 ideal.sync_kit(this);
2950 }
2951 final_sync(ideal);
2952
2953 return true;
2954 }
2955
2956 // Always update the is_disable_suspend bit.
2957 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2958 if (!DoJVMTIVirtualThreadTransitions) {
2959 return true;
2960 }
2961 IdealKit ideal(this);
2962
2963 {
2964 // unconditionally update the is_disable_suspend bit in current JavaThread
2965 Node* thread = ideal.thread();
2966 Node* arg = _gvn.transform(argument(0)); // argument for notification
2967 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2968 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2969
2970 sync_kit(ideal);
2971 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2972 ideal.sync_kit(this);
2973 }
2974 final_sync(ideal);
2975
2976 return true;
2977 }
2978
2979 #endif // INCLUDE_JVMTI
2980
2981 #ifdef JFR_HAVE_INTRINSICS
2982
2983 /**
2984 * if oop->klass != null
2985 * // normal class
2986 * epoch = _epoch_state ? 2 : 1
2987 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2988 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2989 * }
2990 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2991 * else
2992 * // primitive class
2993 * if oop->array_klass != null
2994 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2995 * else
2996 * id = LAST_TYPE_ID + 1 // void class path
2997 * if (!signaled)
2998 * signaled = true
2999 */
3000 bool LibraryCallKit::inline_native_classID() {
3001 Node* cls = argument(0);
3002
3003 IdealKit ideal(this);
3004 #define __ ideal.
3005 IdealVariable result(ideal); __ declarations_done();
3006 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3007 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3008 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3009
3010
3011 __ if_then(kls, BoolTest::ne, null()); {
3012 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3013 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3014
3015 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3016 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3017 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3018 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3019 mask = _gvn.transform(new OrLNode(mask, epoch));
3020 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3021
3022 float unlikely = PROB_UNLIKELY(0.999);
3023 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3024 sync_kit(ideal);
3025 make_runtime_call(RC_LEAF,
3026 OptoRuntime::class_id_load_barrier_Type(),
3027 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3028 "class id load barrier",
3029 TypePtr::BOTTOM,
3030 kls);
3031 ideal.sync_kit(this);
3032 } __ end_if();
3033
3034 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3035 } __ else_(); {
3036 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3037 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3038 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3039 __ if_then(array_kls, BoolTest::ne, null()); {
3040 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3041 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3042 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3043 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3044 } __ else_(); {
3045 // void class case
3046 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3047 } __ end_if();
3048
3049 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3050 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3051 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3052 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3053 } __ end_if();
3054 } __ end_if();
3055
3056 final_sync(ideal);
3057 set_result(ideal.value(result));
3058 #undef __
3059 return true;
3060 }
3061
3062 //------------------------inline_native_jvm_commit------------------
3063 bool LibraryCallKit::inline_native_jvm_commit() {
3064 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3065
3066 // Save input memory and i_o state.
3067 Node* input_memory_state = reset_memory();
3068 set_all_memory(input_memory_state);
3069 Node* input_io_state = i_o();
3070
3071 // TLS.
3072 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3073 // Jfr java buffer.
3074 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3075 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3076 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3077
3078 // Load the current value of the notified field in the JfrThreadLocal.
3079 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3080 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3081
3082 // Test for notification.
3083 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3084 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3085 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3086
3087 // True branch, is notified.
3088 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3089 set_control(is_notified);
3090
3091 // Reset notified state.
3092 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3093
3094 // 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.
3095 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3096 // Convert the machine-word to a long.
3097 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3098
3099 // False branch, not notified.
3100 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3101 set_control(not_notified);
3102 set_all_memory(input_memory_state);
3103
3104 // Arg is the next position as a long.
3105 Node* arg = argument(0);
3106 // Convert long to machine-word.
3107 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3108
3109 // Store the next_position to the underlying jfr java buffer.
3110 Node* commit_memory;
3111 #ifdef _LP64
3112 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3113 #else
3114 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3115 #endif
3116
3117 // 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.
3118 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3119 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3120 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3121
3122 // And flags with lease constant.
3123 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3124
3125 // Branch on lease to conditionalize returning the leased java buffer.
3126 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3127 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3128 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3129
3130 // False branch, not a lease.
3131 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3132
3133 // True branch, is lease.
3134 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3135 set_control(is_lease);
3136
3137 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3138 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3139 OptoRuntime::void_void_Type(),
3140 StubRoutines::jfr_return_lease(),
3141 "return_lease", TypePtr::BOTTOM);
3142 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3143
3144 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3145 record_for_igvn(lease_compare_rgn);
3146 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3147 record_for_igvn(lease_compare_mem);
3148 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3149 record_for_igvn(lease_compare_io);
3150 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3151 record_for_igvn(lease_result_value);
3152
3153 // Update control and phi nodes.
3154 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3155 lease_compare_rgn->init_req(_false_path, not_lease);
3156
3157 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3158 lease_compare_mem->init_req(_false_path, commit_memory);
3159
3160 lease_compare_io->init_req(_true_path, i_o());
3161 lease_compare_io->init_req(_false_path, input_io_state);
3162
3163 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3164 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3165
3166 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3167 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3168 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3169 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3170
3171 // Update control and phi nodes.
3172 result_rgn->init_req(_true_path, is_notified);
3173 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3174
3175 result_mem->init_req(_true_path, notified_reset_memory);
3176 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3177
3178 result_io->init_req(_true_path, input_io_state);
3179 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3180
3181 result_value->init_req(_true_path, current_pos);
3182 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3183
3184 // Set output state.
3185 set_control(_gvn.transform(result_rgn));
3186 set_all_memory(_gvn.transform(result_mem));
3187 set_i_o(_gvn.transform(result_io));
3188 set_result(result_rgn, result_value);
3189 return true;
3190 }
3191
3192 /*
3193 * The intrinsic is a model of this pseudo-code:
3194 *
3195 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3196 * jobject h_event_writer = tl->java_event_writer();
3197 * if (h_event_writer == nullptr) {
3198 * return nullptr;
3199 * }
3200 * oop threadObj = Thread::threadObj();
3201 * oop vthread = java_lang_Thread::vthread(threadObj);
3202 * traceid tid;
3203 * bool excluded;
3204 * if (vthread != threadObj) { // i.e. current thread is virtual
3205 * tid = java_lang_Thread::tid(vthread);
3206 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3207 * excluded = vthread_epoch_raw & excluded_mask;
3208 * if (!excluded) {
3209 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3210 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3211 * if (vthread_epoch != current_epoch) {
3212 * write_checkpoint();
3213 * }
3214 * }
3215 * } else {
3216 * tid = java_lang_Thread::tid(threadObj);
3217 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3218 * excluded = thread_epoch_raw & excluded_mask;
3219 * }
3220 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3221 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3222 * if (tid_in_event_writer != tid) {
3223 * setField(event_writer, "threadID", tid);
3224 * setField(event_writer, "excluded", excluded);
3225 * }
3226 * return event_writer
3227 */
3228 bool LibraryCallKit::inline_native_getEventWriter() {
3229 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3230
3231 // Save input memory and i_o state.
3232 Node* input_memory_state = reset_memory();
3233 set_all_memory(input_memory_state);
3234 Node* input_io_state = i_o();
3235
3236 Node* excluded_mask = _gvn.intcon(32768);
3237 Node* epoch_mask = _gvn.intcon(32767);
3238
3239 // TLS
3240 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3241
3242 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3243 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3244
3245 // Load the eventwriter jobject handle.
3246 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3247
3248 // Null check the jobject handle.
3249 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3250 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3251 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3252
3253 // False path, jobj is null.
3254 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3255
3256 // True path, jobj is not null.
3257 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3258
3259 set_control(jobj_is_not_null);
3260
3261 // Load the threadObj for the CarrierThread.
3262 Node* threadObj = generate_current_thread(tls_ptr);
3263
3264 // Load the vthread.
3265 Node* vthread = generate_virtual_thread(tls_ptr);
3266
3267 // If vthread != threadObj, this is a virtual thread.
3268 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3269 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3270 IfNode* iff_vthread_not_equal_threadObj =
3271 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3272
3273 // False branch, fallback to threadObj.
3274 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3275 set_control(vthread_equal_threadObj);
3276
3277 // Load the tid field from the vthread object.
3278 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3279
3280 // Load the raw epoch value from the threadObj.
3281 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3282 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3283 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3284
3285 // Mask off the excluded information from the epoch.
3286 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3287
3288 // True branch, this is a virtual thread.
3289 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3290 set_control(vthread_not_equal_threadObj);
3291
3292 // Load the tid field from the vthread object.
3293 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3294
3295 // Load the raw epoch value from the vthread.
3296 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3297 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3298 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3299
3300 // Mask off the excluded information from the epoch.
3301 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3302
3303 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3304 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3305 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3306 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3307
3308 // False branch, vthread is excluded, no need to write epoch info.
3309 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3310
3311 // True branch, vthread is included, update epoch info.
3312 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3313 set_control(included);
3314
3315 // Get epoch value.
3316 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3317
3318 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3319 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3320 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3321
3322 // Compare the epoch in the vthread to the current epoch generation.
3323 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3324 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3325 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3326
3327 // False path, epoch is equal, checkpoint information is valid.
3328 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3329
3330 // True path, epoch is not equal, write a checkpoint for the vthread.
3331 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3332
3333 set_control(epoch_is_not_equal);
3334
3335 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3336 // The call also updates the native thread local thread id and the vthread with the current epoch.
3337 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3338 OptoRuntime::jfr_write_checkpoint_Type(),
3339 StubRoutines::jfr_write_checkpoint(),
3340 "write_checkpoint", TypePtr::BOTTOM);
3341 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3342
3343 // vthread epoch != current epoch
3344 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3345 record_for_igvn(epoch_compare_rgn);
3346 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3347 record_for_igvn(epoch_compare_mem);
3348 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3349 record_for_igvn(epoch_compare_io);
3350
3351 // Update control and phi nodes.
3352 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3353 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3354 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3355 epoch_compare_mem->init_req(_false_path, input_memory_state);
3356 epoch_compare_io->init_req(_true_path, i_o());
3357 epoch_compare_io->init_req(_false_path, input_io_state);
3358
3359 // excluded != true
3360 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3361 record_for_igvn(exclude_compare_rgn);
3362 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3363 record_for_igvn(exclude_compare_mem);
3364 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3365 record_for_igvn(exclude_compare_io);
3366
3367 // Update control and phi nodes.
3368 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3369 exclude_compare_rgn->init_req(_false_path, excluded);
3370 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3371 exclude_compare_mem->init_req(_false_path, input_memory_state);
3372 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3373 exclude_compare_io->init_req(_false_path, input_io_state);
3374
3375 // vthread != threadObj
3376 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3377 record_for_igvn(vthread_compare_rgn);
3378 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3379 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3380 record_for_igvn(vthread_compare_io);
3381 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3382 record_for_igvn(tid);
3383 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3384 record_for_igvn(exclusion);
3385
3386 // Update control and phi nodes.
3387 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3388 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3389 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3390 vthread_compare_mem->init_req(_false_path, input_memory_state);
3391 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3392 vthread_compare_io->init_req(_false_path, input_io_state);
3393 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3394 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3395 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3396 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3397
3398 // Update branch state.
3399 set_control(_gvn.transform(vthread_compare_rgn));
3400 set_all_memory(_gvn.transform(vthread_compare_mem));
3401 set_i_o(_gvn.transform(vthread_compare_io));
3402
3403 // Load the event writer oop by dereferencing the jobject handle.
3404 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3405 assert(klass_EventWriter->is_loaded(), "invariant");
3406 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3407 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3408 const TypeOopPtr* const xtype = aklass->as_instance_type();
3409 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3410 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3411
3412 // Load the current thread id from the event writer object.
3413 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3414 // Get the field offset to, conditionally, store an updated tid value later.
3415 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3416 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3417 // Get the field offset to, conditionally, store an updated exclusion value later.
3418 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3419 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3420
3421 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3422 record_for_igvn(event_writer_tid_compare_rgn);
3423 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3424 record_for_igvn(event_writer_tid_compare_mem);
3425 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3426 record_for_igvn(event_writer_tid_compare_io);
3427
3428 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3429 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3430 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3431 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3432
3433 // False path, tids are the same.
3434 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3435
3436 // True path, tid is not equal, need to update the tid in the event writer.
3437 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3438 record_for_igvn(tid_is_not_equal);
3439
3440 // Store the exclusion state to the event writer.
3441 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3442
3443 // Store the tid to the event writer.
3444 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3445
3446 // Update control and phi nodes.
3447 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3448 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3449 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3450 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3451 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3452 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3453
3454 // Result of top level CFG, Memory, IO and Value.
3455 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3456 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3457 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3458 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3459
3460 // Result control.
3461 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3462 result_rgn->init_req(_false_path, jobj_is_null);
3463
3464 // Result memory.
3465 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3466 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3467
3468 // Result IO.
3469 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3470 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3471
3472 // Result value.
3473 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3474 result_value->init_req(_false_path, null()); // return null
3475
3476 // Set output state.
3477 set_control(_gvn.transform(result_rgn));
3478 set_all_memory(_gvn.transform(result_mem));
3479 set_i_o(_gvn.transform(result_io));
3480 set_result(result_rgn, result_value);
3481 return true;
3482 }
3483
3484 /*
3485 * The intrinsic is a model of this pseudo-code:
3486 *
3487 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3488 * if (carrierThread != thread) { // is virtual thread
3489 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3490 * bool excluded = vthread_epoch_raw & excluded_mask;
3491 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3492 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3493 * if (!excluded) {
3494 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3495 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3496 * }
3497 * Atomic::release_store(&tl->_vthread, true);
3498 * return;
3499 * }
3500 * Atomic::release_store(&tl->_vthread, false);
3501 */
3502 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3503 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3504
3505 Node* input_memory_state = reset_memory();
3506 set_all_memory(input_memory_state);
3507
3508 Node* excluded_mask = _gvn.intcon(32768);
3509 Node* epoch_mask = _gvn.intcon(32767);
3510
3511 Node* const carrierThread = generate_current_thread(jt);
3512 // If thread != carrierThread, this is a virtual thread.
3513 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3514 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3515 IfNode* iff_thread_not_equal_carrierThread =
3516 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3517
3518 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3519
3520 // False branch, is carrierThread.
3521 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3522 // Store release
3523 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3524
3525 set_all_memory(input_memory_state);
3526
3527 // True branch, is virtual thread.
3528 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3529 set_control(thread_not_equal_carrierThread);
3530
3531 // Load the raw epoch value from the vthread.
3532 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3533 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3534 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3535
3536 // Mask off the excluded information from the epoch.
3537 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3538
3539 // Load the tid field from the thread.
3540 Node* tid = load_field_from_object(thread, "tid", "J");
3541
3542 // Store the vthread tid to the jfr thread local.
3543 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3544 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3545
3546 // Branch is_excluded to conditionalize updating the epoch .
3547 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3548 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3549 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3550
3551 // True branch, vthread is excluded, no need to write epoch info.
3552 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3553 set_control(excluded);
3554 Node* vthread_is_excluded = _gvn.intcon(1);
3555
3556 // False branch, vthread is included, update epoch info.
3557 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3558 set_control(included);
3559 Node* vthread_is_included = _gvn.intcon(0);
3560
3561 // Get epoch value.
3562 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3563
3564 // Store the vthread epoch to the jfr thread local.
3565 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3566 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3567
3568 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3569 record_for_igvn(excluded_rgn);
3570 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3571 record_for_igvn(excluded_mem);
3572 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3573 record_for_igvn(exclusion);
3574
3575 // Merge the excluded control and memory.
3576 excluded_rgn->init_req(_true_path, excluded);
3577 excluded_rgn->init_req(_false_path, included);
3578 excluded_mem->init_req(_true_path, tid_memory);
3579 excluded_mem->init_req(_false_path, included_memory);
3580 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3581 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3582
3583 // Set intermediate state.
3584 set_control(_gvn.transform(excluded_rgn));
3585 set_all_memory(excluded_mem);
3586
3587 // Store the vthread exclusion state to the jfr thread local.
3588 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3589 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3590
3591 // Store release
3592 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3593
3594 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3595 record_for_igvn(thread_compare_rgn);
3596 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3597 record_for_igvn(thread_compare_mem);
3598 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3599 record_for_igvn(vthread);
3600
3601 // Merge the thread_compare control and memory.
3602 thread_compare_rgn->init_req(_true_path, control());
3603 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3604 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3605 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3606
3607 // Set output state.
3608 set_control(_gvn.transform(thread_compare_rgn));
3609 set_all_memory(_gvn.transform(thread_compare_mem));
3610 }
3611
3612 #endif // JFR_HAVE_INTRINSICS
3613
3614 //------------------------inline_native_currentCarrierThread------------------
3615 bool LibraryCallKit::inline_native_currentCarrierThread() {
3616 Node* junk = nullptr;
3617 set_result(generate_current_thread(junk));
3618 return true;
3619 }
3620
3621 //------------------------inline_native_currentThread------------------
3622 bool LibraryCallKit::inline_native_currentThread() {
3623 Node* junk = nullptr;
3624 set_result(generate_virtual_thread(junk));
3625 return true;
3626 }
3627
3628 //------------------------inline_native_setVthread------------------
3629 bool LibraryCallKit::inline_native_setCurrentThread() {
3630 assert(C->method()->changes_current_thread(),
3631 "method changes current Thread but is not annotated ChangesCurrentThread");
3632 Node* arr = argument(1);
3633 Node* thread = _gvn.transform(new ThreadLocalNode());
3634 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3635 Node* thread_obj_handle
3636 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3637 thread_obj_handle = _gvn.transform(thread_obj_handle);
3638 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3639 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3640 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3641 return true;
3642 }
3643
3644 const Type* LibraryCallKit::scopedValueCache_type() {
3645 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3646 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3647 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3648
3649 // Because we create the scopedValue cache lazily we have to make the
3650 // type of the result BotPTR.
3651 bool xk = etype->klass_is_exact();
3652 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3653 return objects_type;
3654 }
3655
3656 Node* LibraryCallKit::scopedValueCache_helper() {
3657 Node* thread = _gvn.transform(new ThreadLocalNode());
3658 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3659 // We cannot use immutable_memory() because we might flip onto a
3660 // different carrier thread, at which point we'll need to use that
3661 // carrier thread's cache.
3662 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3663 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3664 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3665 }
3666
3667 //------------------------inline_native_scopedValueCache------------------
3668 bool LibraryCallKit::inline_native_scopedValueCache() {
3669 Node* cache_obj_handle = scopedValueCache_helper();
3670 const Type* objects_type = scopedValueCache_type();
3671 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3672
3673 return true;
3674 }
3675
3676 //------------------------inline_native_setScopedValueCache------------------
3677 bool LibraryCallKit::inline_native_setScopedValueCache() {
3678 Node* arr = argument(0);
3679 Node* cache_obj_handle = scopedValueCache_helper();
3680 const Type* objects_type = scopedValueCache_type();
3681
3682 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3683 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3684
3685 return true;
3686 }
3687
3688 //---------------------------load_mirror_from_klass----------------------------
3689 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3690 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3691 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3692 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3693 // mirror = ((OopHandle)mirror)->resolve();
3694 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3695 }
3696
3697 //-----------------------load_klass_from_mirror_common-------------------------
3698 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3699 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3700 // and branch to the given path on the region.
3701 // If never_see_null, take an uncommon trap on null, so we can optimistically
3702 // compile for the non-null case.
3703 // If the region is null, force never_see_null = true.
3704 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3705 bool never_see_null,
3706 RegionNode* region,
3707 int null_path,
3708 int offset) {
3709 if (region == nullptr) never_see_null = true;
3710 Node* p = basic_plus_adr(mirror, offset);
3711 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3712 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3713 Node* null_ctl = top();
3714 kls = null_check_oop(kls, &null_ctl, never_see_null);
3715 if (region != nullptr) {
3716 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3717 region->init_req(null_path, null_ctl);
3718 } else {
3719 assert(null_ctl == top(), "no loose ends");
3720 }
3721 return kls;
3722 }
3723
3724 //--------------------(inline_native_Class_query helpers)---------------------
3725 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3726 // Fall through if (mods & mask) == bits, take the guard otherwise.
3727 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3728 // Branch around if the given klass has the given modifier bit set.
3729 // Like generate_guard, adds a new path onto the region.
3730 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3731 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3732 Node* mask = intcon(modifier_mask);
3733 Node* bits = intcon(modifier_bits);
3734 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3735 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3736 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3737 return generate_fair_guard(bol, region);
3738 }
3739 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3740 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3741 }
3742 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3743 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3744 }
3745
3746 //-------------------------inline_native_Class_query-------------------
3747 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3748 const Type* return_type = TypeInt::BOOL;
3749 Node* prim_return_value = top(); // what happens if it's a primitive class?
3750 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3751 bool expect_prim = false; // most of these guys expect to work on refs
3752
3753 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3754
3755 Node* mirror = argument(0);
3756 Node* obj = top();
3757
3758 switch (id) {
3759 case vmIntrinsics::_isInstance:
3760 // nothing is an instance of a primitive type
3761 prim_return_value = intcon(0);
3762 obj = argument(1);
3763 break;
3764 case vmIntrinsics::_getModifiers:
3765 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3766 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3767 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3768 break;
3769 case vmIntrinsics::_isInterface:
3770 prim_return_value = intcon(0);
3771 break;
3772 case vmIntrinsics::_isArray:
3773 prim_return_value = intcon(0);
3774 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3775 break;
3776 case vmIntrinsics::_isPrimitive:
3777 prim_return_value = intcon(1);
3778 expect_prim = true; // obviously
3779 break;
3780 case vmIntrinsics::_isHidden:
3781 prim_return_value = intcon(0);
3782 break;
3783 case vmIntrinsics::_getSuperclass:
3784 prim_return_value = null();
3785 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3786 break;
3787 case vmIntrinsics::_getClassAccessFlags:
3788 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3789 return_type = TypeInt::INT; // not bool! 6297094
3790 break;
3791 default:
3792 fatal_unexpected_iid(id);
3793 break;
3794 }
3795
3796 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3797 if (mirror_con == nullptr) return false; // cannot happen?
3798
3799 #ifndef PRODUCT
3800 if (C->print_intrinsics() || C->print_inlining()) {
3801 ciType* k = mirror_con->java_mirror_type();
3802 if (k) {
3803 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3804 k->print_name();
3805 tty->cr();
3806 }
3807 }
3808 #endif
3809
3810 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3811 RegionNode* region = new RegionNode(PATH_LIMIT);
3812 record_for_igvn(region);
3813 PhiNode* phi = new PhiNode(region, return_type);
3814
3815 // The mirror will never be null of Reflection.getClassAccessFlags, however
3816 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3817 // if it is. See bug 4774291.
3818
3819 // For Reflection.getClassAccessFlags(), the null check occurs in
3820 // the wrong place; see inline_unsafe_access(), above, for a similar
3821 // situation.
3822 mirror = null_check(mirror);
3823 // If mirror or obj is dead, only null-path is taken.
3824 if (stopped()) return true;
3825
3826 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3827
3828 // Now load the mirror's klass metaobject, and null-check it.
3829 // Side-effects region with the control path if the klass is null.
3830 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3831 // If kls is null, we have a primitive mirror.
3832 phi->init_req(_prim_path, prim_return_value);
3833 if (stopped()) { set_result(region, phi); return true; }
3834 bool safe_for_replace = (region->in(_prim_path) == top());
3835
3836 Node* p; // handy temp
3837 Node* null_ctl;
3838
3839 // Now that we have the non-null klass, we can perform the real query.
3840 // For constant classes, the query will constant-fold in LoadNode::Value.
3841 Node* query_value = top();
3842 switch (id) {
3843 case vmIntrinsics::_isInstance:
3844 // nothing is an instance of a primitive type
3845 query_value = gen_instanceof(obj, kls, safe_for_replace);
3846 break;
3847
3848 case vmIntrinsics::_getModifiers:
3849 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3850 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3851 break;
3852
3853 case vmIntrinsics::_isInterface:
3854 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3855 if (generate_interface_guard(kls, region) != nullptr)
3856 // A guard was added. If the guard is taken, it was an interface.
3857 phi->add_req(intcon(1));
3858 // If we fall through, it's a plain class.
3859 query_value = intcon(0);
3860 break;
3861
3862 case vmIntrinsics::_isArray:
3863 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3864 if (generate_array_guard(kls, region) != nullptr)
3865 // A guard was added. If the guard is taken, it was an array.
3866 phi->add_req(intcon(1));
3867 // If we fall through, it's a plain class.
3868 query_value = intcon(0);
3869 break;
3870
3871 case vmIntrinsics::_isPrimitive:
3872 query_value = intcon(0); // "normal" path produces false
3873 break;
3874
3875 case vmIntrinsics::_isHidden:
3876 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3877 if (generate_hidden_class_guard(kls, region) != nullptr)
3878 // A guard was added. If the guard is taken, it was an hidden class.
3879 phi->add_req(intcon(1));
3880 // If we fall through, it's a plain class.
3881 query_value = intcon(0);
3882 break;
3883
3884
3885 case vmIntrinsics::_getSuperclass:
3886 // The rules here are somewhat unfortunate, but we can still do better
3887 // with random logic than with a JNI call.
3888 // Interfaces store null or Object as _super, but must report null.
3889 // Arrays store an intermediate super as _super, but must report Object.
3890 // Other types can report the actual _super.
3891 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3892 if (generate_interface_guard(kls, region) != nullptr)
3893 // A guard was added. If the guard is taken, it was an interface.
3894 phi->add_req(null());
3895 if (generate_array_guard(kls, region) != nullptr)
3896 // A guard was added. If the guard is taken, it was an array.
3897 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3898 // If we fall through, it's a plain class. Get its _super.
3899 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3900 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3901 null_ctl = top();
3902 kls = null_check_oop(kls, &null_ctl);
3903 if (null_ctl != top()) {
3904 // If the guard is taken, Object.superClass is null (both klass and mirror).
3905 region->add_req(null_ctl);
3906 phi ->add_req(null());
3907 }
3908 if (!stopped()) {
3909 query_value = load_mirror_from_klass(kls);
3910 }
3911 break;
3912
3913 case vmIntrinsics::_getClassAccessFlags:
3914 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3915 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3916 break;
3917
3918 default:
3919 fatal_unexpected_iid(id);
3920 break;
3921 }
3922
3923 // Fall-through is the normal case of a query to a real class.
3924 phi->init_req(1, query_value);
3925 region->init_req(1, control());
3926
3927 C->set_has_split_ifs(true); // Has chance for split-if optimization
3928 set_result(region, phi);
3929 return true;
3930 }
3931
3932 //-------------------------inline_Class_cast-------------------
3933 bool LibraryCallKit::inline_Class_cast() {
3934 Node* mirror = argument(0); // Class
3935 Node* obj = argument(1);
3936 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3937 if (mirror_con == nullptr) {
3938 return false; // dead path (mirror->is_top()).
3939 }
3940 if (obj == nullptr || obj->is_top()) {
3941 return false; // dead path
3942 }
3943 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3944
3945 // First, see if Class.cast() can be folded statically.
3946 // java_mirror_type() returns non-null for compile-time Class constants.
3947 ciType* tm = mirror_con->java_mirror_type();
3948 if (tm != nullptr && tm->is_klass() &&
3949 tp != nullptr) {
3950 if (!tp->is_loaded()) {
3951 // Don't use intrinsic when class is not loaded.
3952 return false;
3953 } else {
3954 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3955 if (static_res == Compile::SSC_always_true) {
3956 // isInstance() is true - fold the code.
3957 set_result(obj);
3958 return true;
3959 } else if (static_res == Compile::SSC_always_false) {
3960 // Don't use intrinsic, have to throw ClassCastException.
3961 // If the reference is null, the non-intrinsic bytecode will
3962 // be optimized appropriately.
3963 return false;
3964 }
3965 }
3966 }
3967
3968 // Bailout intrinsic and do normal inlining if exception path is frequent.
3969 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3970 return false;
3971 }
3972
3973 // Generate dynamic checks.
3974 // Class.cast() is java implementation of _checkcast bytecode.
3975 // Do checkcast (Parse::do_checkcast()) optimizations here.
3976
3977 mirror = null_check(mirror);
3978 // If mirror is dead, only null-path is taken.
3979 if (stopped()) {
3980 return true;
3981 }
3982
3983 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
3984 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3985 RegionNode* region = new RegionNode(PATH_LIMIT);
3986 record_for_igvn(region);
3987
3988 // Now load the mirror's klass metaobject, and null-check it.
3989 // If kls is null, we have a primitive mirror and
3990 // nothing is an instance of a primitive type.
3991 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3992
3993 Node* res = top();
3994 if (!stopped()) {
3995 Node* bad_type_ctrl = top();
3996 // Do checkcast optimizations.
3997 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3998 region->init_req(_bad_type_path, bad_type_ctrl);
3999 }
4000 if (region->in(_prim_path) != top() ||
4001 region->in(_bad_type_path) != top()) {
4002 // Let Interpreter throw ClassCastException.
4003 PreserveJVMState pjvms(this);
4004 set_control(_gvn.transform(region));
4005 uncommon_trap(Deoptimization::Reason_intrinsic,
4006 Deoptimization::Action_maybe_recompile);
4007 }
4008 if (!stopped()) {
4009 set_result(res);
4010 }
4011 return true;
4012 }
4013
4014
4015 //--------------------------inline_native_subtype_check------------------------
4016 // This intrinsic takes the JNI calls out of the heart of
4017 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4018 bool LibraryCallKit::inline_native_subtype_check() {
4019 // Pull both arguments off the stack.
4020 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4021 args[0] = argument(0);
4022 args[1] = argument(1);
4023 Node* klasses[2]; // corresponding Klasses: superk, subk
4024 klasses[0] = klasses[1] = top();
4025
4026 enum {
4027 // A full decision tree on {superc is prim, subc is prim}:
4028 _prim_0_path = 1, // {P,N} => false
4029 // {P,P} & superc!=subc => false
4030 _prim_same_path, // {P,P} & superc==subc => true
4031 _prim_1_path, // {N,P} => false
4032 _ref_subtype_path, // {N,N} & subtype check wins => true
4033 _both_ref_path, // {N,N} & subtype check loses => false
4034 PATH_LIMIT
4035 };
4036
4037 RegionNode* region = new RegionNode(PATH_LIMIT);
4038 Node* phi = new PhiNode(region, TypeInt::BOOL);
4039 record_for_igvn(region);
4040
4041 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4042 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4043 int class_klass_offset = java_lang_Class::klass_offset();
4044
4045 // First null-check both mirrors and load each mirror's klass metaobject.
4046 int which_arg;
4047 for (which_arg = 0; which_arg <= 1; which_arg++) {
4048 Node* arg = args[which_arg];
4049 arg = null_check(arg);
4050 if (stopped()) break;
4051 args[which_arg] = arg;
4052
4053 Node* p = basic_plus_adr(arg, class_klass_offset);
4054 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4055 klasses[which_arg] = _gvn.transform(kls);
4056 }
4057
4058 // Having loaded both klasses, test each for null.
4059 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4060 for (which_arg = 0; which_arg <= 1; which_arg++) {
4061 Node* kls = klasses[which_arg];
4062 Node* null_ctl = top();
4063 kls = null_check_oop(kls, &null_ctl, never_see_null);
4064 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4065 region->init_req(prim_path, null_ctl);
4066 if (stopped()) break;
4067 klasses[which_arg] = kls;
4068 }
4069
4070 if (!stopped()) {
4071 // now we have two reference types, in klasses[0..1]
4072 Node* subk = klasses[1]; // the argument to isAssignableFrom
4073 Node* superk = klasses[0]; // the receiver
4074 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4075 // now we have a successful reference subtype check
4076 region->set_req(_ref_subtype_path, control());
4077 }
4078
4079 // If both operands are primitive (both klasses null), then
4080 // we must return true when they are identical primitives.
4081 // It is convenient to test this after the first null klass check.
4082 set_control(region->in(_prim_0_path)); // go back to first null check
4083 if (!stopped()) {
4084 // Since superc is primitive, make a guard for the superc==subc case.
4085 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4086 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4087 generate_guard(bol_eq, region, PROB_FAIR);
4088 if (region->req() == PATH_LIMIT+1) {
4089 // A guard was added. If the added guard is taken, superc==subc.
4090 region->swap_edges(PATH_LIMIT, _prim_same_path);
4091 region->del_req(PATH_LIMIT);
4092 }
4093 region->set_req(_prim_0_path, control()); // Not equal after all.
4094 }
4095
4096 // these are the only paths that produce 'true':
4097 phi->set_req(_prim_same_path, intcon(1));
4098 phi->set_req(_ref_subtype_path, intcon(1));
4099
4100 // pull together the cases:
4101 assert(region->req() == PATH_LIMIT, "sane region");
4102 for (uint i = 1; i < region->req(); i++) {
4103 Node* ctl = region->in(i);
4104 if (ctl == nullptr || ctl == top()) {
4105 region->set_req(i, top());
4106 phi ->set_req(i, top());
4107 } else if (phi->in(i) == nullptr) {
4108 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4109 }
4110 }
4111
4112 set_control(_gvn.transform(region));
4113 set_result(_gvn.transform(phi));
4114 return true;
4115 }
4116
4117 //---------------------generate_array_guard_common------------------------
4118 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4119 bool obj_array, bool not_array) {
4120
4121 if (stopped()) {
4122 return nullptr;
4123 }
4124
4125 // If obj_array/non_array==false/false:
4126 // Branch around if the given klass is in fact an array (either obj or prim).
4127 // If obj_array/non_array==false/true:
4128 // Branch around if the given klass is not an array klass of any kind.
4129 // If obj_array/non_array==true/true:
4130 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4131 // If obj_array/non_array==true/false:
4132 // Branch around if the kls is an oop array (Object[] or subtype)
4133 //
4134 // Like generate_guard, adds a new path onto the region.
4135 jint layout_con = 0;
4136 Node* layout_val = get_layout_helper(kls, layout_con);
4137 if (layout_val == nullptr) {
4138 bool query = (obj_array
4139 ? Klass::layout_helper_is_objArray(layout_con)
4140 : Klass::layout_helper_is_array(layout_con));
4141 if (query == not_array) {
4142 return nullptr; // never a branch
4143 } else { // always a branch
4144 Node* always_branch = control();
4145 if (region != nullptr)
4146 region->add_req(always_branch);
4147 set_control(top());
4148 return always_branch;
4149 }
4150 }
4151 // Now test the correct condition.
4152 jint nval = (obj_array
4153 ? (jint)(Klass::_lh_array_tag_type_value
4154 << Klass::_lh_array_tag_shift)
4155 : Klass::_lh_neutral_value);
4156 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4157 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4158 // invert the test if we are looking for a non-array
4159 if (not_array) btest = BoolTest(btest).negate();
4160 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4161 return generate_fair_guard(bol, region);
4162 }
4163
4164
4165 //-----------------------inline_native_newArray--------------------------
4166 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4167 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4168 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4169 Node* mirror;
4170 Node* count_val;
4171 if (uninitialized) {
4172 null_check_receiver();
4173 mirror = argument(1);
4174 count_val = argument(2);
4175 } else {
4176 mirror = argument(0);
4177 count_val = argument(1);
4178 }
4179
4180 mirror = null_check(mirror);
4181 // If mirror or obj is dead, only null-path is taken.
4182 if (stopped()) return true;
4183
4184 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4185 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4186 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4187 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4188 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4189
4190 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4191 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4192 result_reg, _slow_path);
4193 Node* normal_ctl = control();
4194 Node* no_array_ctl = result_reg->in(_slow_path);
4195
4196 // Generate code for the slow case. We make a call to newArray().
4197 set_control(no_array_ctl);
4198 if (!stopped()) {
4199 // Either the input type is void.class, or else the
4200 // array klass has not yet been cached. Either the
4201 // ensuing call will throw an exception, or else it
4202 // will cache the array klass for next time.
4203 PreserveJVMState pjvms(this);
4204 CallJavaNode* slow_call = nullptr;
4205 if (uninitialized) {
4206 // Generate optimized virtual call (holder class 'Unsafe' is final)
4207 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4208 } else {
4209 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4210 }
4211 Node* slow_result = set_results_for_java_call(slow_call);
4212 // this->control() comes from set_results_for_java_call
4213 result_reg->set_req(_slow_path, control());
4214 result_val->set_req(_slow_path, slow_result);
4215 result_io ->set_req(_slow_path, i_o());
4216 result_mem->set_req(_slow_path, reset_memory());
4217 }
4218
4219 set_control(normal_ctl);
4220 if (!stopped()) {
4221 // Normal case: The array type has been cached in the java.lang.Class.
4222 // The following call works fine even if the array type is polymorphic.
4223 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4224 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4225 result_reg->init_req(_normal_path, control());
4226 result_val->init_req(_normal_path, obj);
4227 result_io ->init_req(_normal_path, i_o());
4228 result_mem->init_req(_normal_path, reset_memory());
4229
4230 if (uninitialized) {
4231 // Mark the allocation so that zeroing is skipped
4232 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4233 alloc->maybe_set_complete(&_gvn);
4234 }
4235 }
4236
4237 // Return the combined state.
4238 set_i_o( _gvn.transform(result_io) );
4239 set_all_memory( _gvn.transform(result_mem));
4240
4241 C->set_has_split_ifs(true); // Has chance for split-if optimization
4242 set_result(result_reg, result_val);
4243 return true;
4244 }
4245
4246 //----------------------inline_native_getLength--------------------------
4247 // public static native int java.lang.reflect.Array.getLength(Object array);
4248 bool LibraryCallKit::inline_native_getLength() {
4249 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4250
4251 Node* array = null_check(argument(0));
4252 // If array is dead, only null-path is taken.
4253 if (stopped()) return true;
4254
4255 // Deoptimize if it is a non-array.
4256 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4257
4258 if (non_array != nullptr) {
4259 PreserveJVMState pjvms(this);
4260 set_control(non_array);
4261 uncommon_trap(Deoptimization::Reason_intrinsic,
4262 Deoptimization::Action_maybe_recompile);
4263 }
4264
4265 // If control is dead, only non-array-path is taken.
4266 if (stopped()) return true;
4267
4268 // The works fine even if the array type is polymorphic.
4269 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4270 Node* result = load_array_length(array);
4271
4272 C->set_has_split_ifs(true); // Has chance for split-if optimization
4273 set_result(result);
4274 return true;
4275 }
4276
4277 //------------------------inline_array_copyOf----------------------------
4278 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4279 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4280 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4281 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4282
4283 // Get the arguments.
4284 Node* original = argument(0);
4285 Node* start = is_copyOfRange? argument(1): intcon(0);
4286 Node* end = is_copyOfRange? argument(2): argument(1);
4287 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4288
4289 Node* newcopy = nullptr;
4290
4291 // Set the original stack and the reexecute bit for the interpreter to reexecute
4292 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4293 { PreserveReexecuteState preexecs(this);
4294 jvms()->set_should_reexecute(true);
4295
4296 array_type_mirror = null_check(array_type_mirror);
4297 original = null_check(original);
4298
4299 // Check if a null path was taken unconditionally.
4300 if (stopped()) return true;
4301
4302 Node* orig_length = load_array_length(original);
4303
4304 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4305 klass_node = null_check(klass_node);
4306
4307 RegionNode* bailout = new RegionNode(1);
4308 record_for_igvn(bailout);
4309
4310 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4311 // Bail out if that is so.
4312 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4313 if (not_objArray != nullptr) {
4314 // Improve the klass node's type from the new optimistic assumption:
4315 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4316 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4317 Node* cast = new CastPPNode(control(), klass_node, akls);
4318 klass_node = _gvn.transform(cast);
4319 }
4320
4321 // Bail out if either start or end is negative.
4322 generate_negative_guard(start, bailout, &start);
4323 generate_negative_guard(end, bailout, &end);
4324
4325 Node* length = end;
4326 if (_gvn.type(start) != TypeInt::ZERO) {
4327 length = _gvn.transform(new SubINode(end, start));
4328 }
4329
4330 // Bail out if length is negative (i.e., if start > end).
4331 // Without this the new_array would throw
4332 // NegativeArraySizeException but IllegalArgumentException is what
4333 // should be thrown
4334 generate_negative_guard(length, bailout, &length);
4335
4336 // Bail out if start is larger than the original length
4337 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4338 generate_negative_guard(orig_tail, bailout, &orig_tail);
4339
4340 if (bailout->req() > 1) {
4341 PreserveJVMState pjvms(this);
4342 set_control(_gvn.transform(bailout));
4343 uncommon_trap(Deoptimization::Reason_intrinsic,
4344 Deoptimization::Action_maybe_recompile);
4345 }
4346
4347 if (!stopped()) {
4348 // How many elements will we copy from the original?
4349 // The answer is MinI(orig_tail, length).
4350 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4351
4352 // Generate a direct call to the right arraycopy function(s).
4353 // We know the copy is disjoint but we might not know if the
4354 // oop stores need checking.
4355 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4356 // This will fail a store-check if x contains any non-nulls.
4357
4358 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4359 // loads/stores but it is legal only if we're sure the
4360 // Arrays.copyOf would succeed. So we need all input arguments
4361 // to the copyOf to be validated, including that the copy to the
4362 // new array won't trigger an ArrayStoreException. That subtype
4363 // check can be optimized if we know something on the type of
4364 // the input array from type speculation.
4365 if (_gvn.type(klass_node)->singleton()) {
4366 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4367 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4368
4369 int test = C->static_subtype_check(superk, subk);
4370 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4371 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4372 if (t_original->speculative_type() != nullptr) {
4373 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4374 }
4375 }
4376 }
4377
4378 bool validated = false;
4379 // Reason_class_check rather than Reason_intrinsic because we
4380 // want to intrinsify even if this traps.
4381 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4382 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4383
4384 if (not_subtype_ctrl != top()) {
4385 PreserveJVMState pjvms(this);
4386 set_control(not_subtype_ctrl);
4387 uncommon_trap(Deoptimization::Reason_class_check,
4388 Deoptimization::Action_make_not_entrant);
4389 assert(stopped(), "Should be stopped");
4390 }
4391 validated = true;
4392 }
4393
4394 if (!stopped()) {
4395 newcopy = new_array(klass_node, length, 0); // no arguments to push
4396
4397 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4398 load_object_klass(original), klass_node);
4399 if (!is_copyOfRange) {
4400 ac->set_copyof(validated);
4401 } else {
4402 ac->set_copyofrange(validated);
4403 }
4404 Node* n = _gvn.transform(ac);
4405 if (n == ac) {
4406 ac->connect_outputs(this);
4407 } else {
4408 assert(validated, "shouldn't transform if all arguments not validated");
4409 set_all_memory(n);
4410 }
4411 }
4412 }
4413 } // original reexecute is set back here
4414
4415 C->set_has_split_ifs(true); // Has chance for split-if optimization
4416 if (!stopped()) {
4417 set_result(newcopy);
4418 }
4419 return true;
4420 }
4421
4422
4423 //----------------------generate_virtual_guard---------------------------
4424 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4425 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4426 RegionNode* slow_region) {
4427 ciMethod* method = callee();
4428 int vtable_index = method->vtable_index();
4429 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4430 "bad index %d", vtable_index);
4431 // Get the Method* out of the appropriate vtable entry.
4432 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4433 vtable_index*vtableEntry::size_in_bytes() +
4434 in_bytes(vtableEntry::method_offset());
4435 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4436 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4437
4438 // Compare the target method with the expected method (e.g., Object.hashCode).
4439 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4440
4441 Node* native_call = makecon(native_call_addr);
4442 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4443 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4444
4445 return generate_slow_guard(test_native, slow_region);
4446 }
4447
4448 //-----------------------generate_method_call----------------------------
4449 // Use generate_method_call to make a slow-call to the real
4450 // method if the fast path fails. An alternative would be to
4451 // use a stub like OptoRuntime::slow_arraycopy_Java.
4452 // This only works for expanding the current library call,
4453 // not another intrinsic. (E.g., don't use this for making an
4454 // arraycopy call inside of the copyOf intrinsic.)
4455 CallJavaNode*
4456 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4457 // When compiling the intrinsic method itself, do not use this technique.
4458 guarantee(callee() != C->method(), "cannot make slow-call to self");
4459
4460 ciMethod* method = callee();
4461 // ensure the JVMS we have will be correct for this call
4462 guarantee(method_id == method->intrinsic_id(), "must match");
4463
4464 const TypeFunc* tf = TypeFunc::make(method);
4465 if (res_not_null) {
4466 assert(tf->return_type() == T_OBJECT, "");
4467 const TypeTuple* range = tf->range();
4468 const Type** fields = TypeTuple::fields(range->cnt());
4469 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4470 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4471 tf = TypeFunc::make(tf->domain(), new_range);
4472 }
4473 CallJavaNode* slow_call;
4474 if (is_static) {
4475 assert(!is_virtual, "");
4476 slow_call = new CallStaticJavaNode(C, tf,
4477 SharedRuntime::get_resolve_static_call_stub(), method);
4478 } else if (is_virtual) {
4479 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4480 int vtable_index = Method::invalid_vtable_index;
4481 if (UseInlineCaches) {
4482 // Suppress the vtable call
4483 } else {
4484 // hashCode and clone are not a miranda methods,
4485 // so the vtable index is fixed.
4486 // No need to use the linkResolver to get it.
4487 vtable_index = method->vtable_index();
4488 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4489 "bad index %d", vtable_index);
4490 }
4491 slow_call = new CallDynamicJavaNode(tf,
4492 SharedRuntime::get_resolve_virtual_call_stub(),
4493 method, vtable_index);
4494 } else { // neither virtual nor static: opt_virtual
4495 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4496 slow_call = new CallStaticJavaNode(C, tf,
4497 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4498 slow_call->set_optimized_virtual(true);
4499 }
4500 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4501 // To be able to issue a direct call (optimized virtual or virtual)
4502 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4503 // about the method being invoked should be attached to the call site to
4504 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4505 slow_call->set_override_symbolic_info(true);
4506 }
4507 set_arguments_for_java_call(slow_call);
4508 set_edges_for_java_call(slow_call);
4509 return slow_call;
4510 }
4511
4512
4513 /**
4514 * Build special case code for calls to hashCode on an object. This call may
4515 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4516 * slightly different code.
4517 */
4518 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4519 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4520 assert(!(is_virtual && is_static), "either virtual, special, or static");
4521
4522 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4523
4524 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4525 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4526 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4527 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4528 Node* obj = nullptr;
4529 if (!is_static) {
4530 // Check for hashing null object
4531 obj = null_check_receiver();
4532 if (stopped()) return true; // unconditionally null
4533 result_reg->init_req(_null_path, top());
4534 result_val->init_req(_null_path, top());
4535 } else {
4536 // Do a null check, and return zero if null.
4537 // System.identityHashCode(null) == 0
4538 obj = argument(0);
4539 Node* null_ctl = top();
4540 obj = null_check_oop(obj, &null_ctl);
4541 result_reg->init_req(_null_path, null_ctl);
4542 result_val->init_req(_null_path, _gvn.intcon(0));
4543 }
4544
4545 // Unconditionally null? Then return right away.
4546 if (stopped()) {
4547 set_control( result_reg->in(_null_path));
4548 if (!stopped())
4549 set_result(result_val->in(_null_path));
4550 return true;
4551 }
4552
4553 // We only go to the fast case code if we pass a number of guards. The
4554 // paths which do not pass are accumulated in the slow_region.
4555 RegionNode* slow_region = new RegionNode(1);
4556 record_for_igvn(slow_region);
4557
4558 // If this is a virtual call, we generate a funny guard. We pull out
4559 // the vtable entry corresponding to hashCode() from the target object.
4560 // If the target method which we are calling happens to be the native
4561 // Object hashCode() method, we pass the guard. We do not need this
4562 // guard for non-virtual calls -- the caller is known to be the native
4563 // Object hashCode().
4564 if (is_virtual) {
4565 // After null check, get the object's klass.
4566 Node* obj_klass = load_object_klass(obj);
4567 generate_virtual_guard(obj_klass, slow_region);
4568 }
4569
4570 // Get the header out of the object, use LoadMarkNode when available
4571 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4572 // The control of the load must be null. Otherwise, the load can move before
4573 // the null check after castPP removal.
4574 Node* no_ctrl = nullptr;
4575 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4576
4577 // Test the header to see if it is safe to read w.r.t. locking.
4578 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4579 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4580 if (LockingMode == LM_LIGHTWEIGHT) {
4581 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4582 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4583 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4584
4585 generate_slow_guard(test_monitor, slow_region);
4586 } else {
4587 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4588 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4589 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4590
4591 generate_slow_guard(test_not_unlocked, slow_region);
4592 }
4593
4594 // Get the hash value and check to see that it has been properly assigned.
4595 // We depend on hash_mask being at most 32 bits and avoid the use of
4596 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4597 // vm: see markWord.hpp.
4598 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4599 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4600 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4601 // This hack lets the hash bits live anywhere in the mark object now, as long
4602 // as the shift drops the relevant bits into the low 32 bits. Note that
4603 // Java spec says that HashCode is an int so there's no point in capturing
4604 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4605 hshifted_header = ConvX2I(hshifted_header);
4606 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4607
4608 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4609 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4610 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4611
4612 generate_slow_guard(test_assigned, slow_region);
4613
4614 Node* init_mem = reset_memory();
4615 // fill in the rest of the null path:
4616 result_io ->init_req(_null_path, i_o());
4617 result_mem->init_req(_null_path, init_mem);
4618
4619 result_val->init_req(_fast_path, hash_val);
4620 result_reg->init_req(_fast_path, control());
4621 result_io ->init_req(_fast_path, i_o());
4622 result_mem->init_req(_fast_path, init_mem);
4623
4624 // Generate code for the slow case. We make a call to hashCode().
4625 set_control(_gvn.transform(slow_region));
4626 if (!stopped()) {
4627 // No need for PreserveJVMState, because we're using up the present state.
4628 set_all_memory(init_mem);
4629 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4630 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4631 Node* slow_result = set_results_for_java_call(slow_call);
4632 // this->control() comes from set_results_for_java_call
4633 result_reg->init_req(_slow_path, control());
4634 result_val->init_req(_slow_path, slow_result);
4635 result_io ->set_req(_slow_path, i_o());
4636 result_mem ->set_req(_slow_path, reset_memory());
4637 }
4638
4639 // Return the combined state.
4640 set_i_o( _gvn.transform(result_io) );
4641 set_all_memory( _gvn.transform(result_mem));
4642
4643 set_result(result_reg, result_val);
4644 return true;
4645 }
4646
4647 //---------------------------inline_native_getClass----------------------------
4648 // public final native Class<?> java.lang.Object.getClass();
4649 //
4650 // Build special case code for calls to getClass on an object.
4651 bool LibraryCallKit::inline_native_getClass() {
4652 Node* obj = null_check_receiver();
4653 if (stopped()) return true;
4654 set_result(load_mirror_from_klass(load_object_klass(obj)));
4655 return true;
4656 }
4657
4658 //-----------------inline_native_Reflection_getCallerClass---------------------
4659 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4660 //
4661 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4662 //
4663 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4664 // in that it must skip particular security frames and checks for
4665 // caller sensitive methods.
4666 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4667 #ifndef PRODUCT
4668 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4669 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4670 }
4671 #endif
4672
4673 if (!jvms()->has_method()) {
4674 #ifndef PRODUCT
4675 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4676 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4677 }
4678 #endif
4679 return false;
4680 }
4681
4682 // Walk back up the JVM state to find the caller at the required
4683 // depth.
4684 JVMState* caller_jvms = jvms();
4685
4686 // Cf. JVM_GetCallerClass
4687 // NOTE: Start the loop at depth 1 because the current JVM state does
4688 // not include the Reflection.getCallerClass() frame.
4689 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4690 ciMethod* m = caller_jvms->method();
4691 switch (n) {
4692 case 0:
4693 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4694 break;
4695 case 1:
4696 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4697 if (!m->caller_sensitive()) {
4698 #ifndef PRODUCT
4699 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4700 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4701 }
4702 #endif
4703 return false; // bail-out; let JVM_GetCallerClass do the work
4704 }
4705 break;
4706 default:
4707 if (!m->is_ignored_by_security_stack_walk()) {
4708 // We have reached the desired frame; return the holder class.
4709 // Acquire method holder as java.lang.Class and push as constant.
4710 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4711 ciInstance* caller_mirror = caller_klass->java_mirror();
4712 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4713
4714 #ifndef PRODUCT
4715 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4716 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());
4717 tty->print_cr(" JVM state at this point:");
4718 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4719 ciMethod* m = jvms()->of_depth(i)->method();
4720 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4721 }
4722 }
4723 #endif
4724 return true;
4725 }
4726 break;
4727 }
4728 }
4729
4730 #ifndef PRODUCT
4731 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4732 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4733 tty->print_cr(" JVM state at this point:");
4734 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4735 ciMethod* m = jvms()->of_depth(i)->method();
4736 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4737 }
4738 }
4739 #endif
4740
4741 return false; // bail-out; let JVM_GetCallerClass do the work
4742 }
4743
4744 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4745 Node* arg = argument(0);
4746 Node* result = nullptr;
4747
4748 switch (id) {
4749 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4750 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4751 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4752 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4753 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4754 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4755
4756 case vmIntrinsics::_doubleToLongBits: {
4757 // two paths (plus control) merge in a wood
4758 RegionNode *r = new RegionNode(3);
4759 Node *phi = new PhiNode(r, TypeLong::LONG);
4760
4761 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4762 // Build the boolean node
4763 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4764
4765 // Branch either way.
4766 // NaN case is less traveled, which makes all the difference.
4767 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4768 Node *opt_isnan = _gvn.transform(ifisnan);
4769 assert( opt_isnan->is_If(), "Expect an IfNode");
4770 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4771 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4772
4773 set_control(iftrue);
4774
4775 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4776 Node *slow_result = longcon(nan_bits); // return NaN
4777 phi->init_req(1, _gvn.transform( slow_result ));
4778 r->init_req(1, iftrue);
4779
4780 // Else fall through
4781 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4782 set_control(iffalse);
4783
4784 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4785 r->init_req(2, iffalse);
4786
4787 // Post merge
4788 set_control(_gvn.transform(r));
4789 record_for_igvn(r);
4790
4791 C->set_has_split_ifs(true); // Has chance for split-if optimization
4792 result = phi;
4793 assert(result->bottom_type()->isa_long(), "must be");
4794 break;
4795 }
4796
4797 case vmIntrinsics::_floatToIntBits: {
4798 // two paths (plus control) merge in a wood
4799 RegionNode *r = new RegionNode(3);
4800 Node *phi = new PhiNode(r, TypeInt::INT);
4801
4802 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4803 // Build the boolean node
4804 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4805
4806 // Branch either way.
4807 // NaN case is less traveled, which makes all the difference.
4808 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4809 Node *opt_isnan = _gvn.transform(ifisnan);
4810 assert( opt_isnan->is_If(), "Expect an IfNode");
4811 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4812 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4813
4814 set_control(iftrue);
4815
4816 static const jint nan_bits = 0x7fc00000;
4817 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4818 phi->init_req(1, _gvn.transform( slow_result ));
4819 r->init_req(1, iftrue);
4820
4821 // Else fall through
4822 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4823 set_control(iffalse);
4824
4825 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4826 r->init_req(2, iffalse);
4827
4828 // Post merge
4829 set_control(_gvn.transform(r));
4830 record_for_igvn(r);
4831
4832 C->set_has_split_ifs(true); // Has chance for split-if optimization
4833 result = phi;
4834 assert(result->bottom_type()->isa_int(), "must be");
4835 break;
4836 }
4837
4838 default:
4839 fatal_unexpected_iid(id);
4840 break;
4841 }
4842 set_result(_gvn.transform(result));
4843 return true;
4844 }
4845
4846 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4847 Node* arg = argument(0);
4848 Node* result = nullptr;
4849
4850 switch (id) {
4851 case vmIntrinsics::_floatIsInfinite:
4852 result = new IsInfiniteFNode(arg);
4853 break;
4854 case vmIntrinsics::_floatIsFinite:
4855 result = new IsFiniteFNode(arg);
4856 break;
4857 case vmIntrinsics::_doubleIsInfinite:
4858 result = new IsInfiniteDNode(arg);
4859 break;
4860 case vmIntrinsics::_doubleIsFinite:
4861 result = new IsFiniteDNode(arg);
4862 break;
4863 default:
4864 fatal_unexpected_iid(id);
4865 break;
4866 }
4867 set_result(_gvn.transform(result));
4868 return true;
4869 }
4870
4871 //----------------------inline_unsafe_copyMemory-------------------------
4872 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4873
4874 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4875 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4876 const Type* base_t = gvn.type(base);
4877
4878 bool in_native = (base_t == TypePtr::NULL_PTR);
4879 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4880 bool is_mixed = !in_heap && !in_native;
4881
4882 if (is_mixed) {
4883 return true; // mixed accesses can touch both on-heap and off-heap memory
4884 }
4885 if (in_heap) {
4886 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4887 if (!is_prim_array) {
4888 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4889 // there's not enough type information available to determine proper memory slice for it.
4890 return true;
4891 }
4892 }
4893 return false;
4894 }
4895
4896 bool LibraryCallKit::inline_unsafe_copyMemory() {
4897 if (callee()->is_static()) return false; // caller must have the capability!
4898 null_check_receiver(); // null-check receiver
4899 if (stopped()) return true;
4900
4901 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4902
4903 Node* src_base = argument(1); // type: oop
4904 Node* src_off = ConvL2X(argument(2)); // type: long
4905 Node* dst_base = argument(4); // type: oop
4906 Node* dst_off = ConvL2X(argument(5)); // type: long
4907 Node* size = ConvL2X(argument(7)); // type: long
4908
4909 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4910 "fieldOffset must be byte-scaled");
4911
4912 Node* src_addr = make_unsafe_address(src_base, src_off);
4913 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4914
4915 Node* thread = _gvn.transform(new ThreadLocalNode());
4916 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4917 BasicType doing_unsafe_access_bt = T_BYTE;
4918 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4919
4920 // update volatile field
4921 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4922
4923 int flags = RC_LEAF | RC_NO_FP;
4924
4925 const TypePtr* dst_type = TypePtr::BOTTOM;
4926
4927 // Adjust memory effects of the runtime call based on input values.
4928 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4929 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4930 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4931
4932 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4933 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4934 flags |= RC_NARROW_MEM; // narrow in memory
4935 }
4936 }
4937
4938 // Call it. Note that the length argument is not scaled.
4939 make_runtime_call(flags,
4940 OptoRuntime::fast_arraycopy_Type(),
4941 StubRoutines::unsafe_arraycopy(),
4942 "unsafe_arraycopy",
4943 dst_type,
4944 src_addr, dst_addr, size XTOP);
4945
4946 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4947
4948 return true;
4949 }
4950
4951 #undef XTOP
4952
4953 //------------------------clone_coping-----------------------------------
4954 // Helper function for inline_native_clone.
4955 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
4956 assert(obj_size != nullptr, "");
4957 Node* raw_obj = alloc_obj->in(1);
4958 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
4959
4960 AllocateNode* alloc = nullptr;
4961 if (ReduceBulkZeroing) {
4962 // We will be completely responsible for initializing this object -
4963 // mark Initialize node as complete.
4964 alloc = AllocateNode::Ideal_allocation(alloc_obj);
4965 // The object was just allocated - there should be no any stores!
4966 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
4967 // Mark as complete_with_arraycopy so that on AllocateNode
4968 // expansion, we know this AllocateNode is initialized by an array
4969 // copy and a StoreStore barrier exists after the array copy.
4970 alloc->initialization()->set_complete_with_arraycopy();
4971 }
4972
4973 Node* size = _gvn.transform(obj_size);
4974 access_clone(obj, alloc_obj, size, is_array);
4975
4976 // Do not let reads from the cloned object float above the arraycopy.
4977 if (alloc != nullptr) {
4978 // Do not let stores that initialize this object be reordered with
4979 // a subsequent store that would make this object accessible by
4980 // other threads.
4981 // Record what AllocateNode this StoreStore protects so that
4982 // escape analysis can go from the MemBarStoreStoreNode to the
4983 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
4984 // based on the escape status of the AllocateNode.
4985 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
4986 } else {
4987 insert_mem_bar(Op_MemBarCPUOrder);
4988 }
4989 }
4990
4991 //------------------------inline_native_clone----------------------------
4992 // protected native Object java.lang.Object.clone();
4993 //
4994 // Here are the simple edge cases:
4995 // null receiver => normal trap
4996 // virtual and clone was overridden => slow path to out-of-line clone
4997 // not cloneable or finalizer => slow path to out-of-line Object.clone
4998 //
4999 // The general case has two steps, allocation and copying.
5000 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5001 //
5002 // Copying also has two cases, oop arrays and everything else.
5003 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5004 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5005 //
5006 // These steps fold up nicely if and when the cloned object's klass
5007 // can be sharply typed as an object array, a type array, or an instance.
5008 //
5009 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5010 PhiNode* result_val;
5011
5012 // Set the reexecute bit for the interpreter to reexecute
5013 // the bytecode that invokes Object.clone if deoptimization happens.
5014 { PreserveReexecuteState preexecs(this);
5015 jvms()->set_should_reexecute(true);
5016
5017 Node* obj = null_check_receiver();
5018 if (stopped()) return true;
5019
5020 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5021
5022 // If we are going to clone an instance, we need its exact type to
5023 // know the number and types of fields to convert the clone to
5024 // loads/stores. Maybe a speculative type can help us.
5025 if (!obj_type->klass_is_exact() &&
5026 obj_type->speculative_type() != nullptr &&
5027 obj_type->speculative_type()->is_instance_klass()) {
5028 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5029 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5030 !spec_ik->has_injected_fields()) {
5031 if (!obj_type->isa_instptr() ||
5032 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5033 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5034 }
5035 }
5036 }
5037
5038 // Conservatively insert a memory barrier on all memory slices.
5039 // Do not let writes into the original float below the clone.
5040 insert_mem_bar(Op_MemBarCPUOrder);
5041
5042 // paths into result_reg:
5043 enum {
5044 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5045 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5046 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5047 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5048 PATH_LIMIT
5049 };
5050 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5051 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5052 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5053 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5054 record_for_igvn(result_reg);
5055
5056 Node* obj_klass = load_object_klass(obj);
5057 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5058 if (array_ctl != nullptr) {
5059 // It's an array.
5060 PreserveJVMState pjvms(this);
5061 set_control(array_ctl);
5062 Node* obj_length = load_array_length(obj);
5063 Node* array_size = nullptr; // Size of the array without object alignment padding.
5064 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5065
5066 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5067 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5068 // If it is an oop array, it requires very special treatment,
5069 // because gc barriers are required when accessing the array.
5070 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5071 if (is_obja != nullptr) {
5072 PreserveJVMState pjvms2(this);
5073 set_control(is_obja);
5074 // Generate a direct call to the right arraycopy function(s).
5075 // Clones are always tightly coupled.
5076 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5077 ac->set_clone_oop_array();
5078 Node* n = _gvn.transform(ac);
5079 assert(n == ac, "cannot disappear");
5080 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5081
5082 result_reg->init_req(_objArray_path, control());
5083 result_val->init_req(_objArray_path, alloc_obj);
5084 result_i_o ->set_req(_objArray_path, i_o());
5085 result_mem ->set_req(_objArray_path, reset_memory());
5086 }
5087 }
5088 // Otherwise, there are no barriers to worry about.
5089 // (We can dispense with card marks if we know the allocation
5090 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5091 // causes the non-eden paths to take compensating steps to
5092 // simulate a fresh allocation, so that no further
5093 // card marks are required in compiled code to initialize
5094 // the object.)
5095
5096 if (!stopped()) {
5097 copy_to_clone(obj, alloc_obj, array_size, true);
5098
5099 // Present the results of the copy.
5100 result_reg->init_req(_array_path, control());
5101 result_val->init_req(_array_path, alloc_obj);
5102 result_i_o ->set_req(_array_path, i_o());
5103 result_mem ->set_req(_array_path, reset_memory());
5104 }
5105 }
5106
5107 // We only go to the instance fast case code if we pass a number of guards.
5108 // The paths which do not pass are accumulated in the slow_region.
5109 RegionNode* slow_region = new RegionNode(1);
5110 record_for_igvn(slow_region);
5111 if (!stopped()) {
5112 // It's an instance (we did array above). Make the slow-path tests.
5113 // If this is a virtual call, we generate a funny guard. We grab
5114 // the vtable entry corresponding to clone() from the target object.
5115 // If the target method which we are calling happens to be the
5116 // Object clone() method, we pass the guard. We do not need this
5117 // guard for non-virtual calls; the caller is known to be the native
5118 // Object clone().
5119 if (is_virtual) {
5120 generate_virtual_guard(obj_klass, slow_region);
5121 }
5122
5123 // The object must be easily cloneable and must not have a finalizer.
5124 // Both of these conditions may be checked in a single test.
5125 // We could optimize the test further, but we don't care.
5126 generate_access_flags_guard(obj_klass,
5127 // Test both conditions:
5128 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5129 // Must be cloneable but not finalizer:
5130 JVM_ACC_IS_CLONEABLE_FAST,
5131 slow_region);
5132 }
5133
5134 if (!stopped()) {
5135 // It's an instance, and it passed the slow-path tests.
5136 PreserveJVMState pjvms(this);
5137 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5138 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5139 // is reexecuted if deoptimization occurs and there could be problems when merging
5140 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5141 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5142
5143 copy_to_clone(obj, alloc_obj, obj_size, false);
5144
5145 // Present the results of the slow call.
5146 result_reg->init_req(_instance_path, control());
5147 result_val->init_req(_instance_path, alloc_obj);
5148 result_i_o ->set_req(_instance_path, i_o());
5149 result_mem ->set_req(_instance_path, reset_memory());
5150 }
5151
5152 // Generate code for the slow case. We make a call to clone().
5153 set_control(_gvn.transform(slow_region));
5154 if (!stopped()) {
5155 PreserveJVMState pjvms(this);
5156 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5157 // We need to deoptimize on exception (see comment above)
5158 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5159 // this->control() comes from set_results_for_java_call
5160 result_reg->init_req(_slow_path, control());
5161 result_val->init_req(_slow_path, slow_result);
5162 result_i_o ->set_req(_slow_path, i_o());
5163 result_mem ->set_req(_slow_path, reset_memory());
5164 }
5165
5166 // Return the combined state.
5167 set_control( _gvn.transform(result_reg));
5168 set_i_o( _gvn.transform(result_i_o));
5169 set_all_memory( _gvn.transform(result_mem));
5170 } // original reexecute is set back here
5171
5172 set_result(_gvn.transform(result_val));
5173 return true;
5174 }
5175
5176 // If we have a tightly coupled allocation, the arraycopy may take care
5177 // of the array initialization. If one of the guards we insert between
5178 // the allocation and the arraycopy causes a deoptimization, an
5179 // uninitialized array will escape the compiled method. To prevent that
5180 // we set the JVM state for uncommon traps between the allocation and
5181 // the arraycopy to the state before the allocation so, in case of
5182 // deoptimization, we'll reexecute the allocation and the
5183 // initialization.
5184 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5185 if (alloc != nullptr) {
5186 ciMethod* trap_method = alloc->jvms()->method();
5187 int trap_bci = alloc->jvms()->bci();
5188
5189 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5190 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5191 // Make sure there's no store between the allocation and the
5192 // arraycopy otherwise visible side effects could be rexecuted
5193 // in case of deoptimization and cause incorrect execution.
5194 bool no_interfering_store = true;
5195 Node* mem = alloc->in(TypeFunc::Memory);
5196 if (mem->is_MergeMem()) {
5197 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5198 Node* n = mms.memory();
5199 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5200 assert(n->is_Store(), "what else?");
5201 no_interfering_store = false;
5202 break;
5203 }
5204 }
5205 } else {
5206 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5207 Node* n = mms.memory();
5208 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5209 assert(n->is_Store(), "what else?");
5210 no_interfering_store = false;
5211 break;
5212 }
5213 }
5214 }
5215
5216 if (no_interfering_store) {
5217 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5218
5219 JVMState* saved_jvms = jvms();
5220 saved_reexecute_sp = _reexecute_sp;
5221
5222 set_jvms(sfpt->jvms());
5223 _reexecute_sp = jvms()->sp();
5224
5225 return saved_jvms;
5226 }
5227 }
5228 }
5229 return nullptr;
5230 }
5231
5232 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5233 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5234 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5235 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5236 uint size = alloc->req();
5237 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5238 old_jvms->set_map(sfpt);
5239 for (uint i = 0; i < size; i++) {
5240 sfpt->init_req(i, alloc->in(i));
5241 }
5242 // re-push array length for deoptimization
5243 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5244 old_jvms->set_sp(old_jvms->sp()+1);
5245 old_jvms->set_monoff(old_jvms->monoff()+1);
5246 old_jvms->set_scloff(old_jvms->scloff()+1);
5247 old_jvms->set_endoff(old_jvms->endoff()+1);
5248 old_jvms->set_should_reexecute(true);
5249
5250 sfpt->set_i_o(map()->i_o());
5251 sfpt->set_memory(map()->memory());
5252 sfpt->set_control(map()->control());
5253 return sfpt;
5254 }
5255
5256 // In case of a deoptimization, we restart execution at the
5257 // allocation, allocating a new array. We would leave an uninitialized
5258 // array in the heap that GCs wouldn't expect. Move the allocation
5259 // after the traps so we don't allocate the array if we
5260 // deoptimize. This is possible because tightly_coupled_allocation()
5261 // guarantees there's no observer of the allocated array at this point
5262 // and the control flow is simple enough.
5263 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5264 int saved_reexecute_sp, uint new_idx) {
5265 if (saved_jvms_before_guards != nullptr && !stopped()) {
5266 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5267
5268 assert(alloc != nullptr, "only with a tightly coupled allocation");
5269 // restore JVM state to the state at the arraycopy
5270 saved_jvms_before_guards->map()->set_control(map()->control());
5271 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5272 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5273 // If we've improved the types of some nodes (null check) while
5274 // emitting the guards, propagate them to the current state
5275 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5276 set_jvms(saved_jvms_before_guards);
5277 _reexecute_sp = saved_reexecute_sp;
5278
5279 // Remove the allocation from above the guards
5280 CallProjections callprojs;
5281 alloc->extract_projections(&callprojs, true);
5282 InitializeNode* init = alloc->initialization();
5283 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5284 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5285 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5286
5287 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5288 // the allocation (i.e. is only valid if the allocation succeeds):
5289 // 1) replace CastIINode with AllocateArrayNode's length here
5290 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5291 //
5292 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5293 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5294 Node* init_control = init->proj_out(TypeFunc::Control);
5295 Node* alloc_length = alloc->Ideal_length();
5296 #ifdef ASSERT
5297 Node* prev_cast = nullptr;
5298 #endif
5299 for (uint i = 0; i < init_control->outcnt(); i++) {
5300 Node* init_out = init_control->raw_out(i);
5301 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5302 #ifdef ASSERT
5303 if (prev_cast == nullptr) {
5304 prev_cast = init_out;
5305 } else {
5306 if (prev_cast->cmp(*init_out) == false) {
5307 prev_cast->dump();
5308 init_out->dump();
5309 assert(false, "not equal CastIINode");
5310 }
5311 }
5312 #endif
5313 C->gvn_replace_by(init_out, alloc_length);
5314 }
5315 }
5316 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5317
5318 // move the allocation here (after the guards)
5319 _gvn.hash_delete(alloc);
5320 alloc->set_req(TypeFunc::Control, control());
5321 alloc->set_req(TypeFunc::I_O, i_o());
5322 Node *mem = reset_memory();
5323 set_all_memory(mem);
5324 alloc->set_req(TypeFunc::Memory, mem);
5325 set_control(init->proj_out_or_null(TypeFunc::Control));
5326 set_i_o(callprojs.fallthrough_ioproj);
5327
5328 // Update memory as done in GraphKit::set_output_for_allocation()
5329 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5330 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5331 if (ary_type->isa_aryptr() && length_type != nullptr) {
5332 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5333 }
5334 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5335 int elemidx = C->get_alias_index(telemref);
5336 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5337 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5338
5339 Node* allocx = _gvn.transform(alloc);
5340 assert(allocx == alloc, "where has the allocation gone?");
5341 assert(dest->is_CheckCastPP(), "not an allocation result?");
5342
5343 _gvn.hash_delete(dest);
5344 dest->set_req(0, control());
5345 Node* destx = _gvn.transform(dest);
5346 assert(destx == dest, "where has the allocation result gone?");
5347
5348 array_ideal_length(alloc, ary_type, true);
5349 }
5350 }
5351
5352 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5353 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5354 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5355 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5356 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5357 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5358 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5359 JVMState* saved_jvms_before_guards) {
5360 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5361 // There is at least one unrelated uncommon trap which needs to be replaced.
5362 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5363
5364 JVMState* saved_jvms = jvms();
5365 const int saved_reexecute_sp = _reexecute_sp;
5366 set_jvms(sfpt->jvms());
5367 _reexecute_sp = jvms()->sp();
5368
5369 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5370
5371 // Restore state
5372 set_jvms(saved_jvms);
5373 _reexecute_sp = saved_reexecute_sp;
5374 }
5375 }
5376
5377 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5378 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5379 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5380 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5381 while (if_proj->is_IfProj()) {
5382 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5383 if (uncommon_trap != nullptr) {
5384 create_new_uncommon_trap(uncommon_trap);
5385 }
5386 assert(if_proj->in(0)->is_If(), "must be If");
5387 if_proj = if_proj->in(0)->in(0);
5388 }
5389 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5390 "must have reached control projection of init node");
5391 }
5392
5393 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5394 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5395 assert(trap_request != 0, "no valid UCT trap request");
5396 PreserveJVMState pjvms(this);
5397 set_control(uncommon_trap_call->in(0));
5398 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5399 Deoptimization::trap_request_action(trap_request));
5400 assert(stopped(), "Should be stopped");
5401 _gvn.hash_delete(uncommon_trap_call);
5402 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5403 }
5404
5405 //------------------------------inline_array_partition-----------------------
5406 bool LibraryCallKit::inline_array_partition() {
5407
5408 Node* elementType = null_check(argument(0));
5409 Node* obj = argument(1);
5410 Node* offset = argument(2);
5411 Node* fromIndex = argument(4);
5412 Node* toIndex = argument(5);
5413 Node* indexPivot1 = argument(6);
5414 Node* indexPivot2 = argument(7);
5415
5416 Node* pivotIndices = nullptr;
5417
5418 // Set the original stack and the reexecute bit for the interpreter to reexecute
5419 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5420 { PreserveReexecuteState preexecs(this);
5421 jvms()->set_should_reexecute(true);
5422
5423 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5424 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5425 BasicType bt = elem_type->basic_type();
5426 // Disable the intrinsic if the CPU does not support SIMD sort
5427 if (!Matcher::supports_simd_sort(bt)) {
5428 return false;
5429 }
5430 address stubAddr = nullptr;
5431 stubAddr = StubRoutines::select_array_partition_function();
5432 // stub not loaded
5433 if (stubAddr == nullptr) {
5434 return false;
5435 }
5436 // get the address of the array
5437 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5438 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5439 return false; // failed input validation
5440 }
5441 Node* obj_adr = make_unsafe_address(obj, offset);
5442
5443 // create the pivotIndices array of type int and size = 2
5444 Node* size = intcon(2);
5445 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5446 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5447 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5448 guarantee(alloc != nullptr, "created above");
5449 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5450
5451 // pass the basic type enum to the stub
5452 Node* elemType = intcon(bt);
5453
5454 // Call the stub
5455 const char *stubName = "array_partition_stub";
5456 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5457 stubAddr, stubName, TypePtr::BOTTOM,
5458 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5459 indexPivot1, indexPivot2);
5460
5461 } // original reexecute is set back here
5462
5463 if (!stopped()) {
5464 set_result(pivotIndices);
5465 }
5466
5467 return true;
5468 }
5469
5470
5471 //------------------------------inline_array_sort-----------------------
5472 bool LibraryCallKit::inline_array_sort() {
5473
5474 Node* elementType = null_check(argument(0));
5475 Node* obj = argument(1);
5476 Node* offset = argument(2);
5477 Node* fromIndex = argument(4);
5478 Node* toIndex = argument(5);
5479
5480 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5481 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5482 BasicType bt = elem_type->basic_type();
5483 // Disable the intrinsic if the CPU does not support SIMD sort
5484 if (!Matcher::supports_simd_sort(bt)) {
5485 return false;
5486 }
5487 address stubAddr = nullptr;
5488 stubAddr = StubRoutines::select_arraysort_function();
5489 //stub not loaded
5490 if (stubAddr == nullptr) {
5491 return false;
5492 }
5493
5494 // get address of the array
5495 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5496 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5497 return false; // failed input validation
5498 }
5499 Node* obj_adr = make_unsafe_address(obj, offset);
5500
5501 // pass the basic type enum to the stub
5502 Node* elemType = intcon(bt);
5503
5504 // Call the stub.
5505 const char *stubName = "arraysort_stub";
5506 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5507 stubAddr, stubName, TypePtr::BOTTOM,
5508 obj_adr, elemType, fromIndex, toIndex);
5509
5510 return true;
5511 }
5512
5513
5514 //------------------------------inline_arraycopy-----------------------
5515 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5516 // Object dest, int destPos,
5517 // int length);
5518 bool LibraryCallKit::inline_arraycopy() {
5519 // Get the arguments.
5520 Node* src = argument(0); // type: oop
5521 Node* src_offset = argument(1); // type: int
5522 Node* dest = argument(2); // type: oop
5523 Node* dest_offset = argument(3); // type: int
5524 Node* length = argument(4); // type: int
5525
5526 uint new_idx = C->unique();
5527
5528 // Check for allocation before we add nodes that would confuse
5529 // tightly_coupled_allocation()
5530 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5531
5532 int saved_reexecute_sp = -1;
5533 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5534 // See arraycopy_restore_alloc_state() comment
5535 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5536 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5537 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5538 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5539
5540 // The following tests must be performed
5541 // (1) src and dest are arrays.
5542 // (2) src and dest arrays must have elements of the same BasicType
5543 // (3) src and dest must not be null.
5544 // (4) src_offset must not be negative.
5545 // (5) dest_offset must not be negative.
5546 // (6) length must not be negative.
5547 // (7) src_offset + length must not exceed length of src.
5548 // (8) dest_offset + length must not exceed length of dest.
5549 // (9) each element of an oop array must be assignable
5550
5551 // (3) src and dest must not be null.
5552 // always do this here because we need the JVM state for uncommon traps
5553 Node* null_ctl = top();
5554 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5555 assert(null_ctl->is_top(), "no null control here");
5556 dest = null_check(dest, T_ARRAY);
5557
5558 if (!can_emit_guards) {
5559 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5560 // guards but the arraycopy node could still take advantage of a
5561 // tightly allocated allocation. tightly_coupled_allocation() is
5562 // called again to make sure it takes the null check above into
5563 // account: the null check is mandatory and if it caused an
5564 // uncommon trap to be emitted then the allocation can't be
5565 // considered tightly coupled in this context.
5566 alloc = tightly_coupled_allocation(dest);
5567 }
5568
5569 bool validated = false;
5570
5571 const Type* src_type = _gvn.type(src);
5572 const Type* dest_type = _gvn.type(dest);
5573 const TypeAryPtr* top_src = src_type->isa_aryptr();
5574 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5575
5576 // Do we have the type of src?
5577 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5578 // Do we have the type of dest?
5579 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5580 // Is the type for src from speculation?
5581 bool src_spec = false;
5582 // Is the type for dest from speculation?
5583 bool dest_spec = false;
5584
5585 if ((!has_src || !has_dest) && can_emit_guards) {
5586 // We don't have sufficient type information, let's see if
5587 // speculative types can help. We need to have types for both src
5588 // and dest so that it pays off.
5589
5590 // Do we already have or could we have type information for src
5591 bool could_have_src = has_src;
5592 // Do we already have or could we have type information for dest
5593 bool could_have_dest = has_dest;
5594
5595 ciKlass* src_k = nullptr;
5596 if (!has_src) {
5597 src_k = src_type->speculative_type_not_null();
5598 if (src_k != nullptr && src_k->is_array_klass()) {
5599 could_have_src = true;
5600 }
5601 }
5602
5603 ciKlass* dest_k = nullptr;
5604 if (!has_dest) {
5605 dest_k = dest_type->speculative_type_not_null();
5606 if (dest_k != nullptr && dest_k->is_array_klass()) {
5607 could_have_dest = true;
5608 }
5609 }
5610
5611 if (could_have_src && could_have_dest) {
5612 // This is going to pay off so emit the required guards
5613 if (!has_src) {
5614 src = maybe_cast_profiled_obj(src, src_k, true);
5615 src_type = _gvn.type(src);
5616 top_src = src_type->isa_aryptr();
5617 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5618 src_spec = true;
5619 }
5620 if (!has_dest) {
5621 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5622 dest_type = _gvn.type(dest);
5623 top_dest = dest_type->isa_aryptr();
5624 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5625 dest_spec = true;
5626 }
5627 }
5628 }
5629
5630 if (has_src && has_dest && can_emit_guards) {
5631 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5632 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5633 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5634 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5635
5636 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5637 // If both arrays are object arrays then having the exact types
5638 // for both will remove the need for a subtype check at runtime
5639 // before the call and may make it possible to pick a faster copy
5640 // routine (without a subtype check on every element)
5641 // Do we have the exact type of src?
5642 bool could_have_src = src_spec;
5643 // Do we have the exact type of dest?
5644 bool could_have_dest = dest_spec;
5645 ciKlass* src_k = nullptr;
5646 ciKlass* dest_k = nullptr;
5647 if (!src_spec) {
5648 src_k = src_type->speculative_type_not_null();
5649 if (src_k != nullptr && src_k->is_array_klass()) {
5650 could_have_src = true;
5651 }
5652 }
5653 if (!dest_spec) {
5654 dest_k = dest_type->speculative_type_not_null();
5655 if (dest_k != nullptr && dest_k->is_array_klass()) {
5656 could_have_dest = true;
5657 }
5658 }
5659 if (could_have_src && could_have_dest) {
5660 // If we can have both exact types, emit the missing guards
5661 if (could_have_src && !src_spec) {
5662 src = maybe_cast_profiled_obj(src, src_k, true);
5663 }
5664 if (could_have_dest && !dest_spec) {
5665 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5666 }
5667 }
5668 }
5669 }
5670
5671 ciMethod* trap_method = method();
5672 int trap_bci = bci();
5673 if (saved_jvms_before_guards != nullptr) {
5674 trap_method = alloc->jvms()->method();
5675 trap_bci = alloc->jvms()->bci();
5676 }
5677
5678 bool negative_length_guard_generated = false;
5679
5680 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5681 can_emit_guards &&
5682 !src->is_top() && !dest->is_top()) {
5683 // validate arguments: enables transformation the ArrayCopyNode
5684 validated = true;
5685
5686 RegionNode* slow_region = new RegionNode(1);
5687 record_for_igvn(slow_region);
5688
5689 // (1) src and dest are arrays.
5690 generate_non_array_guard(load_object_klass(src), slow_region);
5691 generate_non_array_guard(load_object_klass(dest), slow_region);
5692
5693 // (2) src and dest arrays must have elements of the same BasicType
5694 // done at macro expansion or at Ideal transformation time
5695
5696 // (4) src_offset must not be negative.
5697 generate_negative_guard(src_offset, slow_region);
5698
5699 // (5) dest_offset must not be negative.
5700 generate_negative_guard(dest_offset, slow_region);
5701
5702 // (7) src_offset + length must not exceed length of src.
5703 generate_limit_guard(src_offset, length,
5704 load_array_length(src),
5705 slow_region);
5706
5707 // (8) dest_offset + length must not exceed length of dest.
5708 generate_limit_guard(dest_offset, length,
5709 load_array_length(dest),
5710 slow_region);
5711
5712 // (6) length must not be negative.
5713 // This is also checked in generate_arraycopy() during macro expansion, but
5714 // we also have to check it here for the case where the ArrayCopyNode will
5715 // be eliminated by Escape Analysis.
5716 if (EliminateAllocations) {
5717 generate_negative_guard(length, slow_region);
5718 negative_length_guard_generated = true;
5719 }
5720
5721 // (9) each element of an oop array must be assignable
5722 Node* dest_klass = load_object_klass(dest);
5723 if (src != dest) {
5724 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5725
5726 if (not_subtype_ctrl != top()) {
5727 PreserveJVMState pjvms(this);
5728 set_control(not_subtype_ctrl);
5729 uncommon_trap(Deoptimization::Reason_intrinsic,
5730 Deoptimization::Action_make_not_entrant);
5731 assert(stopped(), "Should be stopped");
5732 }
5733 }
5734 {
5735 PreserveJVMState pjvms(this);
5736 set_control(_gvn.transform(slow_region));
5737 uncommon_trap(Deoptimization::Reason_intrinsic,
5738 Deoptimization::Action_make_not_entrant);
5739 assert(stopped(), "Should be stopped");
5740 }
5741
5742 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5743 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5744 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5745 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5746 }
5747
5748 if (stopped()) {
5749 return true;
5750 }
5751
5752 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5753 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5754 // so the compiler has a chance to eliminate them: during macro expansion,
5755 // we have to set their control (CastPP nodes are eliminated).
5756 load_object_klass(src), load_object_klass(dest),
5757 load_array_length(src), load_array_length(dest));
5758
5759 ac->set_arraycopy(validated);
5760
5761 Node* n = _gvn.transform(ac);
5762 if (n == ac) {
5763 ac->connect_outputs(this);
5764 } else {
5765 assert(validated, "shouldn't transform if all arguments not validated");
5766 set_all_memory(n);
5767 }
5768 clear_upper_avx();
5769
5770
5771 return true;
5772 }
5773
5774
5775 // Helper function which determines if an arraycopy immediately follows
5776 // an allocation, with no intervening tests or other escapes for the object.
5777 AllocateArrayNode*
5778 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5779 if (stopped()) return nullptr; // no fast path
5780 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5781
5782 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5783 if (alloc == nullptr) return nullptr;
5784
5785 Node* rawmem = memory(Compile::AliasIdxRaw);
5786 // Is the allocation's memory state untouched?
5787 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5788 // Bail out if there have been raw-memory effects since the allocation.
5789 // (Example: There might have been a call or safepoint.)
5790 return nullptr;
5791 }
5792 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5793 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5794 return nullptr;
5795 }
5796
5797 // There must be no unexpected observers of this allocation.
5798 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5799 Node* obs = ptr->fast_out(i);
5800 if (obs != this->map()) {
5801 return nullptr;
5802 }
5803 }
5804
5805 // This arraycopy must unconditionally follow the allocation of the ptr.
5806 Node* alloc_ctl = ptr->in(0);
5807 Node* ctl = control();
5808 while (ctl != alloc_ctl) {
5809 // There may be guards which feed into the slow_region.
5810 // Any other control flow means that we might not get a chance
5811 // to finish initializing the allocated object.
5812 // Various low-level checks bottom out in uncommon traps. These
5813 // are considered safe since we've already checked above that
5814 // there is no unexpected observer of this allocation.
5815 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5816 assert(ctl->in(0)->is_If(), "must be If");
5817 ctl = ctl->in(0)->in(0);
5818 } else {
5819 return nullptr;
5820 }
5821 }
5822
5823 // If we get this far, we have an allocation which immediately
5824 // precedes the arraycopy, and we can take over zeroing the new object.
5825 // The arraycopy will finish the initialization, and provide
5826 // a new control state to which we will anchor the destination pointer.
5827
5828 return alloc;
5829 }
5830
5831 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5832 if (node->is_IfProj()) {
5833 Node* other_proj = node->as_IfProj()->other_if_proj();
5834 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5835 Node* obs = other_proj->fast_out(j);
5836 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5837 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5838 return obs->as_CallStaticJava();
5839 }
5840 }
5841 }
5842 return nullptr;
5843 }
5844
5845 //-------------inline_encodeISOArray-----------------------------------
5846 // encode char[] to byte[] in ISO_8859_1 or ASCII
5847 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5848 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5849 // no receiver since it is static method
5850 Node *src = argument(0);
5851 Node *src_offset = argument(1);
5852 Node *dst = argument(2);
5853 Node *dst_offset = argument(3);
5854 Node *length = argument(4);
5855
5856 src = must_be_not_null(src, true);
5857 dst = must_be_not_null(dst, true);
5858
5859 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5860 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5861 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5862 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5863 // failed array check
5864 return false;
5865 }
5866
5867 // Figure out the size and type of the elements we will be copying.
5868 BasicType src_elem = src_type->elem()->array_element_basic_type();
5869 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5870 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5871 return false;
5872 }
5873
5874 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5875 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5876 // 'src_start' points to src array + scaled offset
5877 // 'dst_start' points to dst array + scaled offset
5878
5879 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5880 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5881 enc = _gvn.transform(enc);
5882 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5883 set_memory(res_mem, mtype);
5884 set_result(enc);
5885 clear_upper_avx();
5886
5887 return true;
5888 }
5889
5890 //-------------inline_multiplyToLen-----------------------------------
5891 bool LibraryCallKit::inline_multiplyToLen() {
5892 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5893
5894 address stubAddr = StubRoutines::multiplyToLen();
5895 if (stubAddr == nullptr) {
5896 return false; // Intrinsic's stub is not implemented on this platform
5897 }
5898 const char* stubName = "multiplyToLen";
5899
5900 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5901
5902 // no receiver because it is a static method
5903 Node* x = argument(0);
5904 Node* xlen = argument(1);
5905 Node* y = argument(2);
5906 Node* ylen = argument(3);
5907 Node* z = argument(4);
5908
5909 x = must_be_not_null(x, true);
5910 y = must_be_not_null(y, true);
5911
5912 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5913 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5914 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5915 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5916 // failed array check
5917 return false;
5918 }
5919
5920 BasicType x_elem = x_type->elem()->array_element_basic_type();
5921 BasicType y_elem = y_type->elem()->array_element_basic_type();
5922 if (x_elem != T_INT || y_elem != T_INT) {
5923 return false;
5924 }
5925
5926 // Set the original stack and the reexecute bit for the interpreter to reexecute
5927 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5928 // on the return from z array allocation in runtime.
5929 { PreserveReexecuteState preexecs(this);
5930 jvms()->set_should_reexecute(true);
5931
5932 Node* x_start = array_element_address(x, intcon(0), x_elem);
5933 Node* y_start = array_element_address(y, intcon(0), y_elem);
5934 // 'x_start' points to x array + scaled xlen
5935 // 'y_start' points to y array + scaled ylen
5936
5937 // Allocate the result array
5938 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5939 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5940 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5941
5942 IdealKit ideal(this);
5943
5944 #define __ ideal.
5945 Node* one = __ ConI(1);
5946 Node* zero = __ ConI(0);
5947 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
5948 __ set(need_alloc, zero);
5949 __ set(z_alloc, z);
5950 __ if_then(z, BoolTest::eq, null()); {
5951 __ increment (need_alloc, one);
5952 } __ else_(); {
5953 // Update graphKit memory and control from IdealKit.
5954 sync_kit(ideal);
5955 Node* cast = new CastPPNode(control(), z, TypePtr::NOTNULL);
5956 _gvn.set_type(cast, cast->bottom_type());
5957 C->record_for_igvn(cast);
5958
5959 Node* zlen_arg = load_array_length(cast);
5960 // Update IdealKit memory and control from graphKit.
5961 __ sync_kit(this);
5962 __ if_then(zlen_arg, BoolTest::lt, zlen); {
5963 __ increment (need_alloc, one);
5964 } __ end_if();
5965 } __ end_if();
5966
5967 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
5968 // Update graphKit memory and control from IdealKit.
5969 sync_kit(ideal);
5970 Node * narr = new_array(klass_node, zlen, 1);
5971 // Update IdealKit memory and control from graphKit.
5972 __ sync_kit(this);
5973 __ set(z_alloc, narr);
5974 } __ end_if();
5975
5976 sync_kit(ideal);
5977 z = __ value(z_alloc);
5978 // Can't use TypeAryPtr::INTS which uses Bottom offset.
5979 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
5980 // Final sync IdealKit and GraphKit.
5981 final_sync(ideal);
5982 #undef __
5983
5984 Node* z_start = array_element_address(z, intcon(0), T_INT);
5985
5986 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
5987 OptoRuntime::multiplyToLen_Type(),
5988 stubAddr, stubName, TypePtr::BOTTOM,
5989 x_start, xlen, y_start, ylen, z_start, zlen);
5990 } // original reexecute is set back here
5991
5992 C->set_has_split_ifs(true); // Has chance for split-if optimization
5993 set_result(z);
5994 return true;
5995 }
5996
5997 //-------------inline_squareToLen------------------------------------
5998 bool LibraryCallKit::inline_squareToLen() {
5999 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6000
6001 address stubAddr = StubRoutines::squareToLen();
6002 if (stubAddr == nullptr) {
6003 return false; // Intrinsic's stub is not implemented on this platform
6004 }
6005 const char* stubName = "squareToLen";
6006
6007 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6008
6009 Node* x = argument(0);
6010 Node* len = argument(1);
6011 Node* z = argument(2);
6012 Node* zlen = argument(3);
6013
6014 x = must_be_not_null(x, true);
6015 z = must_be_not_null(z, true);
6016
6017 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6018 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6019 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6020 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6021 // failed array check
6022 return false;
6023 }
6024
6025 BasicType x_elem = x_type->elem()->array_element_basic_type();
6026 BasicType z_elem = z_type->elem()->array_element_basic_type();
6027 if (x_elem != T_INT || z_elem != T_INT) {
6028 return false;
6029 }
6030
6031
6032 Node* x_start = array_element_address(x, intcon(0), x_elem);
6033 Node* z_start = array_element_address(z, intcon(0), z_elem);
6034
6035 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6036 OptoRuntime::squareToLen_Type(),
6037 stubAddr, stubName, TypePtr::BOTTOM,
6038 x_start, len, z_start, zlen);
6039
6040 set_result(z);
6041 return true;
6042 }
6043
6044 //-------------inline_mulAdd------------------------------------------
6045 bool LibraryCallKit::inline_mulAdd() {
6046 assert(UseMulAddIntrinsic, "not implemented on this platform");
6047
6048 address stubAddr = StubRoutines::mulAdd();
6049 if (stubAddr == nullptr) {
6050 return false; // Intrinsic's stub is not implemented on this platform
6051 }
6052 const char* stubName = "mulAdd";
6053
6054 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6055
6056 Node* out = argument(0);
6057 Node* in = argument(1);
6058 Node* offset = argument(2);
6059 Node* len = argument(3);
6060 Node* k = argument(4);
6061
6062 in = must_be_not_null(in, true);
6063 out = must_be_not_null(out, true);
6064
6065 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6066 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6067 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6068 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6069 // failed array check
6070 return false;
6071 }
6072
6073 BasicType out_elem = out_type->elem()->array_element_basic_type();
6074 BasicType in_elem = in_type->elem()->array_element_basic_type();
6075 if (out_elem != T_INT || in_elem != T_INT) {
6076 return false;
6077 }
6078
6079 Node* outlen = load_array_length(out);
6080 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6081 Node* out_start = array_element_address(out, intcon(0), out_elem);
6082 Node* in_start = array_element_address(in, intcon(0), in_elem);
6083
6084 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6085 OptoRuntime::mulAdd_Type(),
6086 stubAddr, stubName, TypePtr::BOTTOM,
6087 out_start,in_start, new_offset, len, k);
6088 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6089 set_result(result);
6090 return true;
6091 }
6092
6093 //-------------inline_montgomeryMultiply-----------------------------------
6094 bool LibraryCallKit::inline_montgomeryMultiply() {
6095 address stubAddr = StubRoutines::montgomeryMultiply();
6096 if (stubAddr == nullptr) {
6097 return false; // Intrinsic's stub is not implemented on this platform
6098 }
6099
6100 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6101 const char* stubName = "montgomery_multiply";
6102
6103 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6104
6105 Node* a = argument(0);
6106 Node* b = argument(1);
6107 Node* n = argument(2);
6108 Node* len = argument(3);
6109 Node* inv = argument(4);
6110 Node* m = argument(6);
6111
6112 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6113 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6114 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6115 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6116 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6117 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6118 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6119 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6120 // failed array check
6121 return false;
6122 }
6123
6124 BasicType a_elem = a_type->elem()->array_element_basic_type();
6125 BasicType b_elem = b_type->elem()->array_element_basic_type();
6126 BasicType n_elem = n_type->elem()->array_element_basic_type();
6127 BasicType m_elem = m_type->elem()->array_element_basic_type();
6128 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6129 return false;
6130 }
6131
6132 // Make the call
6133 {
6134 Node* a_start = array_element_address(a, intcon(0), a_elem);
6135 Node* b_start = array_element_address(b, intcon(0), b_elem);
6136 Node* n_start = array_element_address(n, intcon(0), n_elem);
6137 Node* m_start = array_element_address(m, intcon(0), m_elem);
6138
6139 Node* call = make_runtime_call(RC_LEAF,
6140 OptoRuntime::montgomeryMultiply_Type(),
6141 stubAddr, stubName, TypePtr::BOTTOM,
6142 a_start, b_start, n_start, len, inv, top(),
6143 m_start);
6144 set_result(m);
6145 }
6146
6147 return true;
6148 }
6149
6150 bool LibraryCallKit::inline_montgomerySquare() {
6151 address stubAddr = StubRoutines::montgomerySquare();
6152 if (stubAddr == nullptr) {
6153 return false; // Intrinsic's stub is not implemented on this platform
6154 }
6155
6156 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6157 const char* stubName = "montgomery_square";
6158
6159 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6160
6161 Node* a = argument(0);
6162 Node* n = argument(1);
6163 Node* len = argument(2);
6164 Node* inv = argument(3);
6165 Node* m = argument(5);
6166
6167 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6168 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6169 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6170 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6171 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6172 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6173 // failed array check
6174 return false;
6175 }
6176
6177 BasicType a_elem = a_type->elem()->array_element_basic_type();
6178 BasicType n_elem = n_type->elem()->array_element_basic_type();
6179 BasicType m_elem = m_type->elem()->array_element_basic_type();
6180 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6181 return false;
6182 }
6183
6184 // Make the call
6185 {
6186 Node* a_start = array_element_address(a, intcon(0), a_elem);
6187 Node* n_start = array_element_address(n, intcon(0), n_elem);
6188 Node* m_start = array_element_address(m, intcon(0), m_elem);
6189
6190 Node* call = make_runtime_call(RC_LEAF,
6191 OptoRuntime::montgomerySquare_Type(),
6192 stubAddr, stubName, TypePtr::BOTTOM,
6193 a_start, n_start, len, inv, top(),
6194 m_start);
6195 set_result(m);
6196 }
6197
6198 return true;
6199 }
6200
6201 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6202 address stubAddr = nullptr;
6203 const char* stubName = nullptr;
6204
6205 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6206 if (stubAddr == nullptr) {
6207 return false; // Intrinsic's stub is not implemented on this platform
6208 }
6209
6210 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6211
6212 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6213
6214 Node* newArr = argument(0);
6215 Node* oldArr = argument(1);
6216 Node* newIdx = argument(2);
6217 Node* shiftCount = argument(3);
6218 Node* numIter = argument(4);
6219
6220 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6221 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6222 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6223 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6224 return false;
6225 }
6226
6227 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6228 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6229 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6230 return false;
6231 }
6232
6233 // Make the call
6234 {
6235 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6236 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6237
6238 Node* call = make_runtime_call(RC_LEAF,
6239 OptoRuntime::bigIntegerShift_Type(),
6240 stubAddr,
6241 stubName,
6242 TypePtr::BOTTOM,
6243 newArr_start,
6244 oldArr_start,
6245 newIdx,
6246 shiftCount,
6247 numIter);
6248 }
6249
6250 return true;
6251 }
6252
6253 //-------------inline_vectorizedMismatch------------------------------
6254 bool LibraryCallKit::inline_vectorizedMismatch() {
6255 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6256
6257 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6258 Node* obja = argument(0); // Object
6259 Node* aoffset = argument(1); // long
6260 Node* objb = argument(3); // Object
6261 Node* boffset = argument(4); // long
6262 Node* length = argument(6); // int
6263 Node* scale = argument(7); // int
6264
6265 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6266 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6267 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6268 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6269 scale == top()) {
6270 return false; // failed input validation
6271 }
6272
6273 Node* obja_adr = make_unsafe_address(obja, aoffset);
6274 Node* objb_adr = make_unsafe_address(objb, boffset);
6275
6276 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6277 //
6278 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6279 // if (length <= inline_limit) {
6280 // inline_path:
6281 // vmask = VectorMaskGen length
6282 // vload1 = LoadVectorMasked obja, vmask
6283 // vload2 = LoadVectorMasked objb, vmask
6284 // result1 = VectorCmpMasked vload1, vload2, vmask
6285 // } else {
6286 // call_stub_path:
6287 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6288 // }
6289 // exit_block:
6290 // return Phi(result1, result2);
6291 //
6292 enum { inline_path = 1, // input is small enough to process it all at once
6293 stub_path = 2, // input is too large; call into the VM
6294 PATH_LIMIT = 3
6295 };
6296
6297 Node* exit_block = new RegionNode(PATH_LIMIT);
6298 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6299 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6300
6301 Node* call_stub_path = control();
6302
6303 BasicType elem_bt = T_ILLEGAL;
6304
6305 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6306 if (scale_t->is_con()) {
6307 switch (scale_t->get_con()) {
6308 case 0: elem_bt = T_BYTE; break;
6309 case 1: elem_bt = T_SHORT; break;
6310 case 2: elem_bt = T_INT; break;
6311 case 3: elem_bt = T_LONG; break;
6312
6313 default: elem_bt = T_ILLEGAL; break; // not supported
6314 }
6315 }
6316
6317 int inline_limit = 0;
6318 bool do_partial_inline = false;
6319
6320 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6321 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6322 do_partial_inline = inline_limit >= 16;
6323 }
6324
6325 if (do_partial_inline) {
6326 assert(elem_bt != T_ILLEGAL, "sanity");
6327
6328 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6329 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6330 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6331
6332 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6333 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6334 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6335
6336 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6337
6338 if (!stopped()) {
6339 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6340
6341 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6342 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6343 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6344 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6345
6346 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6347 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6348 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6349 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6350
6351 exit_block->init_req(inline_path, control());
6352 memory_phi->init_req(inline_path, map()->memory());
6353 result_phi->init_req(inline_path, result);
6354
6355 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6356 clear_upper_avx();
6357 }
6358 }
6359 }
6360
6361 if (call_stub_path != nullptr) {
6362 set_control(call_stub_path);
6363
6364 Node* call = make_runtime_call(RC_LEAF,
6365 OptoRuntime::vectorizedMismatch_Type(),
6366 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6367 obja_adr, objb_adr, length, scale);
6368
6369 exit_block->init_req(stub_path, control());
6370 memory_phi->init_req(stub_path, map()->memory());
6371 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6372 }
6373
6374 exit_block = _gvn.transform(exit_block);
6375 memory_phi = _gvn.transform(memory_phi);
6376 result_phi = _gvn.transform(result_phi);
6377
6378 set_control(exit_block);
6379 set_all_memory(memory_phi);
6380 set_result(result_phi);
6381
6382 return true;
6383 }
6384
6385 //------------------------------inline_vectorizedHashcode----------------------------
6386 bool LibraryCallKit::inline_vectorizedHashCode() {
6387 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6388
6389 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6390 Node* array = argument(0);
6391 Node* offset = argument(1);
6392 Node* length = argument(2);
6393 Node* initialValue = argument(3);
6394 Node* basic_type = argument(4);
6395
6396 if (basic_type == top()) {
6397 return false; // failed input validation
6398 }
6399
6400 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6401 if (!basic_type_t->is_con()) {
6402 return false; // Only intrinsify if mode argument is constant
6403 }
6404
6405 array = must_be_not_null(array, true);
6406
6407 BasicType bt = (BasicType)basic_type_t->get_con();
6408
6409 // Resolve address of first element
6410 Node* array_start = array_element_address(array, offset, bt);
6411
6412 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6413 array_start, length, initialValue, basic_type)));
6414 clear_upper_avx();
6415
6416 return true;
6417 }
6418
6419 /**
6420 * Calculate CRC32 for byte.
6421 * int java.util.zip.CRC32.update(int crc, int b)
6422 */
6423 bool LibraryCallKit::inline_updateCRC32() {
6424 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6425 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6426 // no receiver since it is static method
6427 Node* crc = argument(0); // type: int
6428 Node* b = argument(1); // type: int
6429
6430 /*
6431 * int c = ~ crc;
6432 * b = timesXtoThe32[(b ^ c) & 0xFF];
6433 * b = b ^ (c >>> 8);
6434 * crc = ~b;
6435 */
6436
6437 Node* M1 = intcon(-1);
6438 crc = _gvn.transform(new XorINode(crc, M1));
6439 Node* result = _gvn.transform(new XorINode(crc, b));
6440 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6441
6442 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6443 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6444 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6445 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6446
6447 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6448 result = _gvn.transform(new XorINode(crc, result));
6449 result = _gvn.transform(new XorINode(result, M1));
6450 set_result(result);
6451 return true;
6452 }
6453
6454 /**
6455 * Calculate CRC32 for byte[] array.
6456 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6457 */
6458 bool LibraryCallKit::inline_updateBytesCRC32() {
6459 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6460 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6461 // no receiver since it is static method
6462 Node* crc = argument(0); // type: int
6463 Node* src = argument(1); // type: oop
6464 Node* offset = argument(2); // type: int
6465 Node* length = argument(3); // type: int
6466
6467 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6468 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6469 // failed array check
6470 return false;
6471 }
6472
6473 // Figure out the size and type of the elements we will be copying.
6474 BasicType src_elem = src_type->elem()->array_element_basic_type();
6475 if (src_elem != T_BYTE) {
6476 return false;
6477 }
6478
6479 // 'src_start' points to src array + scaled offset
6480 src = must_be_not_null(src, true);
6481 Node* src_start = array_element_address(src, offset, src_elem);
6482
6483 // We assume that range check is done by caller.
6484 // TODO: generate range check (offset+length < src.length) in debug VM.
6485
6486 // Call the stub.
6487 address stubAddr = StubRoutines::updateBytesCRC32();
6488 const char *stubName = "updateBytesCRC32";
6489
6490 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6491 stubAddr, stubName, TypePtr::BOTTOM,
6492 crc, src_start, length);
6493 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6494 set_result(result);
6495 return true;
6496 }
6497
6498 /**
6499 * Calculate CRC32 for ByteBuffer.
6500 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6501 */
6502 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6503 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6504 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6505 // no receiver since it is static method
6506 Node* crc = argument(0); // type: int
6507 Node* src = argument(1); // type: long
6508 Node* offset = argument(3); // type: int
6509 Node* length = argument(4); // type: int
6510
6511 src = ConvL2X(src); // adjust Java long to machine word
6512 Node* base = _gvn.transform(new CastX2PNode(src));
6513 offset = ConvI2X(offset);
6514
6515 // 'src_start' points to src array + scaled offset
6516 Node* src_start = basic_plus_adr(top(), base, offset);
6517
6518 // Call the stub.
6519 address stubAddr = StubRoutines::updateBytesCRC32();
6520 const char *stubName = "updateBytesCRC32";
6521
6522 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6523 stubAddr, stubName, TypePtr::BOTTOM,
6524 crc, src_start, length);
6525 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6526 set_result(result);
6527 return true;
6528 }
6529
6530 //------------------------------get_table_from_crc32c_class-----------------------
6531 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6532 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6533 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6534
6535 return table;
6536 }
6537
6538 //------------------------------inline_updateBytesCRC32C-----------------------
6539 //
6540 // Calculate CRC32C for byte[] array.
6541 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6542 //
6543 bool LibraryCallKit::inline_updateBytesCRC32C() {
6544 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6545 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6546 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6547 // no receiver since it is a static method
6548 Node* crc = argument(0); // type: int
6549 Node* src = argument(1); // type: oop
6550 Node* offset = argument(2); // type: int
6551 Node* end = argument(3); // type: int
6552
6553 Node* length = _gvn.transform(new SubINode(end, offset));
6554
6555 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6556 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6557 // failed array check
6558 return false;
6559 }
6560
6561 // Figure out the size and type of the elements we will be copying.
6562 BasicType src_elem = src_type->elem()->array_element_basic_type();
6563 if (src_elem != T_BYTE) {
6564 return false;
6565 }
6566
6567 // 'src_start' points to src array + scaled offset
6568 src = must_be_not_null(src, true);
6569 Node* src_start = array_element_address(src, offset, src_elem);
6570
6571 // static final int[] byteTable in class CRC32C
6572 Node* table = get_table_from_crc32c_class(callee()->holder());
6573 table = must_be_not_null(table, true);
6574 Node* table_start = array_element_address(table, intcon(0), T_INT);
6575
6576 // We assume that range check is done by caller.
6577 // TODO: generate range check (offset+length < src.length) in debug VM.
6578
6579 // Call the stub.
6580 address stubAddr = StubRoutines::updateBytesCRC32C();
6581 const char *stubName = "updateBytesCRC32C";
6582
6583 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6584 stubAddr, stubName, TypePtr::BOTTOM,
6585 crc, src_start, length, table_start);
6586 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6587 set_result(result);
6588 return true;
6589 }
6590
6591 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6592 //
6593 // Calculate CRC32C for DirectByteBuffer.
6594 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6595 //
6596 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6597 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6598 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6599 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6600 // no receiver since it is a static method
6601 Node* crc = argument(0); // type: int
6602 Node* src = argument(1); // type: long
6603 Node* offset = argument(3); // type: int
6604 Node* end = argument(4); // type: int
6605
6606 Node* length = _gvn.transform(new SubINode(end, offset));
6607
6608 src = ConvL2X(src); // adjust Java long to machine word
6609 Node* base = _gvn.transform(new CastX2PNode(src));
6610 offset = ConvI2X(offset);
6611
6612 // 'src_start' points to src array + scaled offset
6613 Node* src_start = basic_plus_adr(top(), base, offset);
6614
6615 // static final int[] byteTable in class CRC32C
6616 Node* table = get_table_from_crc32c_class(callee()->holder());
6617 table = must_be_not_null(table, true);
6618 Node* table_start = array_element_address(table, intcon(0), T_INT);
6619
6620 // Call the stub.
6621 address stubAddr = StubRoutines::updateBytesCRC32C();
6622 const char *stubName = "updateBytesCRC32C";
6623
6624 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6625 stubAddr, stubName, TypePtr::BOTTOM,
6626 crc, src_start, length, table_start);
6627 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6628 set_result(result);
6629 return true;
6630 }
6631
6632 //------------------------------inline_updateBytesAdler32----------------------
6633 //
6634 // Calculate Adler32 checksum for byte[] array.
6635 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6636 //
6637 bool LibraryCallKit::inline_updateBytesAdler32() {
6638 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6639 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6640 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6641 // no receiver since it is static method
6642 Node* crc = argument(0); // type: int
6643 Node* src = argument(1); // type: oop
6644 Node* offset = argument(2); // type: int
6645 Node* length = argument(3); // type: int
6646
6647 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6648 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6649 // failed array check
6650 return false;
6651 }
6652
6653 // Figure out the size and type of the elements we will be copying.
6654 BasicType src_elem = src_type->elem()->array_element_basic_type();
6655 if (src_elem != T_BYTE) {
6656 return false;
6657 }
6658
6659 // 'src_start' points to src array + scaled offset
6660 Node* src_start = array_element_address(src, offset, src_elem);
6661
6662 // We assume that range check is done by caller.
6663 // TODO: generate range check (offset+length < src.length) in debug VM.
6664
6665 // Call the stub.
6666 address stubAddr = StubRoutines::updateBytesAdler32();
6667 const char *stubName = "updateBytesAdler32";
6668
6669 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6670 stubAddr, stubName, TypePtr::BOTTOM,
6671 crc, src_start, length);
6672 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6673 set_result(result);
6674 return true;
6675 }
6676
6677 //------------------------------inline_updateByteBufferAdler32---------------
6678 //
6679 // Calculate Adler32 checksum for DirectByteBuffer.
6680 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6681 //
6682 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6683 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6684 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6685 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6686 // no receiver since it is static method
6687 Node* crc = argument(0); // type: int
6688 Node* src = argument(1); // type: long
6689 Node* offset = argument(3); // type: int
6690 Node* length = argument(4); // type: int
6691
6692 src = ConvL2X(src); // adjust Java long to machine word
6693 Node* base = _gvn.transform(new CastX2PNode(src));
6694 offset = ConvI2X(offset);
6695
6696 // 'src_start' points to src array + scaled offset
6697 Node* src_start = basic_plus_adr(top(), base, offset);
6698
6699 // Call the stub.
6700 address stubAddr = StubRoutines::updateBytesAdler32();
6701 const char *stubName = "updateBytesAdler32";
6702
6703 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6704 stubAddr, stubName, TypePtr::BOTTOM,
6705 crc, src_start, length);
6706
6707 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6708 set_result(result);
6709 return true;
6710 }
6711
6712 //----------------------------inline_reference_get----------------------------
6713 // public T java.lang.ref.Reference.get();
6714 bool LibraryCallKit::inline_reference_get() {
6715 const int referent_offset = java_lang_ref_Reference::referent_offset();
6716
6717 // Get the argument:
6718 Node* reference_obj = null_check_receiver();
6719 if (stopped()) return true;
6720
6721 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6722 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6723 decorators, /*is_static*/ false, nullptr);
6724 if (result == nullptr) return false;
6725
6726 // Add memory barrier to prevent commoning reads from this field
6727 // across safepoint since GC can change its value.
6728 insert_mem_bar(Op_MemBarCPUOrder);
6729
6730 set_result(result);
6731 return true;
6732 }
6733
6734 //----------------------------inline_reference_refersTo0----------------------------
6735 // bool java.lang.ref.Reference.refersTo0();
6736 // bool java.lang.ref.PhantomReference.refersTo0();
6737 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6738 // Get arguments:
6739 Node* reference_obj = null_check_receiver();
6740 Node* other_obj = argument(1);
6741 if (stopped()) return true;
6742
6743 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6744 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6745 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6746 decorators, /*is_static*/ false, nullptr);
6747 if (referent == nullptr) return false;
6748
6749 // Add memory barrier to prevent commoning reads from this field
6750 // across safepoint since GC can change its value.
6751 insert_mem_bar(Op_MemBarCPUOrder);
6752
6753 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6754 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6755 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6756
6757 RegionNode* region = new RegionNode(3);
6758 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6759
6760 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6761 region->init_req(1, if_true);
6762 phi->init_req(1, intcon(1));
6763
6764 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6765 region->init_req(2, if_false);
6766 phi->init_req(2, intcon(0));
6767
6768 set_control(_gvn.transform(region));
6769 record_for_igvn(region);
6770 set_result(_gvn.transform(phi));
6771 return true;
6772 }
6773
6774
6775 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6776 DecoratorSet decorators, bool is_static,
6777 ciInstanceKlass* fromKls) {
6778 if (fromKls == nullptr) {
6779 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6780 assert(tinst != nullptr, "obj is null");
6781 assert(tinst->is_loaded(), "obj is not loaded");
6782 fromKls = tinst->instance_klass();
6783 } else {
6784 assert(is_static, "only for static field access");
6785 }
6786 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6787 ciSymbol::make(fieldTypeString),
6788 is_static);
6789
6790 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6791 if (field == nullptr) return (Node *) nullptr;
6792
6793 if (is_static) {
6794 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6795 fromObj = makecon(tip);
6796 }
6797
6798 // Next code copied from Parse::do_get_xxx():
6799
6800 // Compute address and memory type.
6801 int offset = field->offset_in_bytes();
6802 bool is_vol = field->is_volatile();
6803 ciType* field_klass = field->type();
6804 assert(field_klass->is_loaded(), "should be loaded");
6805 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6806 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6807 BasicType bt = field->layout_type();
6808
6809 // Build the resultant type of the load
6810 const Type *type;
6811 if (bt == T_OBJECT) {
6812 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6813 } else {
6814 type = Type::get_const_basic_type(bt);
6815 }
6816
6817 if (is_vol) {
6818 decorators |= MO_SEQ_CST;
6819 }
6820
6821 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6822 }
6823
6824 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6825 bool is_exact /* true */, bool is_static /* false */,
6826 ciInstanceKlass * fromKls /* nullptr */) {
6827 if (fromKls == nullptr) {
6828 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6829 assert(tinst != nullptr, "obj is null");
6830 assert(tinst->is_loaded(), "obj is not loaded");
6831 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6832 fromKls = tinst->instance_klass();
6833 }
6834 else {
6835 assert(is_static, "only for static field access");
6836 }
6837 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6838 ciSymbol::make(fieldTypeString),
6839 is_static);
6840
6841 assert(field != nullptr, "undefined field");
6842 assert(!field->is_volatile(), "not defined for volatile fields");
6843
6844 if (is_static) {
6845 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6846 fromObj = makecon(tip);
6847 }
6848
6849 // Next code copied from Parse::do_get_xxx():
6850
6851 // Compute address and memory type.
6852 int offset = field->offset_in_bytes();
6853 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6854
6855 return adr;
6856 }
6857
6858 //------------------------------inline_aescrypt_Block-----------------------
6859 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6860 address stubAddr = nullptr;
6861 const char *stubName;
6862 assert(UseAES, "need AES instruction support");
6863
6864 switch(id) {
6865 case vmIntrinsics::_aescrypt_encryptBlock:
6866 stubAddr = StubRoutines::aescrypt_encryptBlock();
6867 stubName = "aescrypt_encryptBlock";
6868 break;
6869 case vmIntrinsics::_aescrypt_decryptBlock:
6870 stubAddr = StubRoutines::aescrypt_decryptBlock();
6871 stubName = "aescrypt_decryptBlock";
6872 break;
6873 default:
6874 break;
6875 }
6876 if (stubAddr == nullptr) return false;
6877
6878 Node* aescrypt_object = argument(0);
6879 Node* src = argument(1);
6880 Node* src_offset = argument(2);
6881 Node* dest = argument(3);
6882 Node* dest_offset = argument(4);
6883
6884 src = must_be_not_null(src, true);
6885 dest = must_be_not_null(dest, true);
6886
6887 // (1) src and dest are arrays.
6888 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6889 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6890 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6891 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6892
6893 // for the quick and dirty code we will skip all the checks.
6894 // we are just trying to get the call to be generated.
6895 Node* src_start = src;
6896 Node* dest_start = dest;
6897 if (src_offset != nullptr || dest_offset != nullptr) {
6898 assert(src_offset != nullptr && dest_offset != nullptr, "");
6899 src_start = array_element_address(src, src_offset, T_BYTE);
6900 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6901 }
6902
6903 // now need to get the start of its expanded key array
6904 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6905 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6906 if (k_start == nullptr) return false;
6907
6908 // Call the stub.
6909 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6910 stubAddr, stubName, TypePtr::BOTTOM,
6911 src_start, dest_start, k_start);
6912
6913 return true;
6914 }
6915
6916 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6917 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6918 address stubAddr = nullptr;
6919 const char *stubName = nullptr;
6920
6921 assert(UseAES, "need AES instruction support");
6922
6923 switch(id) {
6924 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6925 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6926 stubName = "cipherBlockChaining_encryptAESCrypt";
6927 break;
6928 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6929 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6930 stubName = "cipherBlockChaining_decryptAESCrypt";
6931 break;
6932 default:
6933 break;
6934 }
6935 if (stubAddr == nullptr) return false;
6936
6937 Node* cipherBlockChaining_object = argument(0);
6938 Node* src = argument(1);
6939 Node* src_offset = argument(2);
6940 Node* len = argument(3);
6941 Node* dest = argument(4);
6942 Node* dest_offset = argument(5);
6943
6944 src = must_be_not_null(src, false);
6945 dest = must_be_not_null(dest, false);
6946
6947 // (1) src and dest are arrays.
6948 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6949 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6950 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6951 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6952
6953 // checks are the responsibility of the caller
6954 Node* src_start = src;
6955 Node* dest_start = dest;
6956 if (src_offset != nullptr || dest_offset != nullptr) {
6957 assert(src_offset != nullptr && dest_offset != nullptr, "");
6958 src_start = array_element_address(src, src_offset, T_BYTE);
6959 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6960 }
6961
6962 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
6963 // (because of the predicated logic executed earlier).
6964 // so we cast it here safely.
6965 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6966
6967 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
6968 if (embeddedCipherObj == nullptr) return false;
6969
6970 // cast it to what we know it will be at runtime
6971 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
6972 assert(tinst != nullptr, "CBC obj is null");
6973 assert(tinst->is_loaded(), "CBC obj is not loaded");
6974 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
6975 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
6976
6977 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
6978 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
6979 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
6980 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
6981 aescrypt_object = _gvn.transform(aescrypt_object);
6982
6983 // we need to get the start of the aescrypt_object's expanded key array
6984 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6985 if (k_start == nullptr) return false;
6986
6987 // similarly, get the start address of the r vector
6988 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
6989 if (objRvec == nullptr) return false;
6990 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
6991
6992 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
6993 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
6994 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
6995 stubAddr, stubName, TypePtr::BOTTOM,
6996 src_start, dest_start, k_start, r_start, len);
6997
6998 // return cipher length (int)
6999 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7000 set_result(retvalue);
7001 return true;
7002 }
7003
7004 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7005 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7006 address stubAddr = nullptr;
7007 const char *stubName = nullptr;
7008
7009 assert(UseAES, "need AES instruction support");
7010
7011 switch (id) {
7012 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7013 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7014 stubName = "electronicCodeBook_encryptAESCrypt";
7015 break;
7016 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7017 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7018 stubName = "electronicCodeBook_decryptAESCrypt";
7019 break;
7020 default:
7021 break;
7022 }
7023
7024 if (stubAddr == nullptr) return false;
7025
7026 Node* electronicCodeBook_object = argument(0);
7027 Node* src = argument(1);
7028 Node* src_offset = argument(2);
7029 Node* len = argument(3);
7030 Node* dest = argument(4);
7031 Node* dest_offset = argument(5);
7032
7033 // (1) src and dest are arrays.
7034 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7035 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7036 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7037 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7038
7039 // checks are the responsibility of the caller
7040 Node* src_start = src;
7041 Node* dest_start = dest;
7042 if (src_offset != nullptr || dest_offset != nullptr) {
7043 assert(src_offset != nullptr && dest_offset != nullptr, "");
7044 src_start = array_element_address(src, src_offset, T_BYTE);
7045 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7046 }
7047
7048 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7049 // (because of the predicated logic executed earlier).
7050 // so we cast it here safely.
7051 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7052
7053 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7054 if (embeddedCipherObj == nullptr) return false;
7055
7056 // cast it to what we know it will be at runtime
7057 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7058 assert(tinst != nullptr, "ECB obj is null");
7059 assert(tinst->is_loaded(), "ECB obj is not loaded");
7060 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7061 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7062
7063 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7064 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7065 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7066 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7067 aescrypt_object = _gvn.transform(aescrypt_object);
7068
7069 // we need to get the start of the aescrypt_object's expanded key array
7070 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7071 if (k_start == nullptr) return false;
7072
7073 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7074 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7075 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7076 stubAddr, stubName, TypePtr::BOTTOM,
7077 src_start, dest_start, k_start, len);
7078
7079 // return cipher length (int)
7080 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7081 set_result(retvalue);
7082 return true;
7083 }
7084
7085 //------------------------------inline_counterMode_AESCrypt-----------------------
7086 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7087 assert(UseAES, "need AES instruction support");
7088 if (!UseAESCTRIntrinsics) return false;
7089
7090 address stubAddr = nullptr;
7091 const char *stubName = nullptr;
7092 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7093 stubAddr = StubRoutines::counterMode_AESCrypt();
7094 stubName = "counterMode_AESCrypt";
7095 }
7096 if (stubAddr == nullptr) return false;
7097
7098 Node* counterMode_object = argument(0);
7099 Node* src = argument(1);
7100 Node* src_offset = argument(2);
7101 Node* len = argument(3);
7102 Node* dest = argument(4);
7103 Node* dest_offset = argument(5);
7104
7105 // (1) src and dest are arrays.
7106 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7107 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7108 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7109 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7110
7111 // checks are the responsibility of the caller
7112 Node* src_start = src;
7113 Node* dest_start = dest;
7114 if (src_offset != nullptr || dest_offset != nullptr) {
7115 assert(src_offset != nullptr && dest_offset != nullptr, "");
7116 src_start = array_element_address(src, src_offset, T_BYTE);
7117 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7118 }
7119
7120 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7121 // (because of the predicated logic executed earlier).
7122 // so we cast it here safely.
7123 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7124 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7125 if (embeddedCipherObj == nullptr) return false;
7126 // cast it to what we know it will be at runtime
7127 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7128 assert(tinst != nullptr, "CTR obj is null");
7129 assert(tinst->is_loaded(), "CTR obj is not loaded");
7130 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7131 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7132 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7133 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7134 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7135 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7136 aescrypt_object = _gvn.transform(aescrypt_object);
7137 // we need to get the start of the aescrypt_object's expanded key array
7138 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7139 if (k_start == nullptr) return false;
7140 // similarly, get the start address of the r vector
7141 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7142 if (obj_counter == nullptr) return false;
7143 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7144
7145 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7146 if (saved_encCounter == nullptr) return false;
7147 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7148 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7149
7150 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7151 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7152 OptoRuntime::counterMode_aescrypt_Type(),
7153 stubAddr, stubName, TypePtr::BOTTOM,
7154 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7155
7156 // return cipher length (int)
7157 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7158 set_result(retvalue);
7159 return true;
7160 }
7161
7162 //------------------------------get_key_start_from_aescrypt_object-----------------------
7163 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7164 #if defined(PPC64) || defined(S390)
7165 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7166 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7167 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7168 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7169 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7170 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7171 if (objSessionK == nullptr) {
7172 return (Node *) nullptr;
7173 }
7174 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7175 #else
7176 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7177 #endif // PPC64
7178 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7179 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7180
7181 // now have the array, need to get the start address of the K array
7182 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7183 return k_start;
7184 }
7185
7186 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7187 // Return node representing slow path of predicate check.
7188 // the pseudo code we want to emulate with this predicate is:
7189 // for encryption:
7190 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7191 // for decryption:
7192 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7193 // note cipher==plain is more conservative than the original java code but that's OK
7194 //
7195 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7196 // The receiver was checked for null already.
7197 Node* objCBC = argument(0);
7198
7199 Node* src = argument(1);
7200 Node* dest = argument(4);
7201
7202 // Load embeddedCipher field of CipherBlockChaining object.
7203 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7204
7205 // get AESCrypt klass for instanceOf check
7206 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7207 // will have same classloader as CipherBlockChaining object
7208 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7209 assert(tinst != nullptr, "CBCobj is null");
7210 assert(tinst->is_loaded(), "CBCobj is not loaded");
7211
7212 // we want to do an instanceof comparison against the AESCrypt class
7213 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7214 if (!klass_AESCrypt->is_loaded()) {
7215 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7216 Node* ctrl = control();
7217 set_control(top()); // no regular fast path
7218 return ctrl;
7219 }
7220
7221 src = must_be_not_null(src, true);
7222 dest = must_be_not_null(dest, true);
7223
7224 // Resolve oops to stable for CmpP below.
7225 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7226
7227 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7228 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7229 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7230
7231 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7232
7233 // for encryption, we are done
7234 if (!decrypting)
7235 return instof_false; // even if it is null
7236
7237 // for decryption, we need to add a further check to avoid
7238 // taking the intrinsic path when cipher and plain are the same
7239 // see the original java code for why.
7240 RegionNode* region = new RegionNode(3);
7241 region->init_req(1, instof_false);
7242
7243 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7244 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7245 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7246 region->init_req(2, src_dest_conjoint);
7247
7248 record_for_igvn(region);
7249 return _gvn.transform(region);
7250 }
7251
7252 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7253 // Return node representing slow path of predicate check.
7254 // the pseudo code we want to emulate with this predicate is:
7255 // for encryption:
7256 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7257 // for decryption:
7258 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7259 // note cipher==plain is more conservative than the original java code but that's OK
7260 //
7261 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7262 // The receiver was checked for null already.
7263 Node* objECB = argument(0);
7264
7265 // Load embeddedCipher field of ElectronicCodeBook object.
7266 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7267
7268 // get AESCrypt klass for instanceOf check
7269 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7270 // will have same classloader as ElectronicCodeBook object
7271 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7272 assert(tinst != nullptr, "ECBobj is null");
7273 assert(tinst->is_loaded(), "ECBobj is not loaded");
7274
7275 // we want to do an instanceof comparison against the AESCrypt class
7276 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7277 if (!klass_AESCrypt->is_loaded()) {
7278 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7279 Node* ctrl = control();
7280 set_control(top()); // no regular fast path
7281 return ctrl;
7282 }
7283 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7284
7285 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7286 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7287 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7288
7289 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7290
7291 // for encryption, we are done
7292 if (!decrypting)
7293 return instof_false; // even if it is null
7294
7295 // for decryption, we need to add a further check to avoid
7296 // taking the intrinsic path when cipher and plain are the same
7297 // see the original java code for why.
7298 RegionNode* region = new RegionNode(3);
7299 region->init_req(1, instof_false);
7300 Node* src = argument(1);
7301 Node* dest = argument(4);
7302 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7303 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7304 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7305 region->init_req(2, src_dest_conjoint);
7306
7307 record_for_igvn(region);
7308 return _gvn.transform(region);
7309 }
7310
7311 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7312 // Return node representing slow path of predicate check.
7313 // the pseudo code we want to emulate with this predicate is:
7314 // for encryption:
7315 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7316 // for decryption:
7317 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7318 // note cipher==plain is more conservative than the original java code but that's OK
7319 //
7320
7321 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7322 // The receiver was checked for null already.
7323 Node* objCTR = argument(0);
7324
7325 // Load embeddedCipher field of CipherBlockChaining object.
7326 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7327
7328 // get AESCrypt klass for instanceOf check
7329 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7330 // will have same classloader as CipherBlockChaining object
7331 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7332 assert(tinst != nullptr, "CTRobj is null");
7333 assert(tinst->is_loaded(), "CTRobj is not loaded");
7334
7335 // we want to do an instanceof comparison against the AESCrypt class
7336 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7337 if (!klass_AESCrypt->is_loaded()) {
7338 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7339 Node* ctrl = control();
7340 set_control(top()); // no regular fast path
7341 return ctrl;
7342 }
7343
7344 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7345 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7346 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7347 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7348 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7349
7350 return instof_false; // even if it is null
7351 }
7352
7353 //------------------------------inline_ghash_processBlocks
7354 bool LibraryCallKit::inline_ghash_processBlocks() {
7355 address stubAddr;
7356 const char *stubName;
7357 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7358
7359 stubAddr = StubRoutines::ghash_processBlocks();
7360 stubName = "ghash_processBlocks";
7361
7362 Node* data = argument(0);
7363 Node* offset = argument(1);
7364 Node* len = argument(2);
7365 Node* state = argument(3);
7366 Node* subkeyH = argument(4);
7367
7368 state = must_be_not_null(state, true);
7369 subkeyH = must_be_not_null(subkeyH, true);
7370 data = must_be_not_null(data, true);
7371
7372 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7373 assert(state_start, "state is null");
7374 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7375 assert(subkeyH_start, "subkeyH is null");
7376 Node* data_start = array_element_address(data, offset, T_BYTE);
7377 assert(data_start, "data is null");
7378
7379 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7380 OptoRuntime::ghash_processBlocks_Type(),
7381 stubAddr, stubName, TypePtr::BOTTOM,
7382 state_start, subkeyH_start, data_start, len);
7383 return true;
7384 }
7385
7386 //------------------------------inline_chacha20Block
7387 bool LibraryCallKit::inline_chacha20Block() {
7388 address stubAddr;
7389 const char *stubName;
7390 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7391
7392 stubAddr = StubRoutines::chacha20Block();
7393 stubName = "chacha20Block";
7394
7395 Node* state = argument(0);
7396 Node* result = argument(1);
7397
7398 state = must_be_not_null(state, true);
7399 result = must_be_not_null(result, true);
7400
7401 Node* state_start = array_element_address(state, intcon(0), T_INT);
7402 assert(state_start, "state is null");
7403 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7404 assert(result_start, "result is null");
7405
7406 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7407 OptoRuntime::chacha20Block_Type(),
7408 stubAddr, stubName, TypePtr::BOTTOM,
7409 state_start, result_start);
7410 // return key stream length (int)
7411 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7412 set_result(retvalue);
7413 return true;
7414 }
7415
7416 bool LibraryCallKit::inline_base64_encodeBlock() {
7417 address stubAddr;
7418 const char *stubName;
7419 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7420 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7421 stubAddr = StubRoutines::base64_encodeBlock();
7422 stubName = "encodeBlock";
7423
7424 if (!stubAddr) return false;
7425 Node* base64obj = argument(0);
7426 Node* src = argument(1);
7427 Node* offset = argument(2);
7428 Node* len = argument(3);
7429 Node* dest = argument(4);
7430 Node* dp = argument(5);
7431 Node* isURL = argument(6);
7432
7433 src = must_be_not_null(src, true);
7434 dest = must_be_not_null(dest, true);
7435
7436 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7437 assert(src_start, "source array is null");
7438 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7439 assert(dest_start, "destination array is null");
7440
7441 Node* base64 = make_runtime_call(RC_LEAF,
7442 OptoRuntime::base64_encodeBlock_Type(),
7443 stubAddr, stubName, TypePtr::BOTTOM,
7444 src_start, offset, len, dest_start, dp, isURL);
7445 return true;
7446 }
7447
7448 bool LibraryCallKit::inline_base64_decodeBlock() {
7449 address stubAddr;
7450 const char *stubName;
7451 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7452 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7453 stubAddr = StubRoutines::base64_decodeBlock();
7454 stubName = "decodeBlock";
7455
7456 if (!stubAddr) return false;
7457 Node* base64obj = argument(0);
7458 Node* src = argument(1);
7459 Node* src_offset = argument(2);
7460 Node* len = argument(3);
7461 Node* dest = argument(4);
7462 Node* dest_offset = argument(5);
7463 Node* isURL = argument(6);
7464 Node* isMIME = argument(7);
7465
7466 src = must_be_not_null(src, true);
7467 dest = must_be_not_null(dest, true);
7468
7469 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7470 assert(src_start, "source array is null");
7471 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7472 assert(dest_start, "destination array is null");
7473
7474 Node* call = make_runtime_call(RC_LEAF,
7475 OptoRuntime::base64_decodeBlock_Type(),
7476 stubAddr, stubName, TypePtr::BOTTOM,
7477 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7478 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7479 set_result(result);
7480 return true;
7481 }
7482
7483 bool LibraryCallKit::inline_poly1305_processBlocks() {
7484 address stubAddr;
7485 const char *stubName;
7486 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7487 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7488 stubAddr = StubRoutines::poly1305_processBlocks();
7489 stubName = "poly1305_processBlocks";
7490
7491 if (!stubAddr) return false;
7492 null_check_receiver(); // null-check receiver
7493 if (stopped()) return true;
7494
7495 Node* input = argument(1);
7496 Node* input_offset = argument(2);
7497 Node* len = argument(3);
7498 Node* alimbs = argument(4);
7499 Node* rlimbs = argument(5);
7500
7501 input = must_be_not_null(input, true);
7502 alimbs = must_be_not_null(alimbs, true);
7503 rlimbs = must_be_not_null(rlimbs, true);
7504
7505 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7506 assert(input_start, "input array is null");
7507 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7508 assert(acc_start, "acc array is null");
7509 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7510 assert(r_start, "r array is null");
7511
7512 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7513 OptoRuntime::poly1305_processBlocks_Type(),
7514 stubAddr, stubName, TypePtr::BOTTOM,
7515 input_start, len, acc_start, r_start);
7516 return true;
7517 }
7518
7519 //------------------------------inline_digestBase_implCompress-----------------------
7520 //
7521 // Calculate MD5 for single-block byte[] array.
7522 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7523 //
7524 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7525 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7526 //
7527 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7528 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7529 //
7530 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7531 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7532 //
7533 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7534 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7535 //
7536 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7537 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7538
7539 Node* digestBase_obj = argument(0);
7540 Node* src = argument(1); // type oop
7541 Node* ofs = argument(2); // type int
7542
7543 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7544 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7545 // failed array check
7546 return false;
7547 }
7548 // Figure out the size and type of the elements we will be copying.
7549 BasicType src_elem = src_type->elem()->array_element_basic_type();
7550 if (src_elem != T_BYTE) {
7551 return false;
7552 }
7553 // 'src_start' points to src array + offset
7554 src = must_be_not_null(src, true);
7555 Node* src_start = array_element_address(src, ofs, src_elem);
7556 Node* state = nullptr;
7557 Node* block_size = nullptr;
7558 address stubAddr;
7559 const char *stubName;
7560
7561 switch(id) {
7562 case vmIntrinsics::_md5_implCompress:
7563 assert(UseMD5Intrinsics, "need MD5 instruction support");
7564 state = get_state_from_digest_object(digestBase_obj, T_INT);
7565 stubAddr = StubRoutines::md5_implCompress();
7566 stubName = "md5_implCompress";
7567 break;
7568 case vmIntrinsics::_sha_implCompress:
7569 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7570 state = get_state_from_digest_object(digestBase_obj, T_INT);
7571 stubAddr = StubRoutines::sha1_implCompress();
7572 stubName = "sha1_implCompress";
7573 break;
7574 case vmIntrinsics::_sha2_implCompress:
7575 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7576 state = get_state_from_digest_object(digestBase_obj, T_INT);
7577 stubAddr = StubRoutines::sha256_implCompress();
7578 stubName = "sha256_implCompress";
7579 break;
7580 case vmIntrinsics::_sha5_implCompress:
7581 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7582 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7583 stubAddr = StubRoutines::sha512_implCompress();
7584 stubName = "sha512_implCompress";
7585 break;
7586 case vmIntrinsics::_sha3_implCompress:
7587 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7588 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7589 stubAddr = StubRoutines::sha3_implCompress();
7590 stubName = "sha3_implCompress";
7591 block_size = get_block_size_from_digest_object(digestBase_obj);
7592 if (block_size == nullptr) return false;
7593 break;
7594 default:
7595 fatal_unexpected_iid(id);
7596 return false;
7597 }
7598 if (state == nullptr) return false;
7599
7600 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7601 if (stubAddr == nullptr) return false;
7602
7603 // Call the stub.
7604 Node* call;
7605 if (block_size == nullptr) {
7606 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7607 stubAddr, stubName, TypePtr::BOTTOM,
7608 src_start, state);
7609 } else {
7610 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7611 stubAddr, stubName, TypePtr::BOTTOM,
7612 src_start, state, block_size);
7613 }
7614
7615 return true;
7616 }
7617
7618 //------------------------------inline_digestBase_implCompressMB-----------------------
7619 //
7620 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7621 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7622 //
7623 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7624 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7625 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7626 assert((uint)predicate < 5, "sanity");
7627 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7628
7629 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7630 Node* src = argument(1); // byte[] array
7631 Node* ofs = argument(2); // type int
7632 Node* limit = argument(3); // type int
7633
7634 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7635 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7636 // failed array check
7637 return false;
7638 }
7639 // Figure out the size and type of the elements we will be copying.
7640 BasicType src_elem = src_type->elem()->array_element_basic_type();
7641 if (src_elem != T_BYTE) {
7642 return false;
7643 }
7644 // 'src_start' points to src array + offset
7645 src = must_be_not_null(src, false);
7646 Node* src_start = array_element_address(src, ofs, src_elem);
7647
7648 const char* klass_digestBase_name = nullptr;
7649 const char* stub_name = nullptr;
7650 address stub_addr = nullptr;
7651 BasicType elem_type = T_INT;
7652
7653 switch (predicate) {
7654 case 0:
7655 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7656 klass_digestBase_name = "sun/security/provider/MD5";
7657 stub_name = "md5_implCompressMB";
7658 stub_addr = StubRoutines::md5_implCompressMB();
7659 }
7660 break;
7661 case 1:
7662 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7663 klass_digestBase_name = "sun/security/provider/SHA";
7664 stub_name = "sha1_implCompressMB";
7665 stub_addr = StubRoutines::sha1_implCompressMB();
7666 }
7667 break;
7668 case 2:
7669 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7670 klass_digestBase_name = "sun/security/provider/SHA2";
7671 stub_name = "sha256_implCompressMB";
7672 stub_addr = StubRoutines::sha256_implCompressMB();
7673 }
7674 break;
7675 case 3:
7676 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7677 klass_digestBase_name = "sun/security/provider/SHA5";
7678 stub_name = "sha512_implCompressMB";
7679 stub_addr = StubRoutines::sha512_implCompressMB();
7680 elem_type = T_LONG;
7681 }
7682 break;
7683 case 4:
7684 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7685 klass_digestBase_name = "sun/security/provider/SHA3";
7686 stub_name = "sha3_implCompressMB";
7687 stub_addr = StubRoutines::sha3_implCompressMB();
7688 elem_type = T_BYTE;
7689 }
7690 break;
7691 default:
7692 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7693 }
7694 if (klass_digestBase_name != nullptr) {
7695 assert(stub_addr != nullptr, "Stub is generated");
7696 if (stub_addr == nullptr) return false;
7697
7698 // get DigestBase klass to lookup for SHA klass
7699 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7700 assert(tinst != nullptr, "digestBase_obj is not instance???");
7701 assert(tinst->is_loaded(), "DigestBase is not loaded");
7702
7703 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7704 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7705 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7706 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7707 }
7708 return false;
7709 }
7710
7711 //------------------------------inline_digestBase_implCompressMB-----------------------
7712 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7713 BasicType elem_type, address stubAddr, const char *stubName,
7714 Node* src_start, Node* ofs, Node* limit) {
7715 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7716 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7717 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7718 digest_obj = _gvn.transform(digest_obj);
7719
7720 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7721 if (state == nullptr) return false;
7722
7723 Node* block_size = nullptr;
7724 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7725 block_size = get_block_size_from_digest_object(digest_obj);
7726 if (block_size == nullptr) return false;
7727 }
7728
7729 // Call the stub.
7730 Node* call;
7731 if (block_size == nullptr) {
7732 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7733 OptoRuntime::digestBase_implCompressMB_Type(false),
7734 stubAddr, stubName, TypePtr::BOTTOM,
7735 src_start, state, ofs, limit);
7736 } else {
7737 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7738 OptoRuntime::digestBase_implCompressMB_Type(true),
7739 stubAddr, stubName, TypePtr::BOTTOM,
7740 src_start, state, block_size, ofs, limit);
7741 }
7742
7743 // return ofs (int)
7744 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7745 set_result(result);
7746
7747 return true;
7748 }
7749
7750 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7751 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7752 assert(UseAES, "need AES instruction support");
7753 address stubAddr = nullptr;
7754 const char *stubName = nullptr;
7755 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7756 stubName = "galoisCounterMode_AESCrypt";
7757
7758 if (stubAddr == nullptr) return false;
7759
7760 Node* in = argument(0);
7761 Node* inOfs = argument(1);
7762 Node* len = argument(2);
7763 Node* ct = argument(3);
7764 Node* ctOfs = argument(4);
7765 Node* out = argument(5);
7766 Node* outOfs = argument(6);
7767 Node* gctr_object = argument(7);
7768 Node* ghash_object = argument(8);
7769
7770 // (1) in, ct and out are arrays.
7771 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7772 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7773 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7774 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7775 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7776 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7777
7778 // checks are the responsibility of the caller
7779 Node* in_start = in;
7780 Node* ct_start = ct;
7781 Node* out_start = out;
7782 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7783 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7784 in_start = array_element_address(in, inOfs, T_BYTE);
7785 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7786 out_start = array_element_address(out, outOfs, T_BYTE);
7787 }
7788
7789 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7790 // (because of the predicated logic executed earlier).
7791 // so we cast it here safely.
7792 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7793 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7794 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7795 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7796 Node* state = load_field_from_object(ghash_object, "state", "[J");
7797
7798 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7799 return false;
7800 }
7801 // cast it to what we know it will be at runtime
7802 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7803 assert(tinst != nullptr, "GCTR obj is null");
7804 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7805 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7806 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7807 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7808 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7809 const TypeOopPtr* xtype = aklass->as_instance_type();
7810 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7811 aescrypt_object = _gvn.transform(aescrypt_object);
7812 // we need to get the start of the aescrypt_object's expanded key array
7813 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7814 if (k_start == nullptr) return false;
7815 // similarly, get the start address of the r vector
7816 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7817 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7818 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7819
7820
7821 // Call the stub, passing params
7822 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7823 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7824 stubAddr, stubName, TypePtr::BOTTOM,
7825 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7826
7827 // return cipher length (int)
7828 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7829 set_result(retvalue);
7830
7831 return true;
7832 }
7833
7834 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7835 // Return node representing slow path of predicate check.
7836 // the pseudo code we want to emulate with this predicate is:
7837 // for encryption:
7838 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7839 // for decryption:
7840 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7841 // note cipher==plain is more conservative than the original java code but that's OK
7842 //
7843
7844 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7845 // The receiver was checked for null already.
7846 Node* objGCTR = argument(7);
7847 // Load embeddedCipher field of GCTR object.
7848 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7849 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7850
7851 // get AESCrypt klass for instanceOf check
7852 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7853 // will have same classloader as CipherBlockChaining object
7854 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7855 assert(tinst != nullptr, "GCTR obj is null");
7856 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7857
7858 // we want to do an instanceof comparison against the AESCrypt class
7859 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7860 if (!klass_AESCrypt->is_loaded()) {
7861 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7862 Node* ctrl = control();
7863 set_control(top()); // no regular fast path
7864 return ctrl;
7865 }
7866
7867 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7868 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7869 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7870 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7871 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7872
7873 return instof_false; // even if it is null
7874 }
7875
7876 //------------------------------get_state_from_digest_object-----------------------
7877 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7878 const char* state_type;
7879 switch (elem_type) {
7880 case T_BYTE: state_type = "[B"; break;
7881 case T_INT: state_type = "[I"; break;
7882 case T_LONG: state_type = "[J"; break;
7883 default: ShouldNotReachHere();
7884 }
7885 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7886 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7887 if (digest_state == nullptr) return (Node *) nullptr;
7888
7889 // now have the array, need to get the start address of the state array
7890 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7891 return state;
7892 }
7893
7894 //------------------------------get_block_size_from_sha3_object----------------------------------
7895 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7896 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7897 assert (block_size != nullptr, "sanity");
7898 return block_size;
7899 }
7900
7901 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7902 // Return node representing slow path of predicate check.
7903 // the pseudo code we want to emulate with this predicate is:
7904 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7905 //
7906 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7907 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7908 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7909 assert((uint)predicate < 5, "sanity");
7910
7911 // The receiver was checked for null already.
7912 Node* digestBaseObj = argument(0);
7913
7914 // get DigestBase klass for instanceOf check
7915 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7916 assert(tinst != nullptr, "digestBaseObj is null");
7917 assert(tinst->is_loaded(), "DigestBase is not loaded");
7918
7919 const char* klass_name = nullptr;
7920 switch (predicate) {
7921 case 0:
7922 if (UseMD5Intrinsics) {
7923 // we want to do an instanceof comparison against the MD5 class
7924 klass_name = "sun/security/provider/MD5";
7925 }
7926 break;
7927 case 1:
7928 if (UseSHA1Intrinsics) {
7929 // we want to do an instanceof comparison against the SHA class
7930 klass_name = "sun/security/provider/SHA";
7931 }
7932 break;
7933 case 2:
7934 if (UseSHA256Intrinsics) {
7935 // we want to do an instanceof comparison against the SHA2 class
7936 klass_name = "sun/security/provider/SHA2";
7937 }
7938 break;
7939 case 3:
7940 if (UseSHA512Intrinsics) {
7941 // we want to do an instanceof comparison against the SHA5 class
7942 klass_name = "sun/security/provider/SHA5";
7943 }
7944 break;
7945 case 4:
7946 if (UseSHA3Intrinsics) {
7947 // we want to do an instanceof comparison against the SHA3 class
7948 klass_name = "sun/security/provider/SHA3";
7949 }
7950 break;
7951 default:
7952 fatal("unknown SHA intrinsic predicate: %d", predicate);
7953 }
7954
7955 ciKlass* klass = nullptr;
7956 if (klass_name != nullptr) {
7957 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
7958 }
7959 if ((klass == nullptr) || !klass->is_loaded()) {
7960 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
7961 Node* ctrl = control();
7962 set_control(top()); // no intrinsic path
7963 return ctrl;
7964 }
7965 ciInstanceKlass* instklass = klass->as_instance_klass();
7966
7967 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
7968 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7969 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7970 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7971
7972 return instof_false; // even if it is null
7973 }
7974
7975 //-------------inline_fma-----------------------------------
7976 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
7977 Node *a = nullptr;
7978 Node *b = nullptr;
7979 Node *c = nullptr;
7980 Node* result = nullptr;
7981 switch (id) {
7982 case vmIntrinsics::_fmaD:
7983 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
7984 // no receiver since it is static method
7985 a = round_double_node(argument(0));
7986 b = round_double_node(argument(2));
7987 c = round_double_node(argument(4));
7988 result = _gvn.transform(new FmaDNode(control(), a, b, c));
7989 break;
7990 case vmIntrinsics::_fmaF:
7991 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
7992 a = argument(0);
7993 b = argument(1);
7994 c = argument(2);
7995 result = _gvn.transform(new FmaFNode(control(), a, b, c));
7996 break;
7997 default:
7998 fatal_unexpected_iid(id); break;
7999 }
8000 set_result(result);
8001 return true;
8002 }
8003
8004 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8005 // argument(0) is receiver
8006 Node* codePoint = argument(1);
8007 Node* n = nullptr;
8008
8009 switch (id) {
8010 case vmIntrinsics::_isDigit :
8011 n = new DigitNode(control(), codePoint);
8012 break;
8013 case vmIntrinsics::_isLowerCase :
8014 n = new LowerCaseNode(control(), codePoint);
8015 break;
8016 case vmIntrinsics::_isUpperCase :
8017 n = new UpperCaseNode(control(), codePoint);
8018 break;
8019 case vmIntrinsics::_isWhitespace :
8020 n = new WhitespaceNode(control(), codePoint);
8021 break;
8022 default:
8023 fatal_unexpected_iid(id);
8024 }
8025
8026 set_result(_gvn.transform(n));
8027 return true;
8028 }
8029
8030 //------------------------------inline_fp_min_max------------------------------
8031 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8032 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8033
8034 // The intrinsic should be used only when the API branches aren't predictable,
8035 // the last one performing the most important comparison. The following heuristic
8036 // uses the branch statistics to eventually bail out if necessary.
8037
8038 ciMethodData *md = callee()->method_data();
8039
8040 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8041 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8042
8043 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8044 // Bail out if the call-site didn't contribute enough to the statistics.
8045 return false;
8046 }
8047
8048 uint taken = 0, not_taken = 0;
8049
8050 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8051 if (p->is_BranchData()) {
8052 taken = ((ciBranchData*)p)->taken();
8053 not_taken = ((ciBranchData*)p)->not_taken();
8054 }
8055 }
8056
8057 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8058 balance = balance < 0 ? -balance : balance;
8059 if ( balance > 0.2 ) {
8060 // Bail out if the most important branch is predictable enough.
8061 return false;
8062 }
8063 }
8064 */
8065
8066 Node *a = nullptr;
8067 Node *b = nullptr;
8068 Node *n = nullptr;
8069 switch (id) {
8070 case vmIntrinsics::_maxF:
8071 case vmIntrinsics::_minF:
8072 case vmIntrinsics::_maxF_strict:
8073 case vmIntrinsics::_minF_strict:
8074 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8075 a = argument(0);
8076 b = argument(1);
8077 break;
8078 case vmIntrinsics::_maxD:
8079 case vmIntrinsics::_minD:
8080 case vmIntrinsics::_maxD_strict:
8081 case vmIntrinsics::_minD_strict:
8082 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8083 a = round_double_node(argument(0));
8084 b = round_double_node(argument(2));
8085 break;
8086 default:
8087 fatal_unexpected_iid(id);
8088 break;
8089 }
8090 switch (id) {
8091 case vmIntrinsics::_maxF:
8092 case vmIntrinsics::_maxF_strict:
8093 n = new MaxFNode(a, b);
8094 break;
8095 case vmIntrinsics::_minF:
8096 case vmIntrinsics::_minF_strict:
8097 n = new MinFNode(a, b);
8098 break;
8099 case vmIntrinsics::_maxD:
8100 case vmIntrinsics::_maxD_strict:
8101 n = new MaxDNode(a, b);
8102 break;
8103 case vmIntrinsics::_minD:
8104 case vmIntrinsics::_minD_strict:
8105 n = new MinDNode(a, b);
8106 break;
8107 default:
8108 fatal_unexpected_iid(id);
8109 break;
8110 }
8111 set_result(_gvn.transform(n));
8112 return true;
8113 }
8114
8115 bool LibraryCallKit::inline_profileBoolean() {
8116 Node* counts = argument(1);
8117 const TypeAryPtr* ary = nullptr;
8118 ciArray* aobj = nullptr;
8119 if (counts->is_Con()
8120 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8121 && (aobj = ary->const_oop()->as_array()) != nullptr
8122 && (aobj->length() == 2)) {
8123 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8124 jint false_cnt = aobj->element_value(0).as_int();
8125 jint true_cnt = aobj->element_value(1).as_int();
8126
8127 if (C->log() != nullptr) {
8128 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8129 false_cnt, true_cnt);
8130 }
8131
8132 if (false_cnt + true_cnt == 0) {
8133 // According to profile, never executed.
8134 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8135 Deoptimization::Action_reinterpret);
8136 return true;
8137 }
8138
8139 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8140 // is a number of each value occurrences.
8141 Node* result = argument(0);
8142 if (false_cnt == 0 || true_cnt == 0) {
8143 // According to profile, one value has been never seen.
8144 int expected_val = (false_cnt == 0) ? 1 : 0;
8145
8146 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8147 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8148
8149 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8150 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8151 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8152
8153 { // Slow path: uncommon trap for never seen value and then reexecute
8154 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8155 // the value has been seen at least once.
8156 PreserveJVMState pjvms(this);
8157 PreserveReexecuteState preexecs(this);
8158 jvms()->set_should_reexecute(true);
8159
8160 set_control(slow_path);
8161 set_i_o(i_o());
8162
8163 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8164 Deoptimization::Action_reinterpret);
8165 }
8166 // The guard for never seen value enables sharpening of the result and
8167 // returning a constant. It allows to eliminate branches on the same value
8168 // later on.
8169 set_control(fast_path);
8170 result = intcon(expected_val);
8171 }
8172 // Stop profiling.
8173 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8174 // By replacing method body with profile data (represented as ProfileBooleanNode
8175 // on IR level) we effectively disable profiling.
8176 // It enables full speed execution once optimized code is generated.
8177 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8178 C->record_for_igvn(profile);
8179 set_result(profile);
8180 return true;
8181 } else {
8182 // Continue profiling.
8183 // Profile data isn't available at the moment. So, execute method's bytecode version.
8184 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8185 // is compiled and counters aren't available since corresponding MethodHandle
8186 // isn't a compile-time constant.
8187 return false;
8188 }
8189 }
8190
8191 bool LibraryCallKit::inline_isCompileConstant() {
8192 Node* n = argument(0);
8193 set_result(n->is_Con() ? intcon(1) : intcon(0));
8194 return true;
8195 }
8196
8197 //------------------------------- inline_getObjectSize --------------------------------------
8198 //
8199 // Calculate the runtime size of the object/array.
8200 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8201 //
8202 bool LibraryCallKit::inline_getObjectSize() {
8203 Node* obj = argument(3);
8204 Node* klass_node = load_object_klass(obj);
8205
8206 jint layout_con = Klass::_lh_neutral_value;
8207 Node* layout_val = get_layout_helper(klass_node, layout_con);
8208 int layout_is_con = (layout_val == nullptr);
8209
8210 if (layout_is_con) {
8211 // Layout helper is constant, can figure out things at compile time.
8212
8213 if (Klass::layout_helper_is_instance(layout_con)) {
8214 // Instance case: layout_con contains the size itself.
8215 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8216 set_result(size);
8217 } else {
8218 // Array case: size is round(header + element_size*arraylength).
8219 // Since arraylength is different for every array instance, we have to
8220 // compute the whole thing at runtime.
8221
8222 Node* arr_length = load_array_length(obj);
8223
8224 int round_mask = MinObjAlignmentInBytes - 1;
8225 int hsize = Klass::layout_helper_header_size(layout_con);
8226 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8227
8228 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8229 round_mask = 0; // strength-reduce it if it goes away completely
8230 }
8231 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8232 Node* header_size = intcon(hsize + round_mask);
8233
8234 Node* lengthx = ConvI2X(arr_length);
8235 Node* headerx = ConvI2X(header_size);
8236
8237 Node* abody = lengthx;
8238 if (eshift != 0) {
8239 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8240 }
8241 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8242 if (round_mask != 0) {
8243 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8244 }
8245 size = ConvX2L(size);
8246 set_result(size);
8247 }
8248 } else {
8249 // Layout helper is not constant, need to test for array-ness at runtime.
8250
8251 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8252 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8253 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8254 record_for_igvn(result_reg);
8255
8256 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8257 if (array_ctl != nullptr) {
8258 // Array case: size is round(header + element_size*arraylength).
8259 // Since arraylength is different for every array instance, we have to
8260 // compute the whole thing at runtime.
8261
8262 PreserveJVMState pjvms(this);
8263 set_control(array_ctl);
8264 Node* arr_length = load_array_length(obj);
8265
8266 int round_mask = MinObjAlignmentInBytes - 1;
8267 Node* mask = intcon(round_mask);
8268
8269 Node* hss = intcon(Klass::_lh_header_size_shift);
8270 Node* hsm = intcon(Klass::_lh_header_size_mask);
8271 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8272 header_size = _gvn.transform(new AndINode(header_size, hsm));
8273 header_size = _gvn.transform(new AddINode(header_size, mask));
8274
8275 // There is no need to mask or shift this value.
8276 // The semantics of LShiftINode include an implicit mask to 0x1F.
8277 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8278 Node* elem_shift = layout_val;
8279
8280 Node* lengthx = ConvI2X(arr_length);
8281 Node* headerx = ConvI2X(header_size);
8282
8283 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8284 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8285 if (round_mask != 0) {
8286 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8287 }
8288 size = ConvX2L(size);
8289
8290 result_reg->init_req(_array_path, control());
8291 result_val->init_req(_array_path, size);
8292 }
8293
8294 if (!stopped()) {
8295 // Instance case: the layout helper gives us instance size almost directly,
8296 // but we need to mask out the _lh_instance_slow_path_bit.
8297 Node* size = ConvI2X(layout_val);
8298 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8299 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8300 size = _gvn.transform(new AndXNode(size, mask));
8301 size = ConvX2L(size);
8302
8303 result_reg->init_req(_instance_path, control());
8304 result_val->init_req(_instance_path, size);
8305 }
8306
8307 set_result(result_reg, result_val);
8308 }
8309
8310 return true;
8311 }
8312
8313 //------------------------------- inline_blackhole --------------------------------------
8314 //
8315 // Make sure all arguments to this node are alive.
8316 // This matches methods that were requested to be blackholed through compile commands.
8317 //
8318 bool LibraryCallKit::inline_blackhole() {
8319 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8320 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8321 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8322
8323 // Blackhole node pinches only the control, not memory. This allows
8324 // the blackhole to be pinned in the loop that computes blackholed
8325 // values, but have no other side effects, like breaking the optimizations
8326 // across the blackhole.
8327
8328 Node* bh = _gvn.transform(new BlackholeNode(control()));
8329 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8330
8331 // Bind call arguments as blackhole arguments to keep them alive
8332 uint nargs = callee()->arg_size();
8333 for (uint i = 0; i < nargs; i++) {
8334 bh->add_req(argument(i));
8335 }
8336
8337 return true;
8338 }