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