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/ciFlatArrayKlass.hpp"
28 #include "ci/ciUtilities.inline.hpp"
29 #include "classfile/vmIntrinsics.hpp"
30 #include "compiler/compileBroker.hpp"
31 #include "compiler/compileLog.hpp"
32 #include "gc/shared/barrierSet.hpp"
33 #include "jfr/support/jfrIntrinsics.hpp"
34 #include "memory/resourceArea.hpp"
35 #include "oops/klass.inline.hpp"
36 #include "oops/objArrayKlass.hpp"
37 #include "opto/addnode.hpp"
38 #include "opto/arraycopynode.hpp"
39 #include "opto/c2compiler.hpp"
40 #include "opto/castnode.hpp"
41 #include "opto/cfgnode.hpp"
42 #include "opto/convertnode.hpp"
43 #include "opto/countbitsnode.hpp"
44 #include "opto/idealKit.hpp"
45 #include "opto/library_call.hpp"
46 #include "opto/mathexactnode.hpp"
47 #include "opto/mulnode.hpp"
48 #include "opto/narrowptrnode.hpp"
49 #include "opto/opaquenode.hpp"
50 #include "opto/parse.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/rootnode.hpp"
53 #include "opto/subnode.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, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, 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, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, 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, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, 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, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, 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::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
297 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
298 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
299 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
300
301 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
302 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
303 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
304 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
308 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
309
310 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
311 case vmIntrinsics::_equalsU: return inline_string_equals(StrIntrinsicNode::UU);
312
313 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
314
315 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
316 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
317 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
318 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
319
320 case vmIntrinsics::_compressStringC:
321 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
322 case vmIntrinsics::_inflateStringC:
323 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
324
325 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
326 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
327 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
328 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
329 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
330 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
331 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
332 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
333 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
334 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
335 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
336 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_PRIMITIVE_OBJECT,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 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_PRIMITIVE_OBJECT,Relaxed, false);
348
349 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
350 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
351 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
352 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
353 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
354 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
355 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
356 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
357 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
358
359 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
360 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
361 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
362 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
363 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
364 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
365 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
366 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
367 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
368
369 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
370 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
371 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
372 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
373
374 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
375 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
376 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
377 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
378
379 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
380 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
381 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
382 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
383 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
384 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
385 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
386 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
387 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
388
389 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
390 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
391 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
392 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
393 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
394 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
395 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
396 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
397 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
398
399 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
400 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
401 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
402 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
403 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
404 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
405 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
406 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
407 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
408
409 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
410 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
411 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
412 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
413 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
414 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
415 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
416 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
417 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
418
419 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
423 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
424
425 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
426 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
427 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
428 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
429 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
430 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
431 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
432 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
433 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
434 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
435 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
436 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
437 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
438 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
439 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
440 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
441 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
442 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
443 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
444 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
445
446 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
447 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
448 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
449 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
450 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
451 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
452 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
453 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
454 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
455 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
456 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
457 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
458 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
459 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
460 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
461
462 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
465 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
466
467 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
471 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
472
473 case vmIntrinsics::_loadFence:
474 case vmIntrinsics::_storeFence:
475 case vmIntrinsics::_storeStoreFence:
476 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
477
478 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
479
480 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
481 case vmIntrinsics::_currentThread: return inline_native_currentThread();
482 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
483
484 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
485 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
486
487 #if INCLUDE_JVMTI
488 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
489 "notifyJvmtiStart", true, false);
490 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
491 "notifyJvmtiEnd", false, true);
492 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
493 "notifyJvmtiMount", false, false);
494 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
495 "notifyJvmtiUnmount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
497 #endif
498
499 #ifdef JFR_HAVE_INTRINSICS
500 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
501 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
502 #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::_asPrimaryType:
534 case vmIntrinsics::_asPrimaryTypeArg:
535 case vmIntrinsics::_asValueType:
536 case vmIntrinsics::_asValueTypeArg: return inline_primitive_Class_conversion(intrinsic_id());
537
538 case vmIntrinsics::_floatToRawIntBits:
539 case vmIntrinsics::_floatToIntBits:
540 case vmIntrinsics::_intBitsToFloat:
541 case vmIntrinsics::_doubleToRawLongBits:
542 case vmIntrinsics::_doubleToLongBits:
543 case vmIntrinsics::_longBitsToDouble:
544 case vmIntrinsics::_floatToFloat16:
545 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
546
547 case vmIntrinsics::_floatIsFinite:
548 case vmIntrinsics::_floatIsInfinite:
549 case vmIntrinsics::_doubleIsFinite:
550 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
551
552 case vmIntrinsics::_numberOfLeadingZeros_i:
553 case vmIntrinsics::_numberOfLeadingZeros_l:
554 case vmIntrinsics::_numberOfTrailingZeros_i:
555 case vmIntrinsics::_numberOfTrailingZeros_l:
556 case vmIntrinsics::_bitCount_i:
557 case vmIntrinsics::_bitCount_l:
558 case vmIntrinsics::_reverse_i:
559 case vmIntrinsics::_reverse_l:
560 case vmIntrinsics::_reverseBytes_i:
561 case vmIntrinsics::_reverseBytes_l:
562 case vmIntrinsics::_reverseBytes_s:
563 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
564
565 case vmIntrinsics::_compress_i:
566 case vmIntrinsics::_compress_l:
567 case vmIntrinsics::_expand_i:
568 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
569
570 case vmIntrinsics::_compareUnsigned_i:
571 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
572
573 case vmIntrinsics::_divideUnsigned_i:
574 case vmIntrinsics::_divideUnsigned_l:
575 case vmIntrinsics::_remainderUnsigned_i:
576 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
577
578 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
579
580 case vmIntrinsics::_Reference_get: return inline_reference_get();
581 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
582 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
583
584 case vmIntrinsics::_Class_cast: return inline_Class_cast();
585
586 case vmIntrinsics::_aescrypt_encryptBlock:
587 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
588
589 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
590 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
591 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
592
593 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
594 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
595 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
596
597 case vmIntrinsics::_counterMode_AESCrypt:
598 return inline_counterMode_AESCrypt(intrinsic_id());
599
600 case vmIntrinsics::_galoisCounterMode_AESCrypt:
601 return inline_galoisCounterMode_AESCrypt();
602
603 case vmIntrinsics::_md5_implCompress:
604 case vmIntrinsics::_sha_implCompress:
605 case vmIntrinsics::_sha2_implCompress:
606 case vmIntrinsics::_sha5_implCompress:
607 case vmIntrinsics::_sha3_implCompress:
608 return inline_digestBase_implCompress(intrinsic_id());
609
610 case vmIntrinsics::_digestBase_implCompressMB:
611 return inline_digestBase_implCompressMB(predicate);
612
613 case vmIntrinsics::_multiplyToLen:
614 return inline_multiplyToLen();
615
616 case vmIntrinsics::_squareToLen:
617 return inline_squareToLen();
618
619 case vmIntrinsics::_mulAdd:
620 return inline_mulAdd();
621
622 case vmIntrinsics::_montgomeryMultiply:
623 return inline_montgomeryMultiply();
624 case vmIntrinsics::_montgomerySquare:
625 return inline_montgomerySquare();
626
627 case vmIntrinsics::_bigIntegerRightShiftWorker:
628 return inline_bigIntegerShift(true);
629 case vmIntrinsics::_bigIntegerLeftShiftWorker:
630 return inline_bigIntegerShift(false);
631
632 case vmIntrinsics::_vectorizedMismatch:
633 return inline_vectorizedMismatch();
634
635 case vmIntrinsics::_ghash_processBlocks:
636 return inline_ghash_processBlocks();
637 case vmIntrinsics::_chacha20Block:
638 return inline_chacha20Block();
639 case vmIntrinsics::_base64_encodeBlock:
640 return inline_base64_encodeBlock();
641 case vmIntrinsics::_base64_decodeBlock:
642 return inline_base64_decodeBlock();
643 case vmIntrinsics::_poly1305_processBlocks:
644 return inline_poly1305_processBlocks();
645
646 case vmIntrinsics::_encodeISOArray:
647 case vmIntrinsics::_encodeByteISOArray:
648 return inline_encodeISOArray(false);
649 case vmIntrinsics::_encodeAsciiArray:
650 return inline_encodeISOArray(true);
651
652 case vmIntrinsics::_updateCRC32:
653 return inline_updateCRC32();
654 case vmIntrinsics::_updateBytesCRC32:
655 return inline_updateBytesCRC32();
656 case vmIntrinsics::_updateByteBufferCRC32:
657 return inline_updateByteBufferCRC32();
658
659 case vmIntrinsics::_updateBytesCRC32C:
660 return inline_updateBytesCRC32C();
661 case vmIntrinsics::_updateDirectByteBufferCRC32C:
662 return inline_updateDirectByteBufferCRC32C();
663
664 case vmIntrinsics::_updateBytesAdler32:
665 return inline_updateBytesAdler32();
666 case vmIntrinsics::_updateByteBufferAdler32:
667 return inline_updateByteBufferAdler32();
668
669 case vmIntrinsics::_profileBoolean:
670 return inline_profileBoolean();
671 case vmIntrinsics::_isCompileConstant:
672 return inline_isCompileConstant();
673
674 case vmIntrinsics::_countPositives:
675 return inline_countPositives();
676
677 case vmIntrinsics::_fmaD:
678 case vmIntrinsics::_fmaF:
679 return inline_fma(intrinsic_id());
680
681 case vmIntrinsics::_isDigit:
682 case vmIntrinsics::_isLowerCase:
683 case vmIntrinsics::_isUpperCase:
684 case vmIntrinsics::_isWhitespace:
685 return inline_character_compare(intrinsic_id());
686
687 case vmIntrinsics::_min:
688 case vmIntrinsics::_max:
689 case vmIntrinsics::_min_strict:
690 case vmIntrinsics::_max_strict:
691 return inline_min_max(intrinsic_id());
692
693 case vmIntrinsics::_maxF:
694 case vmIntrinsics::_minF:
695 case vmIntrinsics::_maxD:
696 case vmIntrinsics::_minD:
697 case vmIntrinsics::_maxF_strict:
698 case vmIntrinsics::_minF_strict:
699 case vmIntrinsics::_maxD_strict:
700 case vmIntrinsics::_minD_strict:
701 return inline_fp_min_max(intrinsic_id());
702
703 case vmIntrinsics::_VectorUnaryOp:
704 return inline_vector_nary_operation(1);
705 case vmIntrinsics::_VectorBinaryOp:
706 return inline_vector_nary_operation(2);
707 case vmIntrinsics::_VectorTernaryOp:
708 return inline_vector_nary_operation(3);
709 case vmIntrinsics::_VectorFromBitsCoerced:
710 return inline_vector_frombits_coerced();
711 case vmIntrinsics::_VectorMaskOp:
712 return inline_vector_mask_operation();
713 case vmIntrinsics::_VectorLoadOp:
714 return inline_vector_mem_operation(/*is_store=*/false);
715 case vmIntrinsics::_VectorLoadMaskedOp:
716 return inline_vector_mem_masked_operation(/*is_store*/false);
717 case vmIntrinsics::_VectorStoreOp:
718 return inline_vector_mem_operation(/*is_store=*/true);
719 case vmIntrinsics::_VectorStoreMaskedOp:
720 return inline_vector_mem_masked_operation(/*is_store=*/true);
721 case vmIntrinsics::_VectorGatherOp:
722 return inline_vector_gather_scatter(/*is_scatter*/ false);
723 case vmIntrinsics::_VectorScatterOp:
724 return inline_vector_gather_scatter(/*is_scatter*/ true);
725 case vmIntrinsics::_VectorReductionCoerced:
726 return inline_vector_reduction();
727 case vmIntrinsics::_VectorTest:
728 return inline_vector_test();
729 case vmIntrinsics::_VectorBlend:
730 return inline_vector_blend();
731 case vmIntrinsics::_VectorRearrange:
732 return inline_vector_rearrange();
733 case vmIntrinsics::_VectorCompare:
734 return inline_vector_compare();
735 case vmIntrinsics::_VectorBroadcastInt:
736 return inline_vector_broadcast_int();
737 case vmIntrinsics::_VectorConvert:
738 return inline_vector_convert();
739 case vmIntrinsics::_VectorInsert:
740 return inline_vector_insert();
741 case vmIntrinsics::_VectorExtract:
742 return inline_vector_extract();
743 case vmIntrinsics::_VectorCompressExpand:
744 return inline_vector_compress_expand();
745 case vmIntrinsics::_IndexVector:
746 return inline_index_vector();
747 case vmIntrinsics::_IndexPartiallyInUpperRange:
748 return inline_index_partially_in_upper_range();
749
750 case vmIntrinsics::_getObjectSize:
751 return inline_getObjectSize();
752
753 case vmIntrinsics::_blackhole:
754 return inline_blackhole();
755
756 default:
757 // If you get here, it may be that someone has added a new intrinsic
758 // to the list in vmIntrinsics.hpp without implementing it here.
759 #ifndef PRODUCT
760 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
761 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
762 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
763 }
764 #endif
765 return false;
766 }
767 }
768
769 Node* LibraryCallKit::try_to_predicate(int predicate) {
770 if (!jvms()->has_method()) {
771 // Root JVMState has a null method.
772 assert(map()->memory()->Opcode() == Op_Parm, "");
773 // Insert the memory aliasing node
774 set_all_memory(reset_memory());
775 }
776 assert(merged_memory(), "");
777
778 switch (intrinsic_id()) {
779 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
780 return inline_cipherBlockChaining_AESCrypt_predicate(false);
781 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
782 return inline_cipherBlockChaining_AESCrypt_predicate(true);
783 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
784 return inline_electronicCodeBook_AESCrypt_predicate(false);
785 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
786 return inline_electronicCodeBook_AESCrypt_predicate(true);
787 case vmIntrinsics::_counterMode_AESCrypt:
788 return inline_counterMode_AESCrypt_predicate();
789 case vmIntrinsics::_digestBase_implCompressMB:
790 return inline_digestBase_implCompressMB_predicate(predicate);
791 case vmIntrinsics::_galoisCounterMode_AESCrypt:
792 return inline_galoisCounterMode_AESCrypt_predicate();
793
794 default:
795 // If you get here, it may be that someone has added a new intrinsic
796 // to the list in vmIntrinsics.hpp without implementing it here.
797 #ifndef PRODUCT
798 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
799 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
800 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
801 }
802 #endif
803 Node* slow_ctl = control();
804 set_control(top()); // No fast path intrinsic
805 return slow_ctl;
806 }
807 }
808
809 //------------------------------set_result-------------------------------
810 // Helper function for finishing intrinsics.
811 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
812 record_for_igvn(region);
813 set_control(_gvn.transform(region));
814 set_result( _gvn.transform(value));
815 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
816 }
817
818 //------------------------------generate_guard---------------------------
819 // Helper function for generating guarded fast-slow graph structures.
820 // The given 'test', if true, guards a slow path. If the test fails
821 // then a fast path can be taken. (We generally hope it fails.)
822 // In all cases, GraphKit::control() is updated to the fast path.
823 // The returned value represents the control for the slow path.
824 // The return value is never 'top'; it is either a valid control
825 // or null if it is obvious that the slow path can never be taken.
826 // Also, if region and the slow control are not null, the slow edge
827 // is appended to the region.
828 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
829 if (stopped()) {
830 // Already short circuited.
831 return nullptr;
832 }
833
834 // Build an if node and its projections.
835 // If test is true we take the slow path, which we assume is uncommon.
836 if (_gvn.type(test) == TypeInt::ZERO) {
837 // The slow branch is never taken. No need to build this guard.
838 return nullptr;
839 }
840
841 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
842
843 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
844 if (if_slow == top()) {
845 // The slow branch is never taken. No need to build this guard.
846 return nullptr;
847 }
848
849 if (region != nullptr)
850 region->add_req(if_slow);
851
852 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
853 set_control(if_fast);
854
855 return if_slow;
856 }
857
858 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
859 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
860 }
861 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
862 return generate_guard(test, region, PROB_FAIR);
863 }
864
865 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
866 Node* *pos_index) {
867 if (stopped())
868 return nullptr; // already stopped
869 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
870 return nullptr; // index is already adequately typed
871 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
872 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
873 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
874 if (is_neg != nullptr && pos_index != nullptr) {
875 // Emulate effect of Parse::adjust_map_after_if.
876 Node* ccast = new CastIINode(index, TypeInt::POS);
877 ccast->set_req(0, control());
878 (*pos_index) = _gvn.transform(ccast);
879 }
880 return is_neg;
881 }
882
883 // Make sure that 'position' is a valid limit index, in [0..length].
884 // There are two equivalent plans for checking this:
885 // A. (offset + copyLength) unsigned<= arrayLength
886 // B. offset <= (arrayLength - copyLength)
887 // We require that all of the values above, except for the sum and
888 // difference, are already known to be non-negative.
889 // Plan A is robust in the face of overflow, if offset and copyLength
890 // are both hugely positive.
891 //
892 // Plan B is less direct and intuitive, but it does not overflow at
893 // all, since the difference of two non-negatives is always
894 // representable. Whenever Java methods must perform the equivalent
895 // check they generally use Plan B instead of Plan A.
896 // For the moment we use Plan A.
897 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
898 Node* subseq_length,
899 Node* array_length,
900 RegionNode* region) {
901 if (stopped())
902 return nullptr; // already stopped
903 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
904 if (zero_offset && subseq_length->eqv_uncast(array_length))
905 return nullptr; // common case of whole-array copy
906 Node* last = subseq_length;
907 if (!zero_offset) // last += offset
908 last = _gvn.transform(new AddINode(last, offset));
909 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
910 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
911 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
912 return is_over;
913 }
914
915 // Emit range checks for the given String.value byte array
916 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
917 if (stopped()) {
918 return; // already stopped
919 }
920 RegionNode* bailout = new RegionNode(1);
921 record_for_igvn(bailout);
922 if (char_count) {
923 // Convert char count to byte count
924 count = _gvn.transform(new LShiftINode(count, intcon(1)));
925 }
926
927 // Offset and count must not be negative
928 generate_negative_guard(offset, bailout);
929 generate_negative_guard(count, bailout);
930 // Offset + count must not exceed length of array
931 generate_limit_guard(offset, count, load_array_length(array), bailout);
932
933 if (bailout->req() > 1) {
934 PreserveJVMState pjvms(this);
935 set_control(_gvn.transform(bailout));
936 uncommon_trap(Deoptimization::Reason_intrinsic,
937 Deoptimization::Action_maybe_recompile);
938 }
939 }
940
941 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
942 bool is_immutable) {
943 ciKlass* thread_klass = env()->Thread_klass();
944 const Type* thread_type
945 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
946
947 Node* thread = _gvn.transform(new ThreadLocalNode());
948 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
949 tls_output = thread;
950
951 Node* thread_obj_handle
952 = (is_immutable
953 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
954 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
955 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
956 thread_obj_handle = _gvn.transform(thread_obj_handle);
957
958 DecoratorSet decorators = IN_NATIVE;
959 if (is_immutable) {
960 decorators |= C2_IMMUTABLE_MEMORY;
961 }
962 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
963 }
964
965 //--------------------------generate_current_thread--------------------
966 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
967 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
968 /*is_immutable*/false);
969 }
970
971 //--------------------------generate_virtual_thread--------------------
972 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
973 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
974 !C->method()->changes_current_thread());
975 }
976
977 //------------------------------make_string_method_node------------------------
978 // Helper method for String intrinsic functions. This version is called with
979 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
980 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
981 // containing the lengths of str1 and str2.
982 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
983 Node* result = nullptr;
984 switch (opcode) {
985 case Op_StrIndexOf:
986 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
987 str1_start, cnt1, str2_start, cnt2, ae);
988 break;
989 case Op_StrComp:
990 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
991 str1_start, cnt1, str2_start, cnt2, ae);
992 break;
993 case Op_StrEquals:
994 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
995 // Use the constant length if there is one because optimized match rule may exist.
996 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
997 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
998 break;
999 default:
1000 ShouldNotReachHere();
1001 return nullptr;
1002 }
1003
1004 // All these intrinsics have checks.
1005 C->set_has_split_ifs(true); // Has chance for split-if optimization
1006 clear_upper_avx();
1007
1008 return _gvn.transform(result);
1009 }
1010
1011 //------------------------------inline_string_compareTo------------------------
1012 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1013 Node* arg1 = argument(0);
1014 Node* arg2 = argument(1);
1015
1016 arg1 = must_be_not_null(arg1, true);
1017 arg2 = must_be_not_null(arg2, true);
1018
1019 // Get start addr and length of first argument
1020 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1021 Node* arg1_cnt = load_array_length(arg1);
1022
1023 // Get start addr and length of second argument
1024 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1025 Node* arg2_cnt = load_array_length(arg2);
1026
1027 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1028 set_result(result);
1029 return true;
1030 }
1031
1032 //------------------------------inline_string_equals------------------------
1033 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1034 Node* arg1 = argument(0);
1035 Node* arg2 = argument(1);
1036
1037 // paths (plus control) merge
1038 RegionNode* region = new RegionNode(3);
1039 Node* phi = new PhiNode(region, TypeInt::BOOL);
1040
1041 if (!stopped()) {
1042
1043 arg1 = must_be_not_null(arg1, true);
1044 arg2 = must_be_not_null(arg2, true);
1045
1046 // Get start addr and length of first argument
1047 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1048 Node* arg1_cnt = load_array_length(arg1);
1049
1050 // Get start addr and length of second argument
1051 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1052 Node* arg2_cnt = load_array_length(arg2);
1053
1054 // Check for arg1_cnt != arg2_cnt
1055 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1056 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1057 Node* if_ne = generate_slow_guard(bol, nullptr);
1058 if (if_ne != nullptr) {
1059 phi->init_req(2, intcon(0));
1060 region->init_req(2, if_ne);
1061 }
1062
1063 // Check for count == 0 is done by assembler code for StrEquals.
1064
1065 if (!stopped()) {
1066 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1067 phi->init_req(1, equals);
1068 region->init_req(1, control());
1069 }
1070 }
1071
1072 // post merge
1073 set_control(_gvn.transform(region));
1074 record_for_igvn(region);
1075
1076 set_result(_gvn.transform(phi));
1077 return true;
1078 }
1079
1080 //------------------------------inline_array_equals----------------------------
1081 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1082 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1083 Node* arg1 = argument(0);
1084 Node* arg2 = argument(1);
1085
1086 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1087 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1088 clear_upper_avx();
1089
1090 return true;
1091 }
1092
1093
1094 //------------------------------inline_countPositives------------------------------
1095 bool LibraryCallKit::inline_countPositives() {
1096 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1097 return false;
1098 }
1099
1100 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1101 // no receiver since it is static method
1102 Node* ba = argument(0);
1103 Node* offset = argument(1);
1104 Node* len = argument(2);
1105
1106 ba = must_be_not_null(ba, true);
1107
1108 // Range checks
1109 generate_string_range_check(ba, offset, len, false);
1110 if (stopped()) {
1111 return true;
1112 }
1113 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1114 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1115 set_result(_gvn.transform(result));
1116 return true;
1117 }
1118
1119 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1120 Node* index = argument(0);
1121 Node* length = bt == T_INT ? argument(1) : argument(2);
1122 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1123 return false;
1124 }
1125
1126 // check that length is positive
1127 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1128 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1129
1130 {
1131 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1132 uncommon_trap(Deoptimization::Reason_intrinsic,
1133 Deoptimization::Action_make_not_entrant);
1134 }
1135
1136 if (stopped()) {
1137 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1138 return true;
1139 }
1140
1141 // length is now known positive, add a cast node to make this explicit
1142 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1143 Node* casted_length = ConstraintCastNode::make(control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1144 casted_length = _gvn.transform(casted_length);
1145 replace_in_map(length, casted_length);
1146 length = casted_length;
1147
1148 // Use an unsigned comparison for the range check itself
1149 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1150 BoolTest::mask btest = BoolTest::lt;
1151 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1152 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1153 _gvn.set_type(rc, rc->Value(&_gvn));
1154 if (!rc_bool->is_Con()) {
1155 record_for_igvn(rc);
1156 }
1157 set_control(_gvn.transform(new IfTrueNode(rc)));
1158 {
1159 PreserveJVMState pjvms(this);
1160 set_control(_gvn.transform(new IfFalseNode(rc)));
1161 uncommon_trap(Deoptimization::Reason_range_check,
1162 Deoptimization::Action_make_not_entrant);
1163 }
1164
1165 if (stopped()) {
1166 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1167 return true;
1168 }
1169
1170 // index is now known to be >= 0 and < length, cast it
1171 Node* result = ConstraintCastNode::make(control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt), ConstraintCastNode::RegularDependency, bt);
1172 result = _gvn.transform(result);
1173 set_result(result);
1174 replace_in_map(index, result);
1175 return true;
1176 }
1177
1178 //------------------------------inline_string_indexOf------------------------
1179 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1180 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1181 return false;
1182 }
1183 Node* src = argument(0);
1184 Node* tgt = argument(1);
1185
1186 // Make the merge point
1187 RegionNode* result_rgn = new RegionNode(4);
1188 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1189
1190 src = must_be_not_null(src, true);
1191 tgt = must_be_not_null(tgt, true);
1192
1193 // Get start addr and length of source string
1194 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1195 Node* src_count = load_array_length(src);
1196
1197 // Get start addr and length of substring
1198 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1199 Node* tgt_count = load_array_length(tgt);
1200
1201 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1202 // Divide src size by 2 if String is UTF16 encoded
1203 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1204 }
1205 if (ae == StrIntrinsicNode::UU) {
1206 // Divide substring size by 2 if String is UTF16 encoded
1207 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1208 }
1209
1210 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1211 if (result != nullptr) {
1212 result_phi->init_req(3, result);
1213 result_rgn->init_req(3, control());
1214 }
1215 set_control(_gvn.transform(result_rgn));
1216 record_for_igvn(result_rgn);
1217 set_result(_gvn.transform(result_phi));
1218
1219 return true;
1220 }
1221
1222 //-----------------------------inline_string_indexOf-----------------------
1223 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1224 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1225 return false;
1226 }
1227 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1228 return false;
1229 }
1230 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1231 Node* src = argument(0); // byte[]
1232 Node* src_count = argument(1); // char count
1233 Node* tgt = argument(2); // byte[]
1234 Node* tgt_count = argument(3); // char count
1235 Node* from_index = argument(4); // char index
1236
1237 src = must_be_not_null(src, true);
1238 tgt = must_be_not_null(tgt, true);
1239
1240 // Multiply byte array index by 2 if String is UTF16 encoded
1241 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1242 src_count = _gvn.transform(new SubINode(src_count, from_index));
1243 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1244 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1245
1246 // Range checks
1247 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1248 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1249 if (stopped()) {
1250 return true;
1251 }
1252
1253 RegionNode* region = new RegionNode(5);
1254 Node* phi = new PhiNode(region, TypeInt::INT);
1255
1256 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1257 if (result != nullptr) {
1258 // The result is index relative to from_index if substring was found, -1 otherwise.
1259 // Generate code which will fold into cmove.
1260 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1261 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1262
1263 Node* if_lt = generate_slow_guard(bol, nullptr);
1264 if (if_lt != nullptr) {
1265 // result == -1
1266 phi->init_req(3, result);
1267 region->init_req(3, if_lt);
1268 }
1269 if (!stopped()) {
1270 result = _gvn.transform(new AddINode(result, from_index));
1271 phi->init_req(4, result);
1272 region->init_req(4, control());
1273 }
1274 }
1275
1276 set_control(_gvn.transform(region));
1277 record_for_igvn(region);
1278 set_result(_gvn.transform(phi));
1279 clear_upper_avx();
1280
1281 return true;
1282 }
1283
1284 // Create StrIndexOfNode with fast path checks
1285 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1286 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1287 // Check for substr count > string count
1288 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1289 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1290 Node* if_gt = generate_slow_guard(bol, nullptr);
1291 if (if_gt != nullptr) {
1292 phi->init_req(1, intcon(-1));
1293 region->init_req(1, if_gt);
1294 }
1295 if (!stopped()) {
1296 // Check for substr count == 0
1297 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1298 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1299 Node* if_zero = generate_slow_guard(bol, nullptr);
1300 if (if_zero != nullptr) {
1301 phi->init_req(2, intcon(0));
1302 region->init_req(2, if_zero);
1303 }
1304 }
1305 if (!stopped()) {
1306 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1307 }
1308 return nullptr;
1309 }
1310
1311 //-----------------------------inline_string_indexOfChar-----------------------
1312 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1313 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1314 return false;
1315 }
1316 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1317 return false;
1318 }
1319 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1320 Node* src = argument(0); // byte[]
1321 Node* int_ch = argument(1);
1322 Node* from_index = argument(2);
1323 Node* max = argument(3);
1324
1325 src = must_be_not_null(src, true);
1326
1327 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1328 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1329 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1330
1331 // Range checks
1332 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1333
1334 // Check for int_ch >= 0
1335 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1336 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1337 {
1338 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1339 uncommon_trap(Deoptimization::Reason_intrinsic,
1340 Deoptimization::Action_maybe_recompile);
1341 }
1342 if (stopped()) {
1343 return true;
1344 }
1345
1346 RegionNode* region = new RegionNode(3);
1347 Node* phi = new PhiNode(region, TypeInt::INT);
1348
1349 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1350 C->set_has_split_ifs(true); // Has chance for split-if optimization
1351 _gvn.transform(result);
1352
1353 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1354 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1355
1356 Node* if_lt = generate_slow_guard(bol, nullptr);
1357 if (if_lt != nullptr) {
1358 // result == -1
1359 phi->init_req(2, result);
1360 region->init_req(2, if_lt);
1361 }
1362 if (!stopped()) {
1363 result = _gvn.transform(new AddINode(result, from_index));
1364 phi->init_req(1, result);
1365 region->init_req(1, control());
1366 }
1367 set_control(_gvn.transform(region));
1368 record_for_igvn(region);
1369 set_result(_gvn.transform(phi));
1370
1371 return true;
1372 }
1373 //---------------------------inline_string_copy---------------------
1374 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1375 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1376 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1377 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1378 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1379 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1380 bool LibraryCallKit::inline_string_copy(bool compress) {
1381 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1382 return false;
1383 }
1384 int nargs = 5; // 2 oops, 3 ints
1385 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1386
1387 Node* src = argument(0);
1388 Node* src_offset = argument(1);
1389 Node* dst = argument(2);
1390 Node* dst_offset = argument(3);
1391 Node* length = argument(4);
1392
1393 // Check for allocation before we add nodes that would confuse
1394 // tightly_coupled_allocation()
1395 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1396
1397 // Figure out the size and type of the elements we will be copying.
1398 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1399 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1400 if (src_type == nullptr || dst_type == nullptr) {
1401 return false;
1402 }
1403 BasicType src_elem = src_type->elem()->array_element_basic_type();
1404 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1405 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1406 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1407 "Unsupported array types for inline_string_copy");
1408
1409 src = must_be_not_null(src, true);
1410 dst = must_be_not_null(dst, true);
1411
1412 // Convert char[] offsets to byte[] offsets
1413 bool convert_src = (compress && src_elem == T_BYTE);
1414 bool convert_dst = (!compress && dst_elem == T_BYTE);
1415 if (convert_src) {
1416 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1417 } else if (convert_dst) {
1418 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1419 }
1420
1421 // Range checks
1422 generate_string_range_check(src, src_offset, length, convert_src);
1423 generate_string_range_check(dst, dst_offset, length, convert_dst);
1424 if (stopped()) {
1425 return true;
1426 }
1427
1428 Node* src_start = array_element_address(src, src_offset, src_elem);
1429 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1430 // 'src_start' points to src array + scaled offset
1431 // 'dst_start' points to dst array + scaled offset
1432 Node* count = nullptr;
1433 if (compress) {
1434 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1435 } else {
1436 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1437 }
1438
1439 if (alloc != nullptr) {
1440 if (alloc->maybe_set_complete(&_gvn)) {
1441 // "You break it, you buy it."
1442 InitializeNode* init = alloc->initialization();
1443 assert(init->is_complete(), "we just did this");
1444 init->set_complete_with_arraycopy();
1445 assert(dst->is_CheckCastPP(), "sanity");
1446 assert(dst->in(0)->in(0) == init, "dest pinned");
1447 }
1448 // Do not let stores that initialize this object be reordered with
1449 // a subsequent store that would make this object accessible by
1450 // other threads.
1451 // Record what AllocateNode this StoreStore protects so that
1452 // escape analysis can go from the MemBarStoreStoreNode to the
1453 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1454 // based on the escape status of the AllocateNode.
1455 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1456 }
1457 if (compress) {
1458 set_result(_gvn.transform(count));
1459 }
1460 clear_upper_avx();
1461
1462 return true;
1463 }
1464
1465 #ifdef _LP64
1466 #define XTOP ,top() /*additional argument*/
1467 #else //_LP64
1468 #define XTOP /*no additional argument*/
1469 #endif //_LP64
1470
1471 //------------------------inline_string_toBytesU--------------------------
1472 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1473 bool LibraryCallKit::inline_string_toBytesU() {
1474 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1475 return false;
1476 }
1477 // Get the arguments.
1478 Node* value = argument(0);
1479 Node* offset = argument(1);
1480 Node* length = argument(2);
1481
1482 Node* newcopy = nullptr;
1483
1484 // Set the original stack and the reexecute bit for the interpreter to reexecute
1485 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1486 { PreserveReexecuteState preexecs(this);
1487 jvms()->set_should_reexecute(true);
1488
1489 // Check if a null path was taken unconditionally.
1490 value = null_check(value);
1491
1492 RegionNode* bailout = new RegionNode(1);
1493 record_for_igvn(bailout);
1494
1495 // Range checks
1496 generate_negative_guard(offset, bailout);
1497 generate_negative_guard(length, bailout);
1498 generate_limit_guard(offset, length, load_array_length(value), bailout);
1499 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1500 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1501
1502 if (bailout->req() > 1) {
1503 PreserveJVMState pjvms(this);
1504 set_control(_gvn.transform(bailout));
1505 uncommon_trap(Deoptimization::Reason_intrinsic,
1506 Deoptimization::Action_maybe_recompile);
1507 }
1508 if (stopped()) {
1509 return true;
1510 }
1511
1512 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1513 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1514 newcopy = new_array(klass_node, size, 0); // no arguments to push
1515 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1516 guarantee(alloc != nullptr, "created above");
1517
1518 // Calculate starting addresses.
1519 Node* src_start = array_element_address(value, offset, T_CHAR);
1520 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1521
1522 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1523 const TypeInt* toffset = gvn().type(offset)->is_int();
1524 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1525
1526 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1527 const char* copyfunc_name = "arraycopy";
1528 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1529 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1530 OptoRuntime::fast_arraycopy_Type(),
1531 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1532 src_start, dst_start, ConvI2X(length) XTOP);
1533 // Do not let reads from the cloned object float above the arraycopy.
1534 if (alloc->maybe_set_complete(&_gvn)) {
1535 // "You break it, you buy it."
1536 InitializeNode* init = alloc->initialization();
1537 assert(init->is_complete(), "we just did this");
1538 init->set_complete_with_arraycopy();
1539 assert(newcopy->is_CheckCastPP(), "sanity");
1540 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1541 }
1542 // Do not let stores that initialize this object be reordered with
1543 // a subsequent store that would make this object accessible by
1544 // other threads.
1545 // Record what AllocateNode this StoreStore protects so that
1546 // escape analysis can go from the MemBarStoreStoreNode to the
1547 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1548 // based on the escape status of the AllocateNode.
1549 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1550 } // original reexecute is set back here
1551
1552 C->set_has_split_ifs(true); // Has chance for split-if optimization
1553 if (!stopped()) {
1554 set_result(newcopy);
1555 }
1556 clear_upper_avx();
1557
1558 return true;
1559 }
1560
1561 //------------------------inline_string_getCharsU--------------------------
1562 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1563 bool LibraryCallKit::inline_string_getCharsU() {
1564 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1565 return false;
1566 }
1567
1568 // Get the arguments.
1569 Node* src = argument(0);
1570 Node* src_begin = argument(1);
1571 Node* src_end = argument(2); // exclusive offset (i < src_end)
1572 Node* dst = argument(3);
1573 Node* dst_begin = argument(4);
1574
1575 // Check for allocation before we add nodes that would confuse
1576 // tightly_coupled_allocation()
1577 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1578
1579 // Check if a null path was taken unconditionally.
1580 src = null_check(src);
1581 dst = null_check(dst);
1582 if (stopped()) {
1583 return true;
1584 }
1585
1586 // Get length and convert char[] offset to byte[] offset
1587 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1588 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1589
1590 // Range checks
1591 generate_string_range_check(src, src_begin, length, true);
1592 generate_string_range_check(dst, dst_begin, length, false);
1593 if (stopped()) {
1594 return true;
1595 }
1596
1597 if (!stopped()) {
1598 // Calculate starting addresses.
1599 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1600 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1601
1602 // Check if array addresses are aligned to HeapWordSize
1603 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1604 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1605 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1606 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1607
1608 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1609 const char* copyfunc_name = "arraycopy";
1610 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1611 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1612 OptoRuntime::fast_arraycopy_Type(),
1613 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1614 src_start, dst_start, ConvI2X(length) XTOP);
1615 // Do not let reads from the cloned object float above the arraycopy.
1616 if (alloc != nullptr) {
1617 if (alloc->maybe_set_complete(&_gvn)) {
1618 // "You break it, you buy it."
1619 InitializeNode* init = alloc->initialization();
1620 assert(init->is_complete(), "we just did this");
1621 init->set_complete_with_arraycopy();
1622 assert(dst->is_CheckCastPP(), "sanity");
1623 assert(dst->in(0)->in(0) == init, "dest pinned");
1624 }
1625 // Do not let stores that initialize this object be reordered with
1626 // a subsequent store that would make this object accessible by
1627 // other threads.
1628 // Record what AllocateNode this StoreStore protects so that
1629 // escape analysis can go from the MemBarStoreStoreNode to the
1630 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1631 // based on the escape status of the AllocateNode.
1632 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1633 } else {
1634 insert_mem_bar(Op_MemBarCPUOrder);
1635 }
1636 }
1637
1638 C->set_has_split_ifs(true); // Has chance for split-if optimization
1639 return true;
1640 }
1641
1642 //----------------------inline_string_char_access----------------------------
1643 // Store/Load char to/from byte[] array.
1644 // static void StringUTF16.putChar(byte[] val, int index, int c)
1645 // static char StringUTF16.getChar(byte[] val, int index)
1646 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1647 Node* value = argument(0);
1648 Node* index = argument(1);
1649 Node* ch = is_store ? argument(2) : nullptr;
1650
1651 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1652 // correctly requires matched array shapes.
1653 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1654 "sanity: byte[] and char[] bases agree");
1655 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1656 "sanity: byte[] and char[] scales agree");
1657
1658 // Bail when getChar over constants is requested: constant folding would
1659 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1660 // Java method would constant fold nicely instead.
1661 if (!is_store && value->is_Con() && index->is_Con()) {
1662 return false;
1663 }
1664
1665 // Save state and restore on bailout
1666 uint old_sp = sp();
1667 SafePointNode* old_map = clone_map();
1668
1669 value = must_be_not_null(value, true);
1670
1671 Node* adr = array_element_address(value, index, T_CHAR);
1672 if (adr->is_top()) {
1673 set_map(old_map);
1674 set_sp(old_sp);
1675 return false;
1676 }
1677 destruct_map_clone(old_map);
1678 if (is_store) {
1679 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1680 } else {
1681 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);
1682 set_result(ch);
1683 }
1684 return true;
1685 }
1686
1687 //--------------------------round_double_node--------------------------------
1688 // Round a double node if necessary.
1689 Node* LibraryCallKit::round_double_node(Node* n) {
1690 if (Matcher::strict_fp_requires_explicit_rounding) {
1691 #ifdef IA32
1692 if (UseSSE < 2) {
1693 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1694 }
1695 #else
1696 Unimplemented();
1697 #endif // IA32
1698 }
1699 return n;
1700 }
1701
1702 //------------------------------inline_math-----------------------------------
1703 // public static double Math.abs(double)
1704 // public static double Math.sqrt(double)
1705 // public static double Math.log(double)
1706 // public static double Math.log10(double)
1707 // public static double Math.round(double)
1708 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1709 Node* arg = round_double_node(argument(0));
1710 Node* n = nullptr;
1711 switch (id) {
1712 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1713 case vmIntrinsics::_dsqrt:
1714 case vmIntrinsics::_dsqrt_strict:
1715 n = new SqrtDNode(C, control(), arg); break;
1716 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1717 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1718 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1719 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1720 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1721 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1722 default: fatal_unexpected_iid(id); break;
1723 }
1724 set_result(_gvn.transform(n));
1725 return true;
1726 }
1727
1728 //------------------------------inline_math-----------------------------------
1729 // public static float Math.abs(float)
1730 // public static int Math.abs(int)
1731 // public static long Math.abs(long)
1732 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1733 Node* arg = argument(0);
1734 Node* n = nullptr;
1735 switch (id) {
1736 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1737 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1738 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1739 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1740 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1741 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1742 default: fatal_unexpected_iid(id); break;
1743 }
1744 set_result(_gvn.transform(n));
1745 return true;
1746 }
1747
1748 //------------------------------runtime_math-----------------------------
1749 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1750 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1751 "must be (DD)D or (D)D type");
1752
1753 // Inputs
1754 Node* a = round_double_node(argument(0));
1755 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1756
1757 const TypePtr* no_memory_effects = nullptr;
1758 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1759 no_memory_effects,
1760 a, top(), b, b ? top() : nullptr);
1761 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1762 #ifdef ASSERT
1763 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1764 assert(value_top == top(), "second value must be top");
1765 #endif
1766
1767 set_result(value);
1768 return true;
1769 }
1770
1771 //------------------------------inline_math_pow-----------------------------
1772 bool LibraryCallKit::inline_math_pow() {
1773 Node* exp = round_double_node(argument(2));
1774 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1775 if (d != nullptr) {
1776 if (d->getd() == 2.0) {
1777 // Special case: pow(x, 2.0) => x * x
1778 Node* base = round_double_node(argument(0));
1779 set_result(_gvn.transform(new MulDNode(base, base)));
1780 return true;
1781 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1782 // Special case: pow(x, 0.5) => sqrt(x)
1783 Node* base = round_double_node(argument(0));
1784 Node* zero = _gvn.zerocon(T_DOUBLE);
1785
1786 RegionNode* region = new RegionNode(3);
1787 Node* phi = new PhiNode(region, Type::DOUBLE);
1788
1789 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1790 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1791 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1792 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1793 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1794
1795 Node* if_pow = generate_slow_guard(test, nullptr);
1796 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1797 phi->init_req(1, value_sqrt);
1798 region->init_req(1, control());
1799
1800 if (if_pow != nullptr) {
1801 set_control(if_pow);
1802 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1803 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1804 const TypePtr* no_memory_effects = nullptr;
1805 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1806 no_memory_effects, base, top(), exp, top());
1807 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1808 #ifdef ASSERT
1809 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1810 assert(value_top == top(), "second value must be top");
1811 #endif
1812 phi->init_req(2, value_pow);
1813 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1814 }
1815
1816 C->set_has_split_ifs(true); // Has chance for split-if optimization
1817 set_control(_gvn.transform(region));
1818 record_for_igvn(region);
1819 set_result(_gvn.transform(phi));
1820
1821 return true;
1822 }
1823 }
1824
1825 return StubRoutines::dpow() != nullptr ?
1826 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1827 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1828 }
1829
1830 //------------------------------inline_math_native-----------------------------
1831 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1832 switch (id) {
1833 case vmIntrinsics::_dsin:
1834 return StubRoutines::dsin() != nullptr ?
1835 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1836 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1837 case vmIntrinsics::_dcos:
1838 return StubRoutines::dcos() != nullptr ?
1839 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1840 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1841 case vmIntrinsics::_dtan:
1842 return StubRoutines::dtan() != nullptr ?
1843 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1844 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1845 case vmIntrinsics::_dexp:
1846 return StubRoutines::dexp() != nullptr ?
1847 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1848 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1849 case vmIntrinsics::_dlog:
1850 return StubRoutines::dlog() != nullptr ?
1851 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1852 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1853 case vmIntrinsics::_dlog10:
1854 return StubRoutines::dlog10() != nullptr ?
1855 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1856 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1857
1858 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1859 case vmIntrinsics::_ceil:
1860 case vmIntrinsics::_floor:
1861 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1862
1863 case vmIntrinsics::_dsqrt:
1864 case vmIntrinsics::_dsqrt_strict:
1865 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1866 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1867 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1868 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1869 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1870
1871 case vmIntrinsics::_dpow: return inline_math_pow();
1872 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1873 case vmIntrinsics::_fcopySign: return inline_math(id);
1874 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1875 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1876 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1877
1878 // These intrinsics are not yet correctly implemented
1879 case vmIntrinsics::_datan2:
1880 return false;
1881
1882 default:
1883 fatal_unexpected_iid(id);
1884 return false;
1885 }
1886 }
1887
1888 //----------------------------inline_notify-----------------------------------*
1889 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1890 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1891 address func;
1892 if (id == vmIntrinsics::_notify) {
1893 func = OptoRuntime::monitor_notify_Java();
1894 } else {
1895 func = OptoRuntime::monitor_notifyAll_Java();
1896 }
1897 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1898 make_slow_call_ex(call, env()->Throwable_klass(), false);
1899 return true;
1900 }
1901
1902
1903 //----------------------------inline_min_max-----------------------------------
1904 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1905 set_result(generate_min_max(id, argument(0), argument(1)));
1906 return true;
1907 }
1908
1909 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1910 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1911 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1912 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1913 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1914
1915 {
1916 PreserveJVMState pjvms(this);
1917 PreserveReexecuteState preexecs(this);
1918 jvms()->set_should_reexecute(true);
1919
1920 set_control(slow_path);
1921 set_i_o(i_o());
1922
1923 uncommon_trap(Deoptimization::Reason_intrinsic,
1924 Deoptimization::Action_none);
1925 }
1926
1927 set_control(fast_path);
1928 set_result(math);
1929 }
1930
1931 template <typename OverflowOp>
1932 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1933 typedef typename OverflowOp::MathOp MathOp;
1934
1935 MathOp* mathOp = new MathOp(arg1, arg2);
1936 Node* operation = _gvn.transform( mathOp );
1937 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1938 inline_math_mathExact(operation, ofcheck);
1939 return true;
1940 }
1941
1942 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1943 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1944 }
1945
1946 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1947 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1948 }
1949
1950 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1951 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1952 }
1953
1954 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1955 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1956 }
1957
1958 bool LibraryCallKit::inline_math_negateExactI() {
1959 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1960 }
1961
1962 bool LibraryCallKit::inline_math_negateExactL() {
1963 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1964 }
1965
1966 bool LibraryCallKit::inline_math_multiplyExactI() {
1967 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1968 }
1969
1970 bool LibraryCallKit::inline_math_multiplyExactL() {
1971 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1972 }
1973
1974 bool LibraryCallKit::inline_math_multiplyHigh() {
1975 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1976 return true;
1977 }
1978
1979 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1980 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1981 return true;
1982 }
1983
1984 Node*
1985 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1986 Node* result_val = nullptr;
1987 switch (id) {
1988 case vmIntrinsics::_min:
1989 case vmIntrinsics::_min_strict:
1990 result_val = _gvn.transform(new MinINode(x0, y0));
1991 break;
1992 case vmIntrinsics::_max:
1993 case vmIntrinsics::_max_strict:
1994 result_val = _gvn.transform(new MaxINode(x0, y0));
1995 break;
1996 default:
1997 fatal_unexpected_iid(id);
1998 break;
1999 }
2000 return result_val;
2001 }
2002
2003 inline int
2004 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2005 const TypePtr* base_type = TypePtr::NULL_PTR;
2006 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2007 if (base_type == nullptr) {
2008 // Unknown type.
2009 return Type::AnyPtr;
2010 } else if (base_type == TypePtr::NULL_PTR) {
2011 // Since this is a null+long form, we have to switch to a rawptr.
2012 base = _gvn.transform(new CastX2PNode(offset));
2013 offset = MakeConX(0);
2014 return Type::RawPtr;
2015 } else if (base_type->base() == Type::RawPtr) {
2016 return Type::RawPtr;
2017 } else if (base_type->isa_oopptr()) {
2018 // Base is never null => always a heap address.
2019 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2020 return Type::OopPtr;
2021 }
2022 // Offset is small => always a heap address.
2023 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2024 if (offset_type != nullptr &&
2025 base_type->offset() == 0 && // (should always be?)
2026 offset_type->_lo >= 0 &&
2027 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2028 return Type::OopPtr;
2029 } else if (type == T_OBJECT || type == T_PRIMITIVE_OBJECT) {
2030 // off heap access to an oop doesn't make any sense. Has to be on
2031 // heap.
2032 return Type::OopPtr;
2033 }
2034 // Otherwise, it might either be oop+off or null+addr.
2035 return Type::AnyPtr;
2036 } else {
2037 // No information:
2038 return Type::AnyPtr;
2039 }
2040 }
2041
2042 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2043 Node* uncasted_base = base;
2044 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2045 if (kind == Type::RawPtr) {
2046 return basic_plus_adr(top(), uncasted_base, offset);
2047 } else if (kind == Type::AnyPtr) {
2048 assert(base == uncasted_base, "unexpected base change");
2049 if (can_cast) {
2050 if (!_gvn.type(base)->speculative_maybe_null() &&
2051 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2052 // According to profiling, this access is always on
2053 // heap. Casting the base to not null and thus avoiding membars
2054 // around the access should allow better optimizations
2055 Node* null_ctl = top();
2056 base = null_check_oop(base, &null_ctl, true, true, true);
2057 assert(null_ctl->is_top(), "no null control here");
2058 return basic_plus_adr(base, offset);
2059 } else if (_gvn.type(base)->speculative_always_null() &&
2060 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2061 // According to profiling, this access is always off
2062 // heap.
2063 base = null_assert(base);
2064 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2065 offset = MakeConX(0);
2066 return basic_plus_adr(top(), raw_base, offset);
2067 }
2068 }
2069 // We don't know if it's an on heap or off heap access. Fall back
2070 // to raw memory access.
2071 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2072 return basic_plus_adr(top(), raw, offset);
2073 } else {
2074 assert(base == uncasted_base, "unexpected base change");
2075 // We know it's an on heap access so base can't be null
2076 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2077 base = must_be_not_null(base, true);
2078 }
2079 return basic_plus_adr(base, offset);
2080 }
2081 }
2082
2083 //--------------------------inline_number_methods-----------------------------
2084 // inline int Integer.numberOfLeadingZeros(int)
2085 // inline int Long.numberOfLeadingZeros(long)
2086 //
2087 // inline int Integer.numberOfTrailingZeros(int)
2088 // inline int Long.numberOfTrailingZeros(long)
2089 //
2090 // inline int Integer.bitCount(int)
2091 // inline int Long.bitCount(long)
2092 //
2093 // inline char Character.reverseBytes(char)
2094 // inline short Short.reverseBytes(short)
2095 // inline int Integer.reverseBytes(int)
2096 // inline long Long.reverseBytes(long)
2097 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2098 Node* arg = argument(0);
2099 Node* n = nullptr;
2100 switch (id) {
2101 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2102 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2103 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2104 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2105 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2106 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2107 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2108 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2109 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2110 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2111 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2112 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2113 default: fatal_unexpected_iid(id); break;
2114 }
2115 set_result(_gvn.transform(n));
2116 return true;
2117 }
2118
2119 //--------------------------inline_bitshuffle_methods-----------------------------
2120 // inline int Integer.compress(int, int)
2121 // inline int Integer.expand(int, int)
2122 // inline long Long.compress(long, long)
2123 // inline long Long.expand(long, long)
2124 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2125 Node* n = nullptr;
2126 switch (id) {
2127 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2128 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2129 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2130 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2131 default: fatal_unexpected_iid(id); break;
2132 }
2133 set_result(_gvn.transform(n));
2134 return true;
2135 }
2136
2137 //--------------------------inline_number_methods-----------------------------
2138 // inline int Integer.compareUnsigned(int, int)
2139 // inline int Long.compareUnsigned(long, long)
2140 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2141 Node* arg1 = argument(0);
2142 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2143 Node* n = nullptr;
2144 switch (id) {
2145 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2146 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2147 default: fatal_unexpected_iid(id); break;
2148 }
2149 set_result(_gvn.transform(n));
2150 return true;
2151 }
2152
2153 //--------------------------inline_unsigned_divmod_methods-----------------------------
2154 // inline int Integer.divideUnsigned(int, int)
2155 // inline int Integer.remainderUnsigned(int, int)
2156 // inline long Long.divideUnsigned(long, long)
2157 // inline long Long.remainderUnsigned(long, long)
2158 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2159 Node* n = nullptr;
2160 switch (id) {
2161 case vmIntrinsics::_divideUnsigned_i: {
2162 zero_check_int(argument(1));
2163 // Compile-time detect of null-exception
2164 if (stopped()) {
2165 return true; // keep the graph constructed so far
2166 }
2167 n = new UDivINode(control(), argument(0), argument(1));
2168 break;
2169 }
2170 case vmIntrinsics::_divideUnsigned_l: {
2171 zero_check_long(argument(2));
2172 // Compile-time detect of null-exception
2173 if (stopped()) {
2174 return true; // keep the graph constructed so far
2175 }
2176 n = new UDivLNode(control(), argument(0), argument(2));
2177 break;
2178 }
2179 case vmIntrinsics::_remainderUnsigned_i: {
2180 zero_check_int(argument(1));
2181 // Compile-time detect of null-exception
2182 if (stopped()) {
2183 return true; // keep the graph constructed so far
2184 }
2185 n = new UModINode(control(), argument(0), argument(1));
2186 break;
2187 }
2188 case vmIntrinsics::_remainderUnsigned_l: {
2189 zero_check_long(argument(2));
2190 // Compile-time detect of null-exception
2191 if (stopped()) {
2192 return true; // keep the graph constructed so far
2193 }
2194 n = new UModLNode(control(), argument(0), argument(2));
2195 break;
2196 }
2197 default: fatal_unexpected_iid(id); break;
2198 }
2199 set_result(_gvn.transform(n));
2200 return true;
2201 }
2202
2203 //----------------------------inline_unsafe_access----------------------------
2204
2205 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2206 // Attempt to infer a sharper value type from the offset and base type.
2207 ciKlass* sharpened_klass = nullptr;
2208 bool null_free = false;
2209
2210 // See if it is an instance field, with an object type.
2211 if (alias_type->field() != nullptr) {
2212 if (alias_type->field()->type()->is_klass()) {
2213 sharpened_klass = alias_type->field()->type()->as_klass();
2214 null_free = alias_type->field()->is_null_free();
2215 }
2216 }
2217
2218 const TypeOopPtr* result = nullptr;
2219 // See if it is a narrow oop array.
2220 if (adr_type->isa_aryptr()) {
2221 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2222 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2223 null_free = adr_type->is_aryptr()->is_null_free();
2224 if (elem_type != nullptr && elem_type->is_loaded()) {
2225 // Sharpen the value type.
2226 result = elem_type;
2227 }
2228 }
2229 }
2230
2231 // The sharpened class might be unloaded if there is no class loader
2232 // contraint in place.
2233 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2234 // Sharpen the value type.
2235 result = TypeOopPtr::make_from_klass(sharpened_klass);
2236 if (null_free) {
2237 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2238 }
2239 }
2240 if (result != nullptr) {
2241 #ifndef PRODUCT
2242 if (C->print_intrinsics() || C->print_inlining()) {
2243 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2244 tty->print(" sharpened value: "); result->dump(); tty->cr();
2245 }
2246 #endif
2247 }
2248 return result;
2249 }
2250
2251 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2252 switch (kind) {
2253 case Relaxed:
2254 return MO_UNORDERED;
2255 case Opaque:
2256 return MO_RELAXED;
2257 case Acquire:
2258 return MO_ACQUIRE;
2259 case Release:
2260 return MO_RELEASE;
2261 case Volatile:
2262 return MO_SEQ_CST;
2263 default:
2264 ShouldNotReachHere();
2265 return 0;
2266 }
2267 }
2268
2269 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2270 if (callee()->is_static()) return false; // caller must have the capability!
2271 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2272 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2273 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2274 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2275
2276 if (is_reference_type(type)) {
2277 decorators |= ON_UNKNOWN_OOP_REF;
2278 }
2279
2280 if (unaligned) {
2281 decorators |= C2_UNALIGNED;
2282 }
2283
2284 #ifndef PRODUCT
2285 {
2286 ResourceMark rm;
2287 // Check the signatures.
2288 ciSignature* sig = callee()->signature();
2289 #ifdef ASSERT
2290 if (!is_store) {
2291 // Object getReference(Object base, int/long offset), etc.
2292 BasicType rtype = sig->return_type()->basic_type();
2293 assert(rtype == type || (rtype == T_OBJECT && type == T_PRIMITIVE_OBJECT), "getter must return the expected value");
2294 assert(sig->count() == 2 || (type == T_PRIMITIVE_OBJECT && sig->count() == 3), "oop getter has 2 or 3 arguments");
2295 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2296 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2297 } else {
2298 // void putReference(Object base, int/long offset, Object x), etc.
2299 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2300 assert(sig->count() == 3 || (type == T_PRIMITIVE_OBJECT && sig->count() == 4), "oop putter has 3 arguments");
2301 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2302 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2303 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2304 assert(vtype == type || (type == T_PRIMITIVE_OBJECT && vtype == T_OBJECT), "putter must accept the expected value");
2305 }
2306 #endif // ASSERT
2307 }
2308 #endif //PRODUCT
2309
2310 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2311
2312 Node* receiver = argument(0); // type: oop
2313
2314 // Build address expression.
2315 Node* heap_base_oop = top();
2316
2317 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2318 Node* base = argument(1); // type: oop
2319 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2320 Node* offset = argument(2); // type: long
2321 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2322 // to be plain byte offsets, which are also the same as those accepted
2323 // by oopDesc::field_addr.
2324 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2325 "fieldOffset must be byte-scaled");
2326
2327 ciInlineKlass* inline_klass = nullptr;
2328 if (type == T_PRIMITIVE_OBJECT) {
2329 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2330 if (cls == nullptr || cls->const_oop() == nullptr) {
2331 return false;
2332 }
2333 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2334 if (!mirror_type->is_inlinetype()) {
2335 return false;
2336 }
2337 inline_klass = mirror_type->as_inline_klass();
2338 }
2339
2340 if (base->is_InlineType()) {
2341 InlineTypeNode* vt = base->as_InlineType();
2342 if (is_store) {
2343 if (!vt->is_allocated(&_gvn)) {
2344 return false;
2345 }
2346 base = vt->get_oop();
2347 } else {
2348 if (offset->is_Con()) {
2349 long off = find_long_con(offset, 0);
2350 ciInlineKlass* vk = vt->type()->inline_klass();
2351 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2352 return false;
2353 }
2354
2355 ciField* field = vk->get_non_flattened_field_by_offset(off);
2356 if (field != nullptr) {
2357 BasicType bt = type2field[field->type()->basic_type()];
2358 if (bt == T_ARRAY || bt == T_NARROWOOP || (bt == T_PRIMITIVE_OBJECT && !field->is_flattened())) {
2359 bt = T_OBJECT;
2360 }
2361 if (bt == type && (bt != T_PRIMITIVE_OBJECT || field->type() == inline_klass)) {
2362 Node* value = vt->field_value_by_offset(off, false);
2363 if (value->is_InlineType()) {
2364 value = value->as_InlineType()->adjust_scalarization_depth(this);
2365 }
2366 set_result(value);
2367 return true;
2368 }
2369 }
2370 }
2371 {
2372 // Re-execute the unsafe access if allocation triggers deoptimization.
2373 PreserveReexecuteState preexecs(this);
2374 jvms()->set_should_reexecute(true);
2375 vt = vt->buffer(this);
2376 }
2377 base = vt->get_oop();
2378 }
2379 }
2380
2381 // 32-bit machines ignore the high half!
2382 offset = ConvL2X(offset);
2383
2384 // Save state and restore on bailout
2385 uint old_sp = sp();
2386 SafePointNode* old_map = clone_map();
2387
2388 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2389
2390 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2391 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2392 decorators |= IN_NATIVE; // off-heap primitive access
2393 } else {
2394 set_map(old_map);
2395 set_sp(old_sp);
2396 return false; // off-heap oop accesses are not supported
2397 }
2398 } else {
2399 heap_base_oop = base; // on-heap or mixed access
2400 }
2401
2402 // Can base be null? Otherwise, always on-heap access.
2403 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2404
2405 if (!can_access_non_heap) {
2406 decorators |= IN_HEAP;
2407 }
2408
2409 Node* val = is_store ? argument(4 + (type == T_PRIMITIVE_OBJECT ? 1 : 0)) : nullptr;
2410
2411 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2412 if (adr_type == TypePtr::NULL_PTR) {
2413 set_map(old_map);
2414 set_sp(old_sp);
2415 return false; // off-heap access with zero address
2416 }
2417
2418 // Try to categorize the address.
2419 Compile::AliasType* alias_type = C->alias_type(adr_type);
2420 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2421
2422 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2423 alias_type->adr_type() == TypeAryPtr::RANGE) {
2424 set_map(old_map);
2425 set_sp(old_sp);
2426 return false; // not supported
2427 }
2428
2429 bool mismatched = false;
2430 BasicType bt = T_ILLEGAL;
2431 ciField* field = nullptr;
2432 if (adr_type->isa_instptr()) {
2433 const TypeInstPtr* instptr = adr_type->is_instptr();
2434 ciInstanceKlass* k = instptr->instance_klass();
2435 int off = instptr->offset();
2436 if (instptr->const_oop() != nullptr &&
2437 k == ciEnv::current()->Class_klass() &&
2438 instptr->offset() >= (k->size_helper() * wordSize)) {
2439 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2440 field = k->get_field_by_offset(off, true);
2441 } else {
2442 field = k->get_non_flattened_field_by_offset(off);
2443 }
2444 if (field != nullptr) {
2445 bt = type2field[field->type()->basic_type()];
2446 }
2447 assert(bt == alias_type->basic_type() || bt == T_PRIMITIVE_OBJECT, "should match");
2448 if (field != nullptr && bt == T_PRIMITIVE_OBJECT && !field->is_flattened()) {
2449 bt = T_OBJECT;
2450 }
2451 } else {
2452 bt = alias_type->basic_type();
2453 }
2454
2455 if (bt != T_ILLEGAL) {
2456 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2457 if (adr_type->is_flat()) {
2458 bt = T_PRIMITIVE_OBJECT;
2459 }
2460 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2461 // Alias type doesn't differentiate between byte[] and boolean[]).
2462 // Use address type to get the element type.
2463 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2464 }
2465 if (bt != T_PRIMITIVE_OBJECT && is_reference_type(bt, true)) {
2466 // accessing an array field with getReference is not a mismatch
2467 bt = T_OBJECT;
2468 }
2469 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2470 // Don't intrinsify mismatched object accesses
2471 set_map(old_map);
2472 set_sp(old_sp);
2473 return false;
2474 }
2475 mismatched = (bt != type);
2476 } else if (alias_type->adr_type()->isa_oopptr()) {
2477 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2478 }
2479
2480 if (type == T_PRIMITIVE_OBJECT) {
2481 if (adr_type->isa_instptr()) {
2482 if (field == nullptr || field->type() != inline_klass) {
2483 mismatched = true;
2484 }
2485 } else if (adr_type->isa_aryptr()) {
2486 const Type* elem = adr_type->is_aryptr()->elem();
2487 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2488 mismatched = true;
2489 }
2490 } else {
2491 mismatched = true;
2492 }
2493 if (is_store) {
2494 const Type* val_t = _gvn.type(val);
2495 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2496 set_map(old_map);
2497 set_sp(old_sp);
2498 return false;
2499 }
2500 }
2501 }
2502
2503 destruct_map_clone(old_map);
2504 assert(!mismatched || type == T_PRIMITIVE_OBJECT || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2505
2506 if (mismatched) {
2507 decorators |= C2_MISMATCHED;
2508 }
2509
2510 // First guess at the value type.
2511 const Type *value_type = Type::get_const_basic_type(type);
2512
2513 // Figure out the memory ordering.
2514 decorators |= mo_decorator_for_access_kind(kind);
2515
2516 if (!is_store) {
2517 if (type == T_OBJECT) {
2518 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2519 if (tjp != nullptr) {
2520 value_type = tjp;
2521 }
2522 } else if (type == T_PRIMITIVE_OBJECT) {
2523 value_type = nullptr;
2524 }
2525 }
2526
2527 receiver = null_check(receiver);
2528 if (stopped()) {
2529 return true;
2530 }
2531 // Heap pointers get a null-check from the interpreter,
2532 // as a courtesy. However, this is not guaranteed by Unsafe,
2533 // and it is not possible to fully distinguish unintended nulls
2534 // from intended ones in this API.
2535
2536 if (!is_store) {
2537 Node* p = nullptr;
2538 // Try to constant fold a load from a constant field
2539
2540 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flattened() && !mismatched) {
2541 // final or stable field
2542 p = make_constant_from_field(field, heap_base_oop);
2543 }
2544
2545 if (p == nullptr) { // Could not constant fold the load
2546 if (type == T_PRIMITIVE_OBJECT) {
2547 if (adr_type->isa_instptr() && !mismatched) {
2548 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2549 int offset = adr_type->is_instptr()->offset();
2550 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, base, holder, offset, decorators);
2551 } else {
2552 p = InlineTypeNode::make_from_flattened(this, inline_klass, base, adr, nullptr, 0, decorators);
2553 }
2554 } else {
2555 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2556 const TypeOopPtr* ptr = value_type->make_oopptr();
2557 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2558 // Load a non-flattened inline type from memory
2559 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass(), !ptr->maybe_null());
2560 }
2561 }
2562 // Normalize the value returned by getBoolean in the following cases
2563 if (type == T_BOOLEAN &&
2564 (mismatched ||
2565 heap_base_oop == top() || // - heap_base_oop is null or
2566 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2567 // and the unsafe access is made to large offset
2568 // (i.e., larger than the maximum offset necessary for any
2569 // field access)
2570 ) {
2571 IdealKit ideal = IdealKit(this);
2572 #define __ ideal.
2573 IdealVariable normalized_result(ideal);
2574 __ declarations_done();
2575 __ set(normalized_result, p);
2576 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2577 __ set(normalized_result, ideal.ConI(1));
2578 ideal.end_if();
2579 final_sync(ideal);
2580 p = __ value(normalized_result);
2581 #undef __
2582 }
2583 }
2584 if (type == T_ADDRESS) {
2585 p = gvn().transform(new CastP2XNode(nullptr, p));
2586 p = ConvX2UL(p);
2587 }
2588 // The load node has the control of the preceding MemBarCPUOrder. All
2589 // following nodes will have the control of the MemBarCPUOrder inserted at
2590 // the end of this method. So, pushing the load onto the stack at a later
2591 // point is fine.
2592 set_result(p);
2593 } else {
2594 if (bt == T_ADDRESS) {
2595 // Repackage the long as a pointer.
2596 val = ConvL2X(val);
2597 val = gvn().transform(new CastX2PNode(val));
2598 }
2599 if (type == T_PRIMITIVE_OBJECT) {
2600 if (adr_type->isa_instptr() && !mismatched) {
2601 ciInstanceKlass* holder = adr_type->is_instptr()->instance_klass();
2602 int offset = adr_type->is_instptr()->offset();
2603 val->as_InlineType()->store_flattened(this, base, base, holder, offset, decorators);
2604 } else {
2605 val->as_InlineType()->store_flattened(this, base, adr, nullptr, 0, decorators);
2606 }
2607 } else {
2608 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2609 }
2610 }
2611
2612 if (argument(1)->is_InlineType() && is_store) {
2613 InlineTypeNode* value = InlineTypeNode::make_from_oop(this, base, _gvn.type(argument(1))->inline_klass());
2614 value = value->make_larval(this, false);
2615 replace_in_map(argument(1), value);
2616 }
2617
2618 return true;
2619 }
2620
2621 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2622 Node* receiver = argument(0);
2623 Node* value = argument(1);
2624 if (!value->is_InlineType()) {
2625 return false;
2626 }
2627
2628 receiver = null_check(receiver);
2629 if (stopped()) {
2630 return true;
2631 }
2632
2633 set_result(value->as_InlineType()->make_larval(this, true));
2634 return true;
2635 }
2636
2637 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2638 Node* receiver = argument(0);
2639 Node* buffer = argument(1);
2640 if (!buffer->is_InlineType()) {
2641 return false;
2642 }
2643 InlineTypeNode* vt = buffer->as_InlineType();
2644 if (!vt->is_allocated(&_gvn)) {
2645 return false;
2646 }
2647 // TODO 8239003 Why is this needed?
2648 if (AllocateNode::Ideal_allocation(vt->get_oop(), &_gvn) == nullptr) {
2649 return false;
2650 }
2651
2652 receiver = null_check(receiver);
2653 if (stopped()) {
2654 return true;
2655 }
2656
2657 set_result(vt->finish_larval(this));
2658 return true;
2659 }
2660
2661 //----------------------------inline_unsafe_load_store----------------------------
2662 // This method serves a couple of different customers (depending on LoadStoreKind):
2663 //
2664 // LS_cmp_swap:
2665 //
2666 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2667 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2668 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2669 //
2670 // LS_cmp_swap_weak:
2671 //
2672 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2673 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2674 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2675 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2676 //
2677 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2678 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2679 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2680 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2681 //
2682 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2683 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2684 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2685 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2686 //
2687 // LS_cmp_exchange:
2688 //
2689 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2690 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2691 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2692 //
2693 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2694 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2695 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2696 //
2697 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2698 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2699 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2700 //
2701 // LS_get_add:
2702 //
2703 // int getAndAddInt( Object o, long offset, int delta)
2704 // long getAndAddLong(Object o, long offset, long delta)
2705 //
2706 // LS_get_set:
2707 //
2708 // int getAndSet(Object o, long offset, int newValue)
2709 // long getAndSet(Object o, long offset, long newValue)
2710 // Object getAndSet(Object o, long offset, Object newValue)
2711 //
2712 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2713 // This basic scheme here is the same as inline_unsafe_access, but
2714 // differs in enough details that combining them would make the code
2715 // overly confusing. (This is a true fact! I originally combined
2716 // them, but even I was confused by it!) As much code/comments as
2717 // possible are retained from inline_unsafe_access though to make
2718 // the correspondences clearer. - dl
2719
2720 if (callee()->is_static()) return false; // caller must have the capability!
2721
2722 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2723 decorators |= mo_decorator_for_access_kind(access_kind);
2724
2725 #ifndef PRODUCT
2726 BasicType rtype;
2727 {
2728 ResourceMark rm;
2729 // Check the signatures.
2730 ciSignature* sig = callee()->signature();
2731 rtype = sig->return_type()->basic_type();
2732 switch(kind) {
2733 case LS_get_add:
2734 case LS_get_set: {
2735 // Check the signatures.
2736 #ifdef ASSERT
2737 assert(rtype == type, "get and set must return the expected type");
2738 assert(sig->count() == 3, "get and set has 3 arguments");
2739 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2740 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2741 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2742 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2743 #endif // ASSERT
2744 break;
2745 }
2746 case LS_cmp_swap:
2747 case LS_cmp_swap_weak: {
2748 // Check the signatures.
2749 #ifdef ASSERT
2750 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2751 assert(sig->count() == 4, "CAS has 4 arguments");
2752 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2753 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2754 #endif // ASSERT
2755 break;
2756 }
2757 case LS_cmp_exchange: {
2758 // Check the signatures.
2759 #ifdef ASSERT
2760 assert(rtype == type, "CAS must return the expected type");
2761 assert(sig->count() == 4, "CAS has 4 arguments");
2762 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2763 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2764 #endif // ASSERT
2765 break;
2766 }
2767 default:
2768 ShouldNotReachHere();
2769 }
2770 }
2771 #endif //PRODUCT
2772
2773 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2774
2775 // Get arguments:
2776 Node* receiver = nullptr;
2777 Node* base = nullptr;
2778 Node* offset = nullptr;
2779 Node* oldval = nullptr;
2780 Node* newval = nullptr;
2781 switch(kind) {
2782 case LS_cmp_swap:
2783 case LS_cmp_swap_weak:
2784 case LS_cmp_exchange: {
2785 const bool two_slot_type = type2size[type] == 2;
2786 receiver = argument(0); // type: oop
2787 base = argument(1); // type: oop
2788 offset = argument(2); // type: long
2789 oldval = argument(4); // type: oop, int, or long
2790 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2791 break;
2792 }
2793 case LS_get_add:
2794 case LS_get_set: {
2795 receiver = argument(0); // type: oop
2796 base = argument(1); // type: oop
2797 offset = argument(2); // type: long
2798 oldval = nullptr;
2799 newval = argument(4); // type: oop, int, or long
2800 break;
2801 }
2802 default:
2803 ShouldNotReachHere();
2804 }
2805
2806 // Build field offset expression.
2807 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2808 // to be plain byte offsets, which are also the same as those accepted
2809 // by oopDesc::field_addr.
2810 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2811 // 32-bit machines ignore the high half of long offsets
2812 offset = ConvL2X(offset);
2813 // Save state and restore on bailout
2814 uint old_sp = sp();
2815 SafePointNode* old_map = clone_map();
2816 Node* adr = make_unsafe_address(base, offset,type, false);
2817 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2818
2819 Compile::AliasType* alias_type = C->alias_type(adr_type);
2820 BasicType bt = alias_type->basic_type();
2821 if (bt != T_ILLEGAL &&
2822 (is_reference_type(bt) != (type == T_OBJECT))) {
2823 // Don't intrinsify mismatched object accesses.
2824 set_map(old_map);
2825 set_sp(old_sp);
2826 return false;
2827 }
2828
2829 destruct_map_clone(old_map);
2830
2831 // For CAS, unlike inline_unsafe_access, there seems no point in
2832 // trying to refine types. Just use the coarse types here.
2833 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2834 const Type *value_type = Type::get_const_basic_type(type);
2835
2836 switch (kind) {
2837 case LS_get_set:
2838 case LS_cmp_exchange: {
2839 if (type == T_OBJECT) {
2840 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2841 if (tjp != nullptr) {
2842 value_type = tjp;
2843 }
2844 }
2845 break;
2846 }
2847 case LS_cmp_swap:
2848 case LS_cmp_swap_weak:
2849 case LS_get_add:
2850 break;
2851 default:
2852 ShouldNotReachHere();
2853 }
2854
2855 // Null check receiver.
2856 receiver = null_check(receiver);
2857 if (stopped()) {
2858 return true;
2859 }
2860
2861 int alias_idx = C->get_alias_index(adr_type);
2862
2863 if (is_reference_type(type)) {
2864 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2865
2866 if (oldval != nullptr && oldval->is_InlineType()) {
2867 // Re-execute the unsafe access if allocation triggers deoptimization.
2868 PreserveReexecuteState preexecs(this);
2869 jvms()->set_should_reexecute(true);
2870 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
2871 }
2872 if (newval != nullptr && newval->is_InlineType()) {
2873 // Re-execute the unsafe access if allocation triggers deoptimization.
2874 PreserveReexecuteState preexecs(this);
2875 jvms()->set_should_reexecute(true);
2876 newval = newval->as_InlineType()->buffer(this)->get_oop();
2877 }
2878
2879 // Transformation of a value which could be null pointer (CastPP #null)
2880 // could be delayed during Parse (for example, in adjust_map_after_if()).
2881 // Execute transformation here to avoid barrier generation in such case.
2882 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2883 newval = _gvn.makecon(TypePtr::NULL_PTR);
2884
2885 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2886 // Refine the value to a null constant, when it is known to be null
2887 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2888 }
2889 }
2890
2891 Node* result = nullptr;
2892 switch (kind) {
2893 case LS_cmp_exchange: {
2894 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2895 oldval, newval, value_type, type, decorators);
2896 break;
2897 }
2898 case LS_cmp_swap_weak:
2899 decorators |= C2_WEAK_CMPXCHG;
2900 case LS_cmp_swap: {
2901 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2902 oldval, newval, value_type, type, decorators);
2903 break;
2904 }
2905 case LS_get_set: {
2906 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2907 newval, value_type, type, decorators);
2908 break;
2909 }
2910 case LS_get_add: {
2911 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2912 newval, value_type, type, decorators);
2913 break;
2914 }
2915 default:
2916 ShouldNotReachHere();
2917 }
2918
2919 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2920 set_result(result);
2921 return true;
2922 }
2923
2924 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2925 // Regardless of form, don't allow previous ld/st to move down,
2926 // then issue acquire, release, or volatile mem_bar.
2927 insert_mem_bar(Op_MemBarCPUOrder);
2928 switch(id) {
2929 case vmIntrinsics::_loadFence:
2930 insert_mem_bar(Op_LoadFence);
2931 return true;
2932 case vmIntrinsics::_storeFence:
2933 insert_mem_bar(Op_StoreFence);
2934 return true;
2935 case vmIntrinsics::_storeStoreFence:
2936 insert_mem_bar(Op_StoreStoreFence);
2937 return true;
2938 case vmIntrinsics::_fullFence:
2939 insert_mem_bar(Op_MemBarVolatile);
2940 return true;
2941 default:
2942 fatal_unexpected_iid(id);
2943 return false;
2944 }
2945 }
2946
2947 bool LibraryCallKit::inline_onspinwait() {
2948 insert_mem_bar(Op_OnSpinWait);
2949 return true;
2950 }
2951
2952 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2953 if (!kls->is_Con()) {
2954 return true;
2955 }
2956 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2957 if (klsptr == nullptr) {
2958 return true;
2959 }
2960 ciInstanceKlass* ik = klsptr->instance_klass();
2961 // don't need a guard for a klass that is already initialized
2962 return !ik->is_initialized();
2963 }
2964
2965 //----------------------------inline_unsafe_writeback0-------------------------
2966 // public native void Unsafe.writeback0(long address)
2967 bool LibraryCallKit::inline_unsafe_writeback0() {
2968 if (!Matcher::has_match_rule(Op_CacheWB)) {
2969 return false;
2970 }
2971 #ifndef PRODUCT
2972 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2973 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2974 ciSignature* sig = callee()->signature();
2975 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2976 #endif
2977 null_check_receiver(); // null-check, then ignore
2978 Node *addr = argument(1);
2979 addr = new CastX2PNode(addr);
2980 addr = _gvn.transform(addr);
2981 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2982 flush = _gvn.transform(flush);
2983 set_memory(flush, TypeRawPtr::BOTTOM);
2984 return true;
2985 }
2986
2987 //----------------------------inline_unsafe_writeback0-------------------------
2988 // public native void Unsafe.writeback0(long address)
2989 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2990 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2991 return false;
2992 }
2993 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2994 return false;
2995 }
2996 #ifndef PRODUCT
2997 assert(Matcher::has_match_rule(Op_CacheWB),
2998 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2999 : "found match rule for CacheWBPostSync but not CacheWB"));
3000
3001 #endif
3002 null_check_receiver(); // null-check, then ignore
3003 Node *sync;
3004 if (is_pre) {
3005 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3006 } else {
3007 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3008 }
3009 sync = _gvn.transform(sync);
3010 set_memory(sync, TypeRawPtr::BOTTOM);
3011 return true;
3012 }
3013
3014 //----------------------------inline_unsafe_allocate---------------------------
3015 // public native Object Unsafe.allocateInstance(Class<?> cls);
3016 bool LibraryCallKit::inline_unsafe_allocate() {
3017 if (callee()->is_static()) return false; // caller must have the capability!
3018
3019 null_check_receiver(); // null-check, then ignore
3020 Node* cls = null_check(argument(1));
3021 if (stopped()) return true;
3022
3023 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3024 kls = null_check(kls);
3025 if (stopped()) return true; // argument was like int.class
3026
3027 Node* test = nullptr;
3028 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3029 // Note: The argument might still be an illegal value like
3030 // Serializable.class or Object[].class. The runtime will handle it.
3031 // But we must make an explicit check for initialization.
3032 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3033 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3034 // can generate code to load it as unsigned byte.
3035 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
3036 Node* bits = intcon(InstanceKlass::fully_initialized);
3037 test = _gvn.transform(new SubINode(inst, bits));
3038 // The 'test' is non-zero if we need to take a slow path.
3039 }
3040 Node* obj = nullptr;
3041 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3042 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3043 obj = InlineTypeNode::make_default(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3044 } else {
3045 obj = new_instance(kls, test);
3046 }
3047 set_result(obj);
3048 return true;
3049 }
3050
3051 //------------------------inline_native_time_funcs--------------
3052 // inline code for System.currentTimeMillis() and System.nanoTime()
3053 // these have the same type and signature
3054 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3055 const TypeFunc* tf = OptoRuntime::void_long_Type();
3056 const TypePtr* no_memory_effects = nullptr;
3057 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3058 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3059 #ifdef ASSERT
3060 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3061 assert(value_top == top(), "second value must be top");
3062 #endif
3063 set_result(value);
3064 return true;
3065 }
3066
3067
3068 #if INCLUDE_JVMTI
3069
3070 // When notifications are disabled then just update the VTMS transition bit and return.
3071 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3072 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3073 if (!DoJVMTIVirtualThreadTransitions) {
3074 return true;
3075 }
3076 IdealKit ideal(this);
3077
3078 Node* ONE = ideal.ConI(1);
3079 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3080 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3081 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3082
3083 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3084 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3085 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3086 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3087
3088 sync_kit(ideal);
3089 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3090 ideal.sync_kit(this);
3091 } ideal.else_(); {
3092 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3093 Node* vt_oop = _gvn.transform(argument(0)); // this argument - VirtualThread oop
3094 Node* thread = ideal.thread();
3095 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3096 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3097 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3098
3099 sync_kit(ideal);
3100 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3101 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3102
3103 ideal.sync_kit(this);
3104 } ideal.end_if();
3105 final_sync(ideal);
3106
3107 return true;
3108 }
3109
3110 // Always update the temporary VTMS transition bit.
3111 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
3112 if (!DoJVMTIVirtualThreadTransitions) {
3113 return true;
3114 }
3115 IdealKit ideal(this);
3116
3117 {
3118 // unconditionally update the temporary VTMS transition bit in current JavaThread
3119 Node* thread = ideal.thread();
3120 Node* hide = _gvn.transform(argument(1)); // hide argument for temporary VTMS transition notification
3121 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
3122 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3123
3124 sync_kit(ideal);
3125 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3126 ideal.sync_kit(this);
3127 }
3128 final_sync(ideal);
3129
3130 return true;
3131 }
3132
3133 #endif // INCLUDE_JVMTI
3134
3135 #ifdef JFR_HAVE_INTRINSICS
3136
3137 /**
3138 * if oop->klass != null
3139 * // normal class
3140 * epoch = _epoch_state ? 2 : 1
3141 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3142 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3143 * }
3144 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3145 * else
3146 * // primitive class
3147 * if oop->array_klass != null
3148 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3149 * else
3150 * id = LAST_TYPE_ID + 1 // void class path
3151 * if (!signaled)
3152 * signaled = true
3153 */
3154 bool LibraryCallKit::inline_native_classID() {
3155 Node* cls = argument(0);
3156
3157 IdealKit ideal(this);
3158 #define __ ideal.
3159 IdealVariable result(ideal); __ declarations_done();
3160 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3161 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3162 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3163
3164
3165 __ if_then(kls, BoolTest::ne, null()); {
3166 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3167 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3168
3169 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3170 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3171 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3172 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3173 mask = _gvn.transform(new OrLNode(mask, epoch));
3174 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3175
3176 float unlikely = PROB_UNLIKELY(0.999);
3177 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3178 sync_kit(ideal);
3179 make_runtime_call(RC_LEAF,
3180 OptoRuntime::class_id_load_barrier_Type(),
3181 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3182 "class id load barrier",
3183 TypePtr::BOTTOM,
3184 kls);
3185 ideal.sync_kit(this);
3186 } __ end_if();
3187
3188 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3189 } __ else_(); {
3190 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3191 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3192 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3193 __ if_then(array_kls, BoolTest::ne, null()); {
3194 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3195 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3196 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3197 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3198 } __ else_(); {
3199 // void class case
3200 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3201 } __ end_if();
3202
3203 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3204 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3205 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3206 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3207 } __ end_if();
3208 } __ end_if();
3209
3210 final_sync(ideal);
3211 set_result(ideal.value(result));
3212 #undef __
3213 return true;
3214 }
3215
3216 /*
3217 * The intrinsic is a model of this pseudo-code:
3218 *
3219 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3220 * jobject h_event_writer = tl->java_event_writer();
3221 * if (h_event_writer == nullptr) {
3222 * return nullptr;
3223 * }
3224 * oop threadObj = Thread::threadObj();
3225 * oop vthread = java_lang_Thread::vthread(threadObj);
3226 * traceid tid;
3227 * bool excluded;
3228 * if (vthread != threadObj) { // i.e. current thread is virtual
3229 * tid = java_lang_Thread::tid(vthread);
3230 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3231 * excluded = vthread_epoch_raw & excluded_mask;
3232 * if (!excluded) {
3233 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3234 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3235 * if (vthread_epoch != current_epoch) {
3236 * write_checkpoint();
3237 * }
3238 * }
3239 * } else {
3240 * tid = java_lang_Thread::tid(threadObj);
3241 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3242 * excluded = thread_epoch_raw & excluded_mask;
3243 * }
3244 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3245 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3246 * if (tid_in_event_writer != tid) {
3247 * setField(event_writer, "threadID", tid);
3248 * setField(event_writer, "excluded", excluded);
3249 * }
3250 * return event_writer
3251 */
3252 bool LibraryCallKit::inline_native_getEventWriter() {
3253 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3254
3255 // Save input memory and i_o state.
3256 Node* input_memory_state = reset_memory();
3257 set_all_memory(input_memory_state);
3258 Node* input_io_state = i_o();
3259
3260 Node* excluded_mask = _gvn.intcon(32768);
3261 Node* epoch_mask = _gvn.intcon(32767);
3262
3263 // TLS
3264 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3265
3266 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3267 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3268
3269 // Load the eventwriter jobject handle.
3270 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3271
3272 // Null check the jobject handle.
3273 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3274 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3275 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3276
3277 // False path, jobj is null.
3278 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3279
3280 // True path, jobj is not null.
3281 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3282
3283 set_control(jobj_is_not_null);
3284
3285 // Load the threadObj for the CarrierThread.
3286 Node* threadObj = generate_current_thread(tls_ptr);
3287
3288 // Load the vthread.
3289 Node* vthread = generate_virtual_thread(tls_ptr);
3290
3291 // If vthread != threadObj, this is a virtual thread.
3292 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3293 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3294 IfNode* iff_vthread_not_equal_threadObj =
3295 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3296
3297 // False branch, fallback to threadObj.
3298 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3299 set_control(vthread_equal_threadObj);
3300
3301 // Load the tid field from the vthread object.
3302 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3303
3304 // Load the raw epoch value from the threadObj.
3305 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3306 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3307 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3308
3309 // Mask off the excluded information from the epoch.
3310 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3311
3312 // True branch, this is a virtual thread.
3313 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3314 set_control(vthread_not_equal_threadObj);
3315
3316 // Load the tid field from the vthread object.
3317 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3318
3319 // Load the raw epoch value from the vthread.
3320 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3321 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3322 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3323
3324 // Mask off the excluded information from the epoch.
3325 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3326
3327 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3328 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3329 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3330 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3331
3332 // False branch, vthread is excluded, no need to write epoch info.
3333 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3334
3335 // True branch, vthread is included, update epoch info.
3336 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3337 set_control(included);
3338
3339 // Get epoch value.
3340 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3341
3342 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3343 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3344 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3345
3346 // Compare the epoch in the vthread to the current epoch generation.
3347 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3348 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3349 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3350
3351 // False path, epoch is equal, checkpoint information is valid.
3352 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3353
3354 // True path, epoch is not equal, write a checkpoint for the vthread.
3355 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3356
3357 set_control(epoch_is_not_equal);
3358
3359 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3360 // The call also updates the native thread local thread id and the vthread with the current epoch.
3361 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3362 OptoRuntime::jfr_write_checkpoint_Type(),
3363 StubRoutines::jfr_write_checkpoint(),
3364 "write_checkpoint", TypePtr::BOTTOM);
3365 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3366
3367 // vthread epoch != current epoch
3368 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3369 record_for_igvn(epoch_compare_rgn);
3370 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3371 record_for_igvn(epoch_compare_mem);
3372 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3373 record_for_igvn(epoch_compare_io);
3374
3375 // Update control and phi nodes.
3376 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3377 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3378 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3379 epoch_compare_mem->init_req(_false_path, input_memory_state);
3380 epoch_compare_io->init_req(_true_path, i_o());
3381 epoch_compare_io->init_req(_false_path, input_io_state);
3382
3383 // excluded != true
3384 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3385 record_for_igvn(exclude_compare_rgn);
3386 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3387 record_for_igvn(exclude_compare_mem);
3388 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3389 record_for_igvn(exclude_compare_io);
3390
3391 // Update control and phi nodes.
3392 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3393 exclude_compare_rgn->init_req(_false_path, excluded);
3394 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3395 exclude_compare_mem->init_req(_false_path, input_memory_state);
3396 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3397 exclude_compare_io->init_req(_false_path, input_io_state);
3398
3399 // vthread != threadObj
3400 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3401 record_for_igvn(vthread_compare_rgn);
3402 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3403 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3404 record_for_igvn(vthread_compare_io);
3405 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3406 record_for_igvn(tid);
3407 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3408 record_for_igvn(exclusion);
3409
3410 // Update control and phi nodes.
3411 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3412 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3413 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3414 vthread_compare_mem->init_req(_false_path, input_memory_state);
3415 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3416 vthread_compare_io->init_req(_false_path, input_io_state);
3417 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3418 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3419 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3420 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3421
3422 // Update branch state.
3423 set_control(_gvn.transform(vthread_compare_rgn));
3424 set_all_memory(_gvn.transform(vthread_compare_mem));
3425 set_i_o(_gvn.transform(vthread_compare_io));
3426
3427 // Load the event writer oop by dereferencing the jobject handle.
3428 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3429 assert(klass_EventWriter->is_loaded(), "invariant");
3430 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3431 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3432 const TypeOopPtr* const xtype = aklass->as_instance_type();
3433 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3434 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3435
3436 // Load the current thread id from the event writer object.
3437 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3438 // Get the field offset to, conditionally, store an updated tid value later.
3439 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3440 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3441 // Get the field offset to, conditionally, store an updated exclusion value later.
3442 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3443 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3444
3445 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3446 record_for_igvn(event_writer_tid_compare_rgn);
3447 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3448 record_for_igvn(event_writer_tid_compare_mem);
3449 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3450 record_for_igvn(event_writer_tid_compare_io);
3451
3452 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3453 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3454 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3455 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3456
3457 // False path, tids are the same.
3458 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3459
3460 // True path, tid is not equal, need to update the tid in the event writer.
3461 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3462 record_for_igvn(tid_is_not_equal);
3463
3464 // Store the exclusion state to the event writer.
3465 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3466
3467 // Store the tid to the event writer.
3468 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3469
3470 // Update control and phi nodes.
3471 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3472 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3473 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3474 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3475 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3476 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3477
3478 // Result of top level CFG, Memory, IO and Value.
3479 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3480 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3481 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3482 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3483
3484 // Result control.
3485 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3486 result_rgn->init_req(_false_path, jobj_is_null);
3487
3488 // Result memory.
3489 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3490 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3491
3492 // Result IO.
3493 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3494 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3495
3496 // Result value.
3497 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3498 result_value->init_req(_false_path, null()); // return null
3499
3500 // Set output state.
3501 set_control(_gvn.transform(result_rgn));
3502 set_all_memory(_gvn.transform(result_mem));
3503 set_i_o(_gvn.transform(result_io));
3504 set_result(result_rgn, result_value);
3505 return true;
3506 }
3507
3508 /*
3509 * The intrinsic is a model of this pseudo-code:
3510 *
3511 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3512 * if (carrierThread != thread) { // is virtual thread
3513 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3514 * bool excluded = vthread_epoch_raw & excluded_mask;
3515 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3516 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3517 * if (!excluded) {
3518 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3519 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3520 * }
3521 * Atomic::release_store(&tl->_vthread, true);
3522 * return;
3523 * }
3524 * Atomic::release_store(&tl->_vthread, false);
3525 */
3526 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3527 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3528
3529 Node* input_memory_state = reset_memory();
3530 set_all_memory(input_memory_state);
3531
3532 Node* excluded_mask = _gvn.intcon(32768);
3533 Node* epoch_mask = _gvn.intcon(32767);
3534
3535 Node* const carrierThread = generate_current_thread(jt);
3536 // If thread != carrierThread, this is a virtual thread.
3537 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3538 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3539 IfNode* iff_thread_not_equal_carrierThread =
3540 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3541
3542 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3543
3544 // False branch, is carrierThread.
3545 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3546 // Store release
3547 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3548
3549 set_all_memory(input_memory_state);
3550
3551 // True branch, is virtual thread.
3552 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3553 set_control(thread_not_equal_carrierThread);
3554
3555 // Load the raw epoch value from the vthread.
3556 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3557 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3558 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3559
3560 // Mask off the excluded information from the epoch.
3561 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3562
3563 // Load the tid field from the thread.
3564 Node* tid = load_field_from_object(thread, "tid", "J");
3565
3566 // Store the vthread tid to the jfr thread local.
3567 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3568 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3569
3570 // Branch is_excluded to conditionalize updating the epoch .
3571 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3572 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3573 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3574
3575 // True branch, vthread is excluded, no need to write epoch info.
3576 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3577 set_control(excluded);
3578 Node* vthread_is_excluded = _gvn.intcon(1);
3579
3580 // False branch, vthread is included, update epoch info.
3581 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3582 set_control(included);
3583 Node* vthread_is_included = _gvn.intcon(0);
3584
3585 // Get epoch value.
3586 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3587
3588 // Store the vthread epoch to the jfr thread local.
3589 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3590 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3591
3592 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3593 record_for_igvn(excluded_rgn);
3594 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3595 record_for_igvn(excluded_mem);
3596 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3597 record_for_igvn(exclusion);
3598
3599 // Merge the excluded control and memory.
3600 excluded_rgn->init_req(_true_path, excluded);
3601 excluded_rgn->init_req(_false_path, included);
3602 excluded_mem->init_req(_true_path, tid_memory);
3603 excluded_mem->init_req(_false_path, included_memory);
3604 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3605 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3606
3607 // Set intermediate state.
3608 set_control(_gvn.transform(excluded_rgn));
3609 set_all_memory(excluded_mem);
3610
3611 // Store the vthread exclusion state to the jfr thread local.
3612 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3613 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3614
3615 // Store release
3616 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3617
3618 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3619 record_for_igvn(thread_compare_rgn);
3620 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3621 record_for_igvn(thread_compare_mem);
3622 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3623 record_for_igvn(vthread);
3624
3625 // Merge the thread_compare control and memory.
3626 thread_compare_rgn->init_req(_true_path, control());
3627 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3628 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3629 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3630
3631 // Set output state.
3632 set_control(_gvn.transform(thread_compare_rgn));
3633 set_all_memory(_gvn.transform(thread_compare_mem));
3634 }
3635
3636 #endif // JFR_HAVE_INTRINSICS
3637
3638 //------------------------inline_native_currentCarrierThread------------------
3639 bool LibraryCallKit::inline_native_currentCarrierThread() {
3640 Node* junk = nullptr;
3641 set_result(generate_current_thread(junk));
3642 return true;
3643 }
3644
3645 //------------------------inline_native_currentThread------------------
3646 bool LibraryCallKit::inline_native_currentThread() {
3647 Node* junk = nullptr;
3648 set_result(generate_virtual_thread(junk));
3649 return true;
3650 }
3651
3652 //------------------------inline_native_setVthread------------------
3653 bool LibraryCallKit::inline_native_setCurrentThread() {
3654 assert(C->method()->changes_current_thread(),
3655 "method changes current Thread but is not annotated ChangesCurrentThread");
3656 Node* arr = argument(1);
3657 Node* thread = _gvn.transform(new ThreadLocalNode());
3658 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3659 Node* thread_obj_handle
3660 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3661 thread_obj_handle = _gvn.transform(thread_obj_handle);
3662 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3663 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3664 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3665 return true;
3666 }
3667
3668 Node* LibraryCallKit::scopedValueCache_helper() {
3669 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3670 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3671
3672 bool xk = etype->klass_is_exact();
3673
3674 Node* thread = _gvn.transform(new ThreadLocalNode());
3675 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3676 // We cannot use immutable_memory() because we might flip onto a
3677 // different carrier thread, at which point we'll need to use that
3678 // carrier thread's cache.
3679 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3680 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3681 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3682 }
3683
3684 //------------------------inline_native_scopedValueCache------------------
3685 bool LibraryCallKit::inline_native_scopedValueCache() {
3686 ciKlass *objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3687 const TypeOopPtr *etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3688 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3689
3690 // Because we create the scopedValue cache lazily we have to make the
3691 // type of the result BotPTR.
3692 bool xk = etype->klass_is_exact();
3693 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
3694 Node* cache_obj_handle = scopedValueCache_helper();
3695 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3696
3697 return true;
3698 }
3699
3700 //------------------------inline_native_setScopedValueCache------------------
3701 bool LibraryCallKit::inline_native_setScopedValueCache() {
3702 Node* arr = argument(0);
3703 Node* cache_obj_handle = scopedValueCache_helper();
3704
3705 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3706 access_store_at(nullptr, cache_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3707
3708 return true;
3709 }
3710
3711 //-----------------------load_klass_from_mirror_common-------------------------
3712 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3713 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3714 // and branch to the given path on the region.
3715 // If never_see_null, take an uncommon trap on null, so we can optimistically
3716 // compile for the non-null case.
3717 // If the region is null, force never_see_null = true.
3718 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3719 bool never_see_null,
3720 RegionNode* region,
3721 int null_path,
3722 int offset) {
3723 if (region == nullptr) never_see_null = true;
3724 Node* p = basic_plus_adr(mirror, offset);
3725 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3726 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3727 Node* null_ctl = top();
3728 kls = null_check_oop(kls, &null_ctl, never_see_null);
3729 if (region != nullptr) {
3730 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3731 region->init_req(null_path, null_ctl);
3732 } else {
3733 assert(null_ctl == top(), "no loose ends");
3734 }
3735 return kls;
3736 }
3737
3738 //--------------------(inline_native_Class_query helpers)---------------------
3739 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3740 // Fall through if (mods & mask) == bits, take the guard otherwise.
3741 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3742 // Branch around if the given klass has the given modifier bit set.
3743 // Like generate_guard, adds a new path onto the region.
3744 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3745 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3746 Node* mask = intcon(modifier_mask);
3747 Node* bits = intcon(modifier_bits);
3748 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3749 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3750 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3751 return generate_fair_guard(bol, region);
3752 }
3753
3754 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3755 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3756 }
3757 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3758 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3759 }
3760
3761 //-------------------------inline_native_Class_query-------------------
3762 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3763 const Type* return_type = TypeInt::BOOL;
3764 Node* prim_return_value = top(); // what happens if it's a primitive class?
3765 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3766 bool expect_prim = false; // most of these guys expect to work on refs
3767
3768 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3769
3770 Node* mirror = argument(0);
3771 Node* obj = top();
3772
3773 switch (id) {
3774 case vmIntrinsics::_isInstance:
3775 // nothing is an instance of a primitive type
3776 prim_return_value = intcon(0);
3777 obj = argument(1);
3778 break;
3779 case vmIntrinsics::_getModifiers:
3780 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3781 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3782 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3783 break;
3784 case vmIntrinsics::_isInterface:
3785 prim_return_value = intcon(0);
3786 break;
3787 case vmIntrinsics::_isArray:
3788 prim_return_value = intcon(0);
3789 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3790 break;
3791 case vmIntrinsics::_isPrimitive:
3792 prim_return_value = intcon(1);
3793 expect_prim = true; // obviously
3794 break;
3795 case vmIntrinsics::_isHidden:
3796 prim_return_value = intcon(0);
3797 break;
3798 case vmIntrinsics::_getSuperclass:
3799 prim_return_value = null();
3800 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3801 break;
3802 case vmIntrinsics::_getClassAccessFlags:
3803 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3804 return_type = TypeInt::INT; // not bool! 6297094
3805 break;
3806 default:
3807 fatal_unexpected_iid(id);
3808 break;
3809 }
3810
3811 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3812 if (mirror_con == nullptr) return false; // cannot happen?
3813
3814 #ifndef PRODUCT
3815 if (C->print_intrinsics() || C->print_inlining()) {
3816 ciType* k = mirror_con->java_mirror_type();
3817 if (k) {
3818 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3819 k->print_name();
3820 tty->cr();
3821 }
3822 }
3823 #endif
3824
3825 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3826 RegionNode* region = new RegionNode(PATH_LIMIT);
3827 record_for_igvn(region);
3828 PhiNode* phi = new PhiNode(region, return_type);
3829
3830 // The mirror will never be null of Reflection.getClassAccessFlags, however
3831 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3832 // if it is. See bug 4774291.
3833
3834 // For Reflection.getClassAccessFlags(), the null check occurs in
3835 // the wrong place; see inline_unsafe_access(), above, for a similar
3836 // situation.
3837 mirror = null_check(mirror);
3838 // If mirror or obj is dead, only null-path is taken.
3839 if (stopped()) return true;
3840
3841 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3842
3843 // Now load the mirror's klass metaobject, and null-check it.
3844 // Side-effects region with the control path if the klass is null.
3845 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3846 // If kls is null, we have a primitive mirror.
3847 phi->init_req(_prim_path, prim_return_value);
3848 if (stopped()) { set_result(region, phi); return true; }
3849 bool safe_for_replace = (region->in(_prim_path) == top());
3850
3851 Node* p; // handy temp
3852 Node* null_ctl;
3853
3854 // Now that we have the non-null klass, we can perform the real query.
3855 // For constant classes, the query will constant-fold in LoadNode::Value.
3856 Node* query_value = top();
3857 switch (id) {
3858 case vmIntrinsics::_isInstance:
3859 // nothing is an instance of a primitive type
3860 query_value = gen_instanceof(obj, kls, safe_for_replace);
3861 break;
3862
3863 case vmIntrinsics::_getModifiers:
3864 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3865 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3866 break;
3867
3868 case vmIntrinsics::_isInterface:
3869 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3870 if (generate_interface_guard(kls, region) != nullptr)
3871 // A guard was added. If the guard is taken, it was an interface.
3872 phi->add_req(intcon(1));
3873 // If we fall through, it's a plain class.
3874 query_value = intcon(0);
3875 break;
3876
3877 case vmIntrinsics::_isArray:
3878 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3879 if (generate_array_guard(kls, region) != nullptr)
3880 // A guard was added. If the guard is taken, it was an array.
3881 phi->add_req(intcon(1));
3882 // If we fall through, it's a plain class.
3883 query_value = intcon(0);
3884 break;
3885
3886 case vmIntrinsics::_isPrimitive:
3887 query_value = intcon(0); // "normal" path produces false
3888 break;
3889
3890 case vmIntrinsics::_isHidden:
3891 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3892 if (generate_hidden_class_guard(kls, region) != nullptr)
3893 // A guard was added. If the guard is taken, it was an hidden class.
3894 phi->add_req(intcon(1));
3895 // If we fall through, it's a plain class.
3896 query_value = intcon(0);
3897 break;
3898
3899
3900 case vmIntrinsics::_getSuperclass:
3901 // The rules here are somewhat unfortunate, but we can still do better
3902 // with random logic than with a JNI call.
3903 // Interfaces store null or Object as _super, but must report null.
3904 // Arrays store an intermediate super as _super, but must report Object.
3905 // Other types can report the actual _super.
3906 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3907 if (generate_interface_guard(kls, region) != nullptr)
3908 // A guard was added. If the guard is taken, it was an interface.
3909 phi->add_req(null());
3910 if (generate_array_guard(kls, region) != nullptr)
3911 // A guard was added. If the guard is taken, it was an array.
3912 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3913 // If we fall through, it's a plain class. Get its _super.
3914 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3915 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3916 null_ctl = top();
3917 kls = null_check_oop(kls, &null_ctl);
3918 if (null_ctl != top()) {
3919 // If the guard is taken, Object.superClass is null (both klass and mirror).
3920 region->add_req(null_ctl);
3921 phi ->add_req(null());
3922 }
3923 if (!stopped()) {
3924 query_value = load_mirror_from_klass(kls);
3925 }
3926 break;
3927
3928 case vmIntrinsics::_getClassAccessFlags:
3929 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3930 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3931 break;
3932
3933 default:
3934 fatal_unexpected_iid(id);
3935 break;
3936 }
3937
3938 // Fall-through is the normal case of a query to a real class.
3939 phi->init_req(1, query_value);
3940 region->init_req(1, control());
3941
3942 C->set_has_split_ifs(true); // Has chance for split-if optimization
3943 set_result(region, phi);
3944 return true;
3945 }
3946
3947 //-------------------------inline_primitive_Class_conversion-------------------
3948 // Class<T> java.lang.Class .asPrimaryType()
3949 // public static Class<T> jdk.internal.value.PrimitiveClass.asPrimaryType(Class<T>)
3950 // Class<T> java.lang.Class .asValueType()
3951 // public static Class<T> jdk.internal.value.PrimitiveClass.asValueType(Class<T>)
3952 bool LibraryCallKit::inline_primitive_Class_conversion(vmIntrinsics::ID id) {
3953 Node* mirror = argument(0); // Receiver/argument Class
3954 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3955 if (mirror_con == nullptr) {
3956 return false;
3957 }
3958
3959 bool is_val_mirror = true;
3960 ciType* tm = mirror_con->java_mirror_type(&is_val_mirror);
3961 if (tm != nullptr) {
3962 Node* result = mirror;
3963 if ((id == vmIntrinsics::_asPrimaryType || id == vmIntrinsics::_asPrimaryTypeArg) && is_val_mirror) {
3964 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->ref_mirror()));
3965 } else if (id == vmIntrinsics::_asValueType || id == vmIntrinsics::_asValueTypeArg) {
3966 if (!tm->is_inlinetype()) {
3967 return false; // Throw UnsupportedOperationException
3968 } else if (!is_val_mirror) {
3969 result = _gvn.makecon(TypeInstPtr::make(tm->as_inline_klass()->val_mirror()));
3970 }
3971 }
3972 set_result(result);
3973 return true;
3974 }
3975 return false;
3976 }
3977
3978 //-------------------------inline_Class_cast-------------------
3979 bool LibraryCallKit::inline_Class_cast() {
3980 Node* mirror = argument(0); // Class
3981 Node* obj = argument(1);
3982 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3983 if (mirror_con == nullptr) {
3984 return false; // dead path (mirror->is_top()).
3985 }
3986 if (obj == nullptr || obj->is_top()) {
3987 return false; // dead path
3988 }
3989 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3990
3991 // First, see if Class.cast() can be folded statically.
3992 // java_mirror_type() returns non-null for compile-time Class constants.
3993 bool requires_null_check = false;
3994 ciType* tm = mirror_con->java_mirror_type(&requires_null_check);
3995 if (tm != nullptr && tm->is_klass() &&
3996 tp != nullptr) {
3997 if (!tp->is_loaded()) {
3998 // Don't use intrinsic when class is not loaded.
3999 return false;
4000 } else {
4001 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4002 if (static_res == Compile::SSC_always_true) {
4003 // isInstance() is true - fold the code.
4004 if (requires_null_check) {
4005 obj = null_check(obj);
4006 }
4007 set_result(obj);
4008 return true;
4009 } else if (static_res == Compile::SSC_always_false) {
4010 // Don't use intrinsic, have to throw ClassCastException.
4011 // If the reference is null, the non-intrinsic bytecode will
4012 // be optimized appropriately.
4013 return false;
4014 }
4015 }
4016 }
4017
4018 // Bailout intrinsic and do normal inlining if exception path is frequent.
4019 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4020 return false;
4021 }
4022
4023 // Generate dynamic checks.
4024 // Class.cast() is java implementation of _checkcast bytecode.
4025 // Do checkcast (Parse::do_checkcast()) optimizations here.
4026
4027 if (requires_null_check) {
4028 obj = null_check(obj);
4029 }
4030 mirror = null_check(mirror);
4031 // If mirror is dead, only null-path is taken.
4032 if (stopped()) {
4033 return true;
4034 }
4035
4036 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4037 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4038 RegionNode* region = new RegionNode(PATH_LIMIT);
4039 record_for_igvn(region);
4040
4041 // Now load the mirror's klass metaobject, and null-check it.
4042 // If kls is null, we have a primitive mirror and
4043 // nothing is an instance of a primitive type.
4044 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4045
4046 Node* res = top();
4047 Node* io = i_o();
4048 Node* mem = merged_memory();
4049 if (!stopped()) {
4050 if (EnableValhalla && !requires_null_check) {
4051 // Check if we are casting to QMyValue
4052 Node* ctrl_val_mirror = generate_fair_guard(is_val_mirror(mirror), nullptr);
4053 if (ctrl_val_mirror != nullptr) {
4054 RegionNode* r = new RegionNode(3);
4055 record_for_igvn(r);
4056 r->init_req(1, control());
4057
4058 // Casting to QMyValue, check for null
4059 set_control(ctrl_val_mirror);
4060 { // PreserveJVMState because null check replaces obj in map
4061 PreserveJVMState pjvms(this);
4062 Node* null_ctr = top();
4063 null_check_oop(obj, &null_ctr);
4064 region->init_req(_npe_path, null_ctr);
4065 r->init_req(2, control());
4066 }
4067 set_control(_gvn.transform(r));
4068 }
4069 }
4070
4071 Node* bad_type_ctrl = top();
4072 // Do checkcast optimizations.
4073 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4074 region->init_req(_bad_type_path, bad_type_ctrl);
4075 }
4076 if (region->in(_prim_path) != top() ||
4077 region->in(_bad_type_path) != top() ||
4078 region->in(_npe_path) != top()) {
4079 // Let Interpreter throw ClassCastException.
4080 PreserveJVMState pjvms(this);
4081 set_control(_gvn.transform(region));
4082 // Set IO and memory because gen_checkcast may override them when buffering inline types
4083 set_i_o(io);
4084 set_all_memory(mem);
4085 uncommon_trap(Deoptimization::Reason_intrinsic,
4086 Deoptimization::Action_maybe_recompile);
4087 }
4088 if (!stopped()) {
4089 set_result(res);
4090 }
4091 return true;
4092 }
4093
4094
4095 //--------------------------inline_native_subtype_check------------------------
4096 // This intrinsic takes the JNI calls out of the heart of
4097 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4098 bool LibraryCallKit::inline_native_subtype_check() {
4099 // Pull both arguments off the stack.
4100 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4101 args[0] = argument(0);
4102 args[1] = argument(1);
4103 Node* klasses[2]; // corresponding Klasses: superk, subk
4104 klasses[0] = klasses[1] = top();
4105
4106 enum {
4107 // A full decision tree on {superc is prim, subc is prim}:
4108 _prim_0_path = 1, // {P,N} => false
4109 // {P,P} & superc!=subc => false
4110 _prim_same_path, // {P,P} & superc==subc => true
4111 _prim_1_path, // {N,P} => false
4112 _ref_subtype_path, // {N,N} & subtype check wins => true
4113 _both_ref_path, // {N,N} & subtype check loses => false
4114 PATH_LIMIT
4115 };
4116
4117 RegionNode* region = new RegionNode(PATH_LIMIT);
4118 RegionNode* prim_region = new RegionNode(2);
4119 Node* phi = new PhiNode(region, TypeInt::BOOL);
4120 record_for_igvn(region);
4121 record_for_igvn(prim_region);
4122
4123 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4124 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4125 int class_klass_offset = java_lang_Class::klass_offset();
4126
4127 // First null-check both mirrors and load each mirror's klass metaobject.
4128 int which_arg;
4129 for (which_arg = 0; which_arg <= 1; which_arg++) {
4130 Node* arg = args[which_arg];
4131 arg = null_check(arg);
4132 if (stopped()) break;
4133 args[which_arg] = arg;
4134
4135 Node* p = basic_plus_adr(arg, class_klass_offset);
4136 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4137 klasses[which_arg] = _gvn.transform(kls);
4138 }
4139
4140 // Having loaded both klasses, test each for null.
4141 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4142 for (which_arg = 0; which_arg <= 1; which_arg++) {
4143 Node* kls = klasses[which_arg];
4144 Node* null_ctl = top();
4145 kls = null_check_oop(kls, &null_ctl, never_see_null);
4146 if (which_arg == 0) {
4147 prim_region->init_req(1, null_ctl);
4148 } else {
4149 region->init_req(_prim_1_path, null_ctl);
4150 }
4151 if (stopped()) break;
4152 klasses[which_arg] = kls;
4153 }
4154
4155 if (!stopped()) {
4156 // now we have two reference types, in klasses[0..1]
4157 Node* subk = klasses[1]; // the argument to isAssignableFrom
4158 Node* superk = klasses[0]; // the receiver
4159 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4160 // If superc is an inline mirror, we also need to check if superc == subc because LMyValue
4161 // is not a subtype of QMyValue but due to subk == superk the subtype check will pass.
4162 generate_fair_guard(is_val_mirror(args[0]), prim_region);
4163 // now we have a successful reference subtype check
4164 region->set_req(_ref_subtype_path, control());
4165 }
4166
4167 // If both operands are primitive (both klasses null), then
4168 // we must return true when they are identical primitives.
4169 // It is convenient to test this after the first null klass check.
4170 // This path is also used if superc is a value mirror.
4171 set_control(_gvn.transform(prim_region));
4172 if (!stopped()) {
4173 // Since superc is primitive, make a guard for the superc==subc case.
4174 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4175 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4176 generate_fair_guard(bol_eq, region);
4177 if (region->req() == PATH_LIMIT+1) {
4178 // A guard was added. If the added guard is taken, superc==subc.
4179 region->swap_edges(PATH_LIMIT, _prim_same_path);
4180 region->del_req(PATH_LIMIT);
4181 }
4182 region->set_req(_prim_0_path, control()); // Not equal after all.
4183 }
4184
4185 // these are the only paths that produce 'true':
4186 phi->set_req(_prim_same_path, intcon(1));
4187 phi->set_req(_ref_subtype_path, intcon(1));
4188
4189 // pull together the cases:
4190 assert(region->req() == PATH_LIMIT, "sane region");
4191 for (uint i = 1; i < region->req(); i++) {
4192 Node* ctl = region->in(i);
4193 if (ctl == nullptr || ctl == top()) {
4194 region->set_req(i, top());
4195 phi ->set_req(i, top());
4196 } else if (phi->in(i) == nullptr) {
4197 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4198 }
4199 }
4200
4201 set_control(_gvn.transform(region));
4202 set_result(_gvn.transform(phi));
4203 return true;
4204 }
4205
4206 //---------------------generate_array_guard_common------------------------
4207 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind) {
4208
4209 if (stopped()) {
4210 return nullptr;
4211 }
4212
4213 // Like generate_guard, adds a new path onto the region.
4214 jint layout_con = 0;
4215 Node* layout_val = get_layout_helper(kls, layout_con);
4216 if (layout_val == nullptr) {
4217 bool query = 0;
4218 switch(kind) {
4219 case ObjectArray: query = Klass::layout_helper_is_objArray(layout_con); break;
4220 case NonObjectArray: query = !Klass::layout_helper_is_objArray(layout_con); break;
4221 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4222 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4223 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4224 default:
4225 ShouldNotReachHere();
4226 }
4227 if (!query) {
4228 return nullptr; // never a branch
4229 } else { // always a branch
4230 Node* always_branch = control();
4231 if (region != nullptr)
4232 region->add_req(always_branch);
4233 set_control(top());
4234 return always_branch;
4235 }
4236 }
4237 unsigned int value = 0;
4238 BoolTest::mask btest = BoolTest::illegal;
4239 switch(kind) {
4240 case ObjectArray:
4241 case NonObjectArray: {
4242 value = Klass::_lh_array_tag_obj_value;
4243 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4244 btest = (kind == ObjectArray) ? BoolTest::eq : BoolTest::ne;
4245 break;
4246 }
4247 case TypeArray: {
4248 value = Klass::_lh_array_tag_type_value;
4249 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4250 btest = BoolTest::eq;
4251 break;
4252 }
4253 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4254 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4255 default:
4256 ShouldNotReachHere();
4257 }
4258 // Now test the correct condition.
4259 jint nval = (jint)value;
4260 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4261 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4262 return generate_fair_guard(bol, region);
4263 }
4264
4265
4266 //-----------------------inline_native_newArray--------------------------
4267 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4268 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4269 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4270 Node* mirror;
4271 Node* count_val;
4272 if (uninitialized) {
4273 mirror = argument(1);
4274 count_val = argument(2);
4275 } else {
4276 mirror = argument(0);
4277 count_val = argument(1);
4278 }
4279
4280 mirror = null_check(mirror);
4281 // If mirror or obj is dead, only null-path is taken.
4282 if (stopped()) return true;
4283
4284 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4285 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4286 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4287 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4288 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4289
4290 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4291 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4292 result_reg, _slow_path);
4293 Node* normal_ctl = control();
4294 Node* no_array_ctl = result_reg->in(_slow_path);
4295
4296 // Generate code for the slow case. We make a call to newArray().
4297 set_control(no_array_ctl);
4298 if (!stopped()) {
4299 // Either the input type is void.class, or else the
4300 // array klass has not yet been cached. Either the
4301 // ensuing call will throw an exception, or else it
4302 // will cache the array klass for next time.
4303 PreserveJVMState pjvms(this);
4304 CallJavaNode* slow_call = nullptr;
4305 if (uninitialized) {
4306 // Generate optimized virtual call (holder class 'Unsafe' is final)
4307 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false);
4308 } else {
4309 slow_call = generate_method_call_static(vmIntrinsics::_newArray);
4310 }
4311 Node* slow_result = set_results_for_java_call(slow_call);
4312 // this->control() comes from set_results_for_java_call
4313 result_reg->set_req(_slow_path, control());
4314 result_val->set_req(_slow_path, slow_result);
4315 result_io ->set_req(_slow_path, i_o());
4316 result_mem->set_req(_slow_path, reset_memory());
4317 }
4318
4319 set_control(normal_ctl);
4320 if (!stopped()) {
4321 // Normal case: The array type has been cached in the java.lang.Class.
4322 // The following call works fine even if the array type is polymorphic.
4323 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4324 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4325 result_reg->init_req(_normal_path, control());
4326 result_val->init_req(_normal_path, obj);
4327 result_io ->init_req(_normal_path, i_o());
4328 result_mem->init_req(_normal_path, reset_memory());
4329
4330 if (uninitialized) {
4331 // Mark the allocation so that zeroing is skipped
4332 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj, &_gvn);
4333 alloc->maybe_set_complete(&_gvn);
4334 }
4335 }
4336
4337 // Return the combined state.
4338 set_i_o( _gvn.transform(result_io) );
4339 set_all_memory( _gvn.transform(result_mem));
4340
4341 C->set_has_split_ifs(true); // Has chance for split-if optimization
4342 set_result(result_reg, result_val);
4343 return true;
4344 }
4345
4346 //----------------------inline_native_getLength--------------------------
4347 // public static native int java.lang.reflect.Array.getLength(Object array);
4348 bool LibraryCallKit::inline_native_getLength() {
4349 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4350
4351 Node* array = null_check(argument(0));
4352 // If array is dead, only null-path is taken.
4353 if (stopped()) return true;
4354
4355 // Deoptimize if it is a non-array.
4356 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4357
4358 if (non_array != nullptr) {
4359 PreserveJVMState pjvms(this);
4360 set_control(non_array);
4361 uncommon_trap(Deoptimization::Reason_intrinsic,
4362 Deoptimization::Action_maybe_recompile);
4363 }
4364
4365 // If control is dead, only non-array-path is taken.
4366 if (stopped()) return true;
4367
4368 // The works fine even if the array type is polymorphic.
4369 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4370 Node* result = load_array_length(array);
4371
4372 C->set_has_split_ifs(true); // Has chance for split-if optimization
4373 set_result(result);
4374 return true;
4375 }
4376
4377 //------------------------inline_array_copyOf----------------------------
4378 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4379 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4380 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4381 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4382
4383 // Get the arguments.
4384 Node* original = argument(0);
4385 Node* start = is_copyOfRange? argument(1): intcon(0);
4386 Node* end = is_copyOfRange? argument(2): argument(1);
4387 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4388
4389 Node* newcopy = nullptr;
4390
4391 // Set the original stack and the reexecute bit for the interpreter to reexecute
4392 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4393 { PreserveReexecuteState preexecs(this);
4394 jvms()->set_should_reexecute(true);
4395
4396 array_type_mirror = null_check(array_type_mirror);
4397 original = null_check(original);
4398
4399 // Check if a null path was taken unconditionally.
4400 if (stopped()) return true;
4401
4402 Node* orig_length = load_array_length(original);
4403
4404 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4405 klass_node = null_check(klass_node);
4406
4407 RegionNode* bailout = new RegionNode(1);
4408 record_for_igvn(bailout);
4409
4410 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4411 // Bail out if that is so.
4412 // Inline type array may have object field that would require a
4413 // write barrier. Conservatively, go to slow path.
4414 // TODO 8251971: Optimize for the case when flat src/dst are later found
4415 // to not contain oops (i.e., move this check to the macro expansion phase).
4416 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
4417 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
4418 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
4419 bool exclude_flat = UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
4420 // Can src array be flat and contain oops?
4421 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
4422 // Can dest array be flat and contain oops?
4423 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
4424 Node* not_objArray = exclude_flat ? generate_non_objArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
4425 if (not_objArray != nullptr) {
4426 // Improve the klass node's type from the new optimistic assumption:
4427 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4428 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
4429 Node* cast = new CastPPNode(klass_node, akls);
4430 cast->init_req(0, control());
4431 klass_node = _gvn.transform(cast);
4432 }
4433
4434 // Bail out if either start or end is negative.
4435 generate_negative_guard(start, bailout, &start);
4436 generate_negative_guard(end, bailout, &end);
4437
4438 Node* length = end;
4439 if (_gvn.type(start) != TypeInt::ZERO) {
4440 length = _gvn.transform(new SubINode(end, start));
4441 }
4442
4443 // Bail out if length is negative.
4444 // Without this the new_array would throw
4445 // NegativeArraySizeException but IllegalArgumentException is what
4446 // should be thrown
4447 generate_negative_guard(length, bailout, &length);
4448
4449 // Handle inline type arrays
4450 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
4451 if (!stopped()) {
4452 orig_t = _gvn.type(original)->isa_aryptr();
4453 if (orig_t != nullptr && orig_t->is_flat()) {
4454 // Src is flat, check that dest is flat as well
4455 if (exclude_flat) {
4456 // Dest can't be flat, bail out
4457 bailout->add_req(control());
4458 set_control(top());
4459 } else {
4460 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
4461 }
4462 } else if (UseFlatArray && (orig_t == nullptr || !orig_t->is_not_flat()) &&
4463 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
4464 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
4465 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
4466 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
4467 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
4468 if (orig_t != nullptr) {
4469 orig_t = orig_t->cast_to_not_flat();
4470 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
4471 }
4472 }
4473 if (!can_validate) {
4474 // No validation. The subtype check emitted at macro expansion time will not go to the slow
4475 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
4476 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
4477 generate_fair_guard(null_free_array_test(klass_node), bailout);
4478 }
4479 }
4480
4481 if (bailout->req() > 1) {
4482 PreserveJVMState pjvms(this);
4483 set_control(_gvn.transform(bailout));
4484 uncommon_trap(Deoptimization::Reason_intrinsic,
4485 Deoptimization::Action_maybe_recompile);
4486 }
4487
4488 if (!stopped()) {
4489 // How many elements will we copy from the original?
4490 // The answer is MinI(orig_length - start, length).
4491 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4492 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4493
4494 // Generate a direct call to the right arraycopy function(s).
4495 // We know the copy is disjoint but we might not know if the
4496 // oop stores need checking.
4497 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4498 // This will fail a store-check if x contains any non-nulls.
4499
4500 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4501 // loads/stores but it is legal only if we're sure the
4502 // Arrays.copyOf would succeed. So we need all input arguments
4503 // to the copyOf to be validated, including that the copy to the
4504 // new array won't trigger an ArrayStoreException. That subtype
4505 // check can be optimized if we know something on the type of
4506 // the input array from type speculation.
4507 if (_gvn.type(klass_node)->singleton()) {
4508 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4509 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4510
4511 int test = C->static_subtype_check(superk, subk);
4512 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4513 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4514 if (t_original->speculative_type() != nullptr) {
4515 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4516 }
4517 }
4518 }
4519
4520 bool validated = false;
4521 // Reason_class_check rather than Reason_intrinsic because we
4522 // want to intrinsify even if this traps.
4523 if (can_validate) {
4524 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4525
4526 if (not_subtype_ctrl != top()) {
4527 PreserveJVMState pjvms(this);
4528 set_control(not_subtype_ctrl);
4529 uncommon_trap(Deoptimization::Reason_class_check,
4530 Deoptimization::Action_make_not_entrant);
4531 assert(stopped(), "Should be stopped");
4532 }
4533 validated = true;
4534 }
4535
4536 if (!stopped()) {
4537 newcopy = new_array(klass_node, length, 0); // no arguments to push
4538
4539 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, false,
4540 load_object_klass(original), klass_node);
4541 if (!is_copyOfRange) {
4542 ac->set_copyof(validated);
4543 } else {
4544 ac->set_copyofrange(validated);
4545 }
4546 Node* n = _gvn.transform(ac);
4547 if (n == ac) {
4548 ac->connect_outputs(this);
4549 } else {
4550 assert(validated, "shouldn't transform if all arguments not validated");
4551 set_all_memory(n);
4552 }
4553 }
4554 }
4555 } // original reexecute is set back here
4556
4557 C->set_has_split_ifs(true); // Has chance for split-if optimization
4558 if (!stopped()) {
4559 set_result(newcopy);
4560 }
4561 return true;
4562 }
4563
4564
4565 //----------------------generate_virtual_guard---------------------------
4566 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4567 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4568 RegionNode* slow_region) {
4569 ciMethod* method = callee();
4570 int vtable_index = method->vtable_index();
4571 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4572 "bad index %d", vtable_index);
4573 // Get the Method* out of the appropriate vtable entry.
4574 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4575 vtable_index*vtableEntry::size_in_bytes() +
4576 in_bytes(vtableEntry::method_offset());
4577 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4578 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4579
4580 // Compare the target method with the expected method (e.g., Object.hashCode).
4581 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4582
4583 Node* native_call = makecon(native_call_addr);
4584 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4585 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4586
4587 return generate_slow_guard(test_native, slow_region);
4588 }
4589
4590 //-----------------------generate_method_call----------------------------
4591 // Use generate_method_call to make a slow-call to the real
4592 // method if the fast path fails. An alternative would be to
4593 // use a stub like OptoRuntime::slow_arraycopy_Java.
4594 // This only works for expanding the current library call,
4595 // not another intrinsic. (E.g., don't use this for making an
4596 // arraycopy call inside of the copyOf intrinsic.)
4597 CallJavaNode*
4598 LibraryCallKit::generate_method_call(vmIntrinsics::ID method_id, bool is_virtual, bool is_static) {
4599 // When compiling the intrinsic method itself, do not use this technique.
4600 guarantee(callee() != C->method(), "cannot make slow-call to self");
4601
4602 ciMethod* method = callee();
4603 // ensure the JVMS we have will be correct for this call
4604 guarantee(method_id == method->intrinsic_id(), "must match");
4605
4606 const TypeFunc* tf = TypeFunc::make(method);
4607 CallJavaNode* slow_call;
4608 if (is_static) {
4609 assert(!is_virtual, "");
4610 slow_call = new CallStaticJavaNode(C, tf,
4611 SharedRuntime::get_resolve_static_call_stub(), method);
4612 } else if (is_virtual) {
4613 null_check_receiver();
4614 int vtable_index = Method::invalid_vtable_index;
4615 if (UseInlineCaches) {
4616 // Suppress the vtable call
4617 } else {
4618 // hashCode and clone are not a miranda methods,
4619 // so the vtable index is fixed.
4620 // No need to use the linkResolver to get it.
4621 vtable_index = method->vtable_index();
4622 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4623 "bad index %d", vtable_index);
4624 }
4625 slow_call = new CallDynamicJavaNode(tf,
4626 SharedRuntime::get_resolve_virtual_call_stub(),
4627 method, vtable_index);
4628 } else { // neither virtual nor static: opt_virtual
4629 null_check_receiver();
4630 slow_call = new CallStaticJavaNode(C, tf,
4631 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4632 slow_call->set_optimized_virtual(true);
4633 }
4634 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4635 // To be able to issue a direct call (optimized virtual or virtual)
4636 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4637 // about the method being invoked should be attached to the call site to
4638 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4639 slow_call->set_override_symbolic_info(true);
4640 }
4641 set_arguments_for_java_call(slow_call);
4642 set_edges_for_java_call(slow_call);
4643 return slow_call;
4644 }
4645
4646
4647 /**
4648 * Build special case code for calls to hashCode on an object. This call may
4649 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4650 * slightly different code.
4651 */
4652 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4653 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4654 assert(!(is_virtual && is_static), "either virtual, special, or static");
4655
4656 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4657
4658 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4659 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4660 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4661 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4662 Node* obj = argument(0);
4663
4664 if (gvn().type(obj)->is_inlinetypeptr()) {
4665 return false;
4666 }
4667
4668 if (!is_static) {
4669 // Check for hashing null object
4670 obj = null_check_receiver();
4671 if (stopped()) return true; // unconditionally null
4672 result_reg->init_req(_null_path, top());
4673 result_val->init_req(_null_path, top());
4674 } else {
4675 // Do a null check, and return zero if null.
4676 // System.identityHashCode(null) == 0
4677 Node* null_ctl = top();
4678 obj = null_check_oop(obj, &null_ctl);
4679 result_reg->init_req(_null_path, null_ctl);
4680 result_val->init_req(_null_path, _gvn.intcon(0));
4681 }
4682
4683 // Unconditionally null? Then return right away.
4684 if (stopped()) {
4685 set_control( result_reg->in(_null_path));
4686 if (!stopped())
4687 set_result(result_val->in(_null_path));
4688 return true;
4689 }
4690
4691 // We only go to the fast case code if we pass a number of guards. The
4692 // paths which do not pass are accumulated in the slow_region.
4693 RegionNode* slow_region = new RegionNode(1);
4694 record_for_igvn(slow_region);
4695
4696 // If this is a virtual call, we generate a funny guard. We pull out
4697 // the vtable entry corresponding to hashCode() from the target object.
4698 // If the target method which we are calling happens to be the native
4699 // Object hashCode() method, we pass the guard. We do not need this
4700 // guard for non-virtual calls -- the caller is known to be the native
4701 // Object hashCode().
4702 if (is_virtual) {
4703 // After null check, get the object's klass.
4704 Node* obj_klass = load_object_klass(obj);
4705 generate_virtual_guard(obj_klass, slow_region);
4706 }
4707
4708 // Get the header out of the object, use LoadMarkNode when available
4709 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4710 // The control of the load must be null. Otherwise, the load can move before
4711 // the null check after castPP removal.
4712 Node* no_ctrl = nullptr;
4713 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4714
4715 // Test the header to see if it is unlocked.
4716 // This also serves as guard against inline types
4717 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
4718 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4719 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4720 Node *chk_unlocked = _gvn.transform(new CmpXNode( lmasked_header, unlocked_val));
4721 Node *test_unlocked = _gvn.transform(new BoolNode( chk_unlocked, BoolTest::ne));
4722
4723 generate_slow_guard(test_unlocked, slow_region);
4724
4725 // Get the hash value and check to see that it has been properly assigned.
4726 // We depend on hash_mask being at most 32 bits and avoid the use of
4727 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4728 // vm: see markWord.hpp.
4729 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4730 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4731 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4732 // This hack lets the hash bits live anywhere in the mark object now, as long
4733 // as the shift drops the relevant bits into the low 32 bits. Note that
4734 // Java spec says that HashCode is an int so there's no point in capturing
4735 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4736 hshifted_header = ConvX2I(hshifted_header);
4737 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4738
4739 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4740 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4741 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4742
4743 generate_slow_guard(test_assigned, slow_region);
4744
4745 Node* init_mem = reset_memory();
4746 // fill in the rest of the null path:
4747 result_io ->init_req(_null_path, i_o());
4748 result_mem->init_req(_null_path, init_mem);
4749
4750 result_val->init_req(_fast_path, hash_val);
4751 result_reg->init_req(_fast_path, control());
4752 result_io ->init_req(_fast_path, i_o());
4753 result_mem->init_req(_fast_path, init_mem);
4754
4755 // Generate code for the slow case. We make a call to hashCode().
4756 set_control(_gvn.transform(slow_region));
4757 if (!stopped()) {
4758 // No need for PreserveJVMState, because we're using up the present state.
4759 set_all_memory(init_mem);
4760 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4761 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static);
4762 Node* slow_result = set_results_for_java_call(slow_call);
4763 // this->control() comes from set_results_for_java_call
4764 result_reg->init_req(_slow_path, control());
4765 result_val->init_req(_slow_path, slow_result);
4766 result_io ->set_req(_slow_path, i_o());
4767 result_mem ->set_req(_slow_path, reset_memory());
4768 }
4769
4770 // Return the combined state.
4771 set_i_o( _gvn.transform(result_io) );
4772 set_all_memory( _gvn.transform(result_mem));
4773
4774 set_result(result_reg, result_val);
4775 return true;
4776 }
4777
4778 //---------------------------inline_native_getClass----------------------------
4779 // public final native Class<?> java.lang.Object.getClass();
4780 //
4781 // Build special case code for calls to getClass on an object.
4782 bool LibraryCallKit::inline_native_getClass() {
4783 Node* obj = argument(0);
4784 if (obj->is_InlineType()) {
4785 const Type* t = _gvn.type(obj);
4786 if (t->maybe_null()) {
4787 null_check(obj);
4788 }
4789 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
4790 return true;
4791 }
4792 obj = null_check_receiver();
4793 if (stopped()) return true;
4794 set_result(load_mirror_from_klass(load_object_klass(obj)));
4795 return true;
4796 }
4797
4798 //-----------------inline_native_Reflection_getCallerClass---------------------
4799 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4800 //
4801 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4802 //
4803 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4804 // in that it must skip particular security frames and checks for
4805 // caller sensitive methods.
4806 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4807 #ifndef PRODUCT
4808 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4809 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4810 }
4811 #endif
4812
4813 if (!jvms()->has_method()) {
4814 #ifndef PRODUCT
4815 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4816 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4817 }
4818 #endif
4819 return false;
4820 }
4821
4822 // Walk back up the JVM state to find the caller at the required
4823 // depth.
4824 JVMState* caller_jvms = jvms();
4825
4826 // Cf. JVM_GetCallerClass
4827 // NOTE: Start the loop at depth 1 because the current JVM state does
4828 // not include the Reflection.getCallerClass() frame.
4829 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4830 ciMethod* m = caller_jvms->method();
4831 switch (n) {
4832 case 0:
4833 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4834 break;
4835 case 1:
4836 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4837 if (!m->caller_sensitive()) {
4838 #ifndef PRODUCT
4839 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4840 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4841 }
4842 #endif
4843 return false; // bail-out; let JVM_GetCallerClass do the work
4844 }
4845 break;
4846 default:
4847 if (!m->is_ignored_by_security_stack_walk()) {
4848 // We have reached the desired frame; return the holder class.
4849 // Acquire method holder as java.lang.Class and push as constant.
4850 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4851 ciInstance* caller_mirror = caller_klass->java_mirror();
4852 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4853
4854 #ifndef PRODUCT
4855 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4856 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());
4857 tty->print_cr(" JVM state at this point:");
4858 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4859 ciMethod* m = jvms()->of_depth(i)->method();
4860 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4861 }
4862 }
4863 #endif
4864 return true;
4865 }
4866 break;
4867 }
4868 }
4869
4870 #ifndef PRODUCT
4871 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4872 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4873 tty->print_cr(" JVM state at this point:");
4874 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4875 ciMethod* m = jvms()->of_depth(i)->method();
4876 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4877 }
4878 }
4879 #endif
4880
4881 return false; // bail-out; let JVM_GetCallerClass do the work
4882 }
4883
4884 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4885 Node* arg = argument(0);
4886 Node* result = nullptr;
4887
4888 switch (id) {
4889 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4890 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4891 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4892 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4893 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4894 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4895
4896 case vmIntrinsics::_doubleToLongBits: {
4897 // two paths (plus control) merge in a wood
4898 RegionNode *r = new RegionNode(3);
4899 Node *phi = new PhiNode(r, TypeLong::LONG);
4900
4901 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4902 // Build the boolean node
4903 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4904
4905 // Branch either way.
4906 // NaN case is less traveled, which makes all the difference.
4907 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4908 Node *opt_isnan = _gvn.transform(ifisnan);
4909 assert( opt_isnan->is_If(), "Expect an IfNode");
4910 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4911 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4912
4913 set_control(iftrue);
4914
4915 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4916 Node *slow_result = longcon(nan_bits); // return NaN
4917 phi->init_req(1, _gvn.transform( slow_result ));
4918 r->init_req(1, iftrue);
4919
4920 // Else fall through
4921 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4922 set_control(iffalse);
4923
4924 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4925 r->init_req(2, iffalse);
4926
4927 // Post merge
4928 set_control(_gvn.transform(r));
4929 record_for_igvn(r);
4930
4931 C->set_has_split_ifs(true); // Has chance for split-if optimization
4932 result = phi;
4933 assert(result->bottom_type()->isa_long(), "must be");
4934 break;
4935 }
4936
4937 case vmIntrinsics::_floatToIntBits: {
4938 // two paths (plus control) merge in a wood
4939 RegionNode *r = new RegionNode(3);
4940 Node *phi = new PhiNode(r, TypeInt::INT);
4941
4942 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4943 // Build the boolean node
4944 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4945
4946 // Branch either way.
4947 // NaN case is less traveled, which makes all the difference.
4948 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4949 Node *opt_isnan = _gvn.transform(ifisnan);
4950 assert( opt_isnan->is_If(), "Expect an IfNode");
4951 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4952 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4953
4954 set_control(iftrue);
4955
4956 static const jint nan_bits = 0x7fc00000;
4957 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4958 phi->init_req(1, _gvn.transform( slow_result ));
4959 r->init_req(1, iftrue);
4960
4961 // Else fall through
4962 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4963 set_control(iffalse);
4964
4965 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4966 r->init_req(2, iffalse);
4967
4968 // Post merge
4969 set_control(_gvn.transform(r));
4970 record_for_igvn(r);
4971
4972 C->set_has_split_ifs(true); // Has chance for split-if optimization
4973 result = phi;
4974 assert(result->bottom_type()->isa_int(), "must be");
4975 break;
4976 }
4977
4978 default:
4979 fatal_unexpected_iid(id);
4980 break;
4981 }
4982 set_result(_gvn.transform(result));
4983 return true;
4984 }
4985
4986 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4987 Node* arg = argument(0);
4988 Node* result = nullptr;
4989
4990 switch (id) {
4991 case vmIntrinsics::_floatIsInfinite:
4992 result = new IsInfiniteFNode(arg);
4993 break;
4994 case vmIntrinsics::_floatIsFinite:
4995 result = new IsFiniteFNode(arg);
4996 break;
4997 case vmIntrinsics::_doubleIsInfinite:
4998 result = new IsInfiniteDNode(arg);
4999 break;
5000 case vmIntrinsics::_doubleIsFinite:
5001 result = new IsFiniteDNode(arg);
5002 break;
5003 default:
5004 fatal_unexpected_iid(id);
5005 break;
5006 }
5007 set_result(_gvn.transform(result));
5008 return true;
5009 }
5010
5011 //----------------------inline_unsafe_copyMemory-------------------------
5012 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5013
5014 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5015 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5016 const Type* base_t = gvn.type(base);
5017
5018 bool in_native = (base_t == TypePtr::NULL_PTR);
5019 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5020 bool is_mixed = !in_heap && !in_native;
5021
5022 if (is_mixed) {
5023 return true; // mixed accesses can touch both on-heap and off-heap memory
5024 }
5025 if (in_heap) {
5026 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5027 if (!is_prim_array) {
5028 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5029 // there's not enough type information available to determine proper memory slice for it.
5030 return true;
5031 }
5032 }
5033 return false;
5034 }
5035
5036 bool LibraryCallKit::inline_unsafe_copyMemory() {
5037 if (callee()->is_static()) return false; // caller must have the capability!
5038 null_check_receiver(); // null-check receiver
5039 if (stopped()) return true;
5040
5041 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5042
5043 Node* src_base = argument(1); // type: oop
5044 Node* src_off = ConvL2X(argument(2)); // type: long
5045 Node* dst_base = argument(4); // type: oop
5046 Node* dst_off = ConvL2X(argument(5)); // type: long
5047 Node* size = ConvL2X(argument(7)); // type: long
5048
5049 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5050 "fieldOffset must be byte-scaled");
5051
5052 Node* src_addr = make_unsafe_address(src_base, src_off);
5053 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5054
5055 Node* thread = _gvn.transform(new ThreadLocalNode());
5056 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5057 BasicType doing_unsafe_access_bt = T_BYTE;
5058 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5059
5060 // update volatile field
5061 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5062
5063 int flags = RC_LEAF | RC_NO_FP;
5064
5065 const TypePtr* dst_type = TypePtr::BOTTOM;
5066
5067 // Adjust memory effects of the runtime call based on input values.
5068 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5069 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5070 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5071
5072 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5073 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5074 flags |= RC_NARROW_MEM; // narrow in memory
5075 }
5076 }
5077
5078 // Call it. Note that the length argument is not scaled.
5079 make_runtime_call(flags,
5080 OptoRuntime::fast_arraycopy_Type(),
5081 StubRoutines::unsafe_arraycopy(),
5082 "unsafe_arraycopy",
5083 dst_type,
5084 src_addr, dst_addr, size XTOP);
5085
5086 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5087
5088 return true;
5089 }
5090
5091 #undef XTOP
5092
5093 //------------------------clone_coping-----------------------------------
5094 // Helper function for inline_native_clone.
5095 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5096 assert(obj_size != nullptr, "");
5097 Node* raw_obj = alloc_obj->in(1);
5098 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5099
5100 AllocateNode* alloc = nullptr;
5101 if (ReduceBulkZeroing) {
5102 // We will be completely responsible for initializing this object -
5103 // mark Initialize node as complete.
5104 alloc = AllocateNode::Ideal_allocation(alloc_obj, &_gvn);
5105 // The object was just allocated - there should be no any stores!
5106 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5107 // Mark as complete_with_arraycopy so that on AllocateNode
5108 // expansion, we know this AllocateNode is initialized by an array
5109 // copy and a StoreStore barrier exists after the array copy.
5110 alloc->initialization()->set_complete_with_arraycopy();
5111 }
5112
5113 Node* size = _gvn.transform(obj_size);
5114 access_clone(obj, alloc_obj, size, is_array);
5115
5116 // Do not let reads from the cloned object float above the arraycopy.
5117 if (alloc != nullptr) {
5118 // Do not let stores that initialize this object be reordered with
5119 // a subsequent store that would make this object accessible by
5120 // other threads.
5121 // Record what AllocateNode this StoreStore protects so that
5122 // escape analysis can go from the MemBarStoreStoreNode to the
5123 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5124 // based on the escape status of the AllocateNode.
5125 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5126 } else {
5127 insert_mem_bar(Op_MemBarCPUOrder);
5128 }
5129 }
5130
5131 //------------------------inline_native_clone----------------------------
5132 // protected native Object java.lang.Object.clone();
5133 //
5134 // Here are the simple edge cases:
5135 // null receiver => normal trap
5136 // virtual and clone was overridden => slow path to out-of-line clone
5137 // not cloneable or finalizer => slow path to out-of-line Object.clone
5138 //
5139 // The general case has two steps, allocation and copying.
5140 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5141 //
5142 // Copying also has two cases, oop arrays and everything else.
5143 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5144 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5145 //
5146 // These steps fold up nicely if and when the cloned object's klass
5147 // can be sharply typed as an object array, a type array, or an instance.
5148 //
5149 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5150 PhiNode* result_val;
5151
5152 // Set the reexecute bit for the interpreter to reexecute
5153 // the bytecode that invokes Object.clone if deoptimization happens.
5154 { PreserveReexecuteState preexecs(this);
5155 jvms()->set_should_reexecute(true);
5156
5157 Node* obj = argument(0);
5158 obj = null_check_receiver();
5159 if (stopped()) return true;
5160
5161 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5162
5163 // If we are going to clone an instance, we need its exact type to
5164 // know the number and types of fields to convert the clone to
5165 // loads/stores. Maybe a speculative type can help us.
5166 if (!obj_type->klass_is_exact() &&
5167 obj_type->speculative_type() != nullptr &&
5168 obj_type->speculative_type()->is_instance_klass() &&
5169 !obj_type->speculative_type()->is_inlinetype()) {
5170 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5171 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5172 !spec_ik->has_injected_fields()) {
5173 if (!obj_type->isa_instptr() ||
5174 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5175 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5176 }
5177 }
5178 }
5179
5180 // Conservatively insert a memory barrier on all memory slices.
5181 // Do not let writes into the original float below the clone.
5182 insert_mem_bar(Op_MemBarCPUOrder);
5183
5184 // paths into result_reg:
5185 enum {
5186 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5187 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5188 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5189 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5190 PATH_LIMIT
5191 };
5192 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5193 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5194 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5195 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5196 record_for_igvn(result_reg);
5197
5198 Node* obj_klass = load_object_klass(obj);
5199 // We only go to the fast case code if we pass a number of guards.
5200 // The paths which do not pass are accumulated in the slow_region.
5201 RegionNode* slow_region = new RegionNode(1);
5202 record_for_igvn(slow_region);
5203
5204 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5205 if (array_ctl != nullptr) {
5206 // It's an array.
5207 PreserveJVMState pjvms(this);
5208 set_control(array_ctl);
5209
5210 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5211 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5212 if (UseFlatArray && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5213 obj_type->can_be_inline_array() &&
5214 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5215 // Flattened inline type array may have object field that would require a
5216 // write barrier. Conservatively, go to slow path.
5217 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5218 }
5219
5220 if (!stopped()) {
5221 Node* obj_length = load_array_length(obj);
5222 Node* obj_size = nullptr;
5223 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
5224
5225 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5226 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5227 // If it is an oop array, it requires very special treatment,
5228 // because gc barriers are required when accessing the array.
5229 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5230 if (is_obja != nullptr) {
5231 PreserveJVMState pjvms2(this);
5232 set_control(is_obja);
5233 // Generate a direct call to the right arraycopy function(s).
5234 // Clones are always tightly coupled.
5235 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5236 ac->set_clone_oop_array();
5237 Node* n = _gvn.transform(ac);
5238 assert(n == ac, "cannot disappear");
5239 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5240
5241 result_reg->init_req(_objArray_path, control());
5242 result_val->init_req(_objArray_path, alloc_obj);
5243 result_i_o ->set_req(_objArray_path, i_o());
5244 result_mem ->set_req(_objArray_path, reset_memory());
5245 }
5246 }
5247 // Otherwise, there are no barriers to worry about.
5248 // (We can dispense with card marks if we know the allocation
5249 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5250 // causes the non-eden paths to take compensating steps to
5251 // simulate a fresh allocation, so that no further
5252 // card marks are required in compiled code to initialize
5253 // the object.)
5254
5255 if (!stopped()) {
5256 copy_to_clone(obj, alloc_obj, obj_size, true);
5257
5258 // Present the results of the copy.
5259 result_reg->init_req(_array_path, control());
5260 result_val->init_req(_array_path, alloc_obj);
5261 result_i_o ->set_req(_array_path, i_o());
5262 result_mem ->set_req(_array_path, reset_memory());
5263 }
5264 }
5265 }
5266
5267 if (!stopped()) {
5268 // It's an instance (we did array above). Make the slow-path tests.
5269 // If this is a virtual call, we generate a funny guard. We grab
5270 // the vtable entry corresponding to clone() from the target object.
5271 // If the target method which we are calling happens to be the
5272 // Object clone() method, we pass the guard. We do not need this
5273 // guard for non-virtual calls; the caller is known to be the native
5274 // Object clone().
5275 if (is_virtual) {
5276 generate_virtual_guard(obj_klass, slow_region);
5277 }
5278
5279 // The object must be easily cloneable and must not have a finalizer.
5280 // Both of these conditions may be checked in a single test.
5281 // We could optimize the test further, but we don't care.
5282 generate_access_flags_guard(obj_klass,
5283 // Test both conditions:
5284 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5285 // Must be cloneable but not finalizer:
5286 JVM_ACC_IS_CLONEABLE_FAST,
5287 slow_region);
5288 }
5289
5290 if (!stopped()) {
5291 // It's an instance, and it passed the slow-path tests.
5292 PreserveJVMState pjvms(this);
5293 Node* obj_size = nullptr;
5294 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5295 // is reexecuted if deoptimization occurs and there could be problems when merging
5296 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5297 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5298
5299 copy_to_clone(obj, alloc_obj, obj_size, false);
5300
5301 // Present the results of the slow call.
5302 result_reg->init_req(_instance_path, control());
5303 result_val->init_req(_instance_path, alloc_obj);
5304 result_i_o ->set_req(_instance_path, i_o());
5305 result_mem ->set_req(_instance_path, reset_memory());
5306 }
5307
5308 // Generate code for the slow case. We make a call to clone().
5309 set_control(_gvn.transform(slow_region));
5310 if (!stopped()) {
5311 PreserveJVMState pjvms(this);
5312 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual);
5313 // We need to deoptimize on exception (see comment above)
5314 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5315 // this->control() comes from set_results_for_java_call
5316 result_reg->init_req(_slow_path, control());
5317 result_val->init_req(_slow_path, slow_result);
5318 result_i_o ->set_req(_slow_path, i_o());
5319 result_mem ->set_req(_slow_path, reset_memory());
5320 }
5321
5322 // Return the combined state.
5323 set_control( _gvn.transform(result_reg));
5324 set_i_o( _gvn.transform(result_i_o));
5325 set_all_memory( _gvn.transform(result_mem));
5326 } // original reexecute is set back here
5327
5328 set_result(_gvn.transform(result_val));
5329 return true;
5330 }
5331
5332 // If we have a tightly coupled allocation, the arraycopy may take care
5333 // of the array initialization. If one of the guards we insert between
5334 // the allocation and the arraycopy causes a deoptimization, an
5335 // uninitialized array will escape the compiled method. To prevent that
5336 // we set the JVM state for uncommon traps between the allocation and
5337 // the arraycopy to the state before the allocation so, in case of
5338 // deoptimization, we'll reexecute the allocation and the
5339 // initialization.
5340 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5341 if (alloc != nullptr) {
5342 ciMethod* trap_method = alloc->jvms()->method();
5343 int trap_bci = alloc->jvms()->bci();
5344
5345 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5346 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5347 // Make sure there's no store between the allocation and the
5348 // arraycopy otherwise visible side effects could be rexecuted
5349 // in case of deoptimization and cause incorrect execution.
5350 bool no_interfering_store = true;
5351 Node* mem = alloc->in(TypeFunc::Memory);
5352 if (mem->is_MergeMem()) {
5353 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5354 Node* n = mms.memory();
5355 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5356 assert(n->is_Store(), "what else?");
5357 no_interfering_store = false;
5358 break;
5359 }
5360 }
5361 } else {
5362 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5363 Node* n = mms.memory();
5364 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5365 assert(n->is_Store(), "what else?");
5366 no_interfering_store = false;
5367 break;
5368 }
5369 }
5370 }
5371
5372 if (no_interfering_store) {
5373 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5374
5375 JVMState* saved_jvms = jvms();
5376 saved_reexecute_sp = _reexecute_sp;
5377
5378 set_jvms(sfpt->jvms());
5379 _reexecute_sp = jvms()->sp();
5380
5381 return saved_jvms;
5382 }
5383 }
5384 }
5385 return nullptr;
5386 }
5387
5388 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5389 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5390 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5391 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5392 uint size = alloc->req();
5393 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5394 old_jvms->set_map(sfpt);
5395 for (uint i = 0; i < size; i++) {
5396 sfpt->init_req(i, alloc->in(i));
5397 }
5398 // re-push array length for deoptimization
5399 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5400 old_jvms->set_sp(old_jvms->sp()+1);
5401 old_jvms->set_monoff(old_jvms->monoff()+1);
5402 old_jvms->set_scloff(old_jvms->scloff()+1);
5403 old_jvms->set_endoff(old_jvms->endoff()+1);
5404 old_jvms->set_should_reexecute(true);
5405
5406 sfpt->set_i_o(map()->i_o());
5407 sfpt->set_memory(map()->memory());
5408 sfpt->set_control(map()->control());
5409 return sfpt;
5410 }
5411
5412 // In case of a deoptimization, we restart execution at the
5413 // allocation, allocating a new array. We would leave an uninitialized
5414 // array in the heap that GCs wouldn't expect. Move the allocation
5415 // after the traps so we don't allocate the array if we
5416 // deoptimize. This is possible because tightly_coupled_allocation()
5417 // guarantees there's no observer of the allocated array at this point
5418 // and the control flow is simple enough.
5419 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5420 int saved_reexecute_sp, uint new_idx) {
5421 if (saved_jvms_before_guards != nullptr && !stopped()) {
5422 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5423
5424 assert(alloc != nullptr, "only with a tightly coupled allocation");
5425 // restore JVM state to the state at the arraycopy
5426 saved_jvms_before_guards->map()->set_control(map()->control());
5427 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5428 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5429 // If we've improved the types of some nodes (null check) while
5430 // emitting the guards, propagate them to the current state
5431 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5432 set_jvms(saved_jvms_before_guards);
5433 _reexecute_sp = saved_reexecute_sp;
5434
5435 // Remove the allocation from above the guards
5436 CallProjections* callprojs = alloc->extract_projections(true);
5437 InitializeNode* init = alloc->initialization();
5438 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5439 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5440 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5441
5442 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5443 // the allocation (i.e. is only valid if the allocation succeeds):
5444 // 1) replace CastIINode with AllocateArrayNode's length here
5445 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5446 //
5447 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5448 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5449 Node* init_control = init->proj_out(TypeFunc::Control);
5450 Node* alloc_length = alloc->Ideal_length();
5451 #ifdef ASSERT
5452 Node* prev_cast = nullptr;
5453 #endif
5454 for (uint i = 0; i < init_control->outcnt(); i++) {
5455 Node* init_out = init_control->raw_out(i);
5456 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5457 #ifdef ASSERT
5458 if (prev_cast == nullptr) {
5459 prev_cast = init_out;
5460 } else {
5461 if (prev_cast->cmp(*init_out) == false) {
5462 prev_cast->dump();
5463 init_out->dump();
5464 assert(false, "not equal CastIINode");
5465 }
5466 }
5467 #endif
5468 C->gvn_replace_by(init_out, alloc_length);
5469 }
5470 }
5471 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5472
5473 // move the allocation here (after the guards)
5474 _gvn.hash_delete(alloc);
5475 alloc->set_req(TypeFunc::Control, control());
5476 alloc->set_req(TypeFunc::I_O, i_o());
5477 Node *mem = reset_memory();
5478 set_all_memory(mem);
5479 alloc->set_req(TypeFunc::Memory, mem);
5480 set_control(init->proj_out_or_null(TypeFunc::Control));
5481 set_i_o(callprojs->fallthrough_ioproj);
5482
5483 // Update memory as done in GraphKit::set_output_for_allocation()
5484 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5485 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5486 if (ary_type->isa_aryptr() && length_type != nullptr) {
5487 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5488 }
5489 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5490 int elemidx = C->get_alias_index(telemref);
5491 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5492 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5493
5494 Node* allocx = _gvn.transform(alloc);
5495 assert(allocx == alloc, "where has the allocation gone?");
5496 assert(dest->is_CheckCastPP(), "not an allocation result?");
5497
5498 _gvn.hash_delete(dest);
5499 dest->set_req(0, control());
5500 Node* destx = _gvn.transform(dest);
5501 assert(destx == dest, "where has the allocation result gone?");
5502
5503 array_ideal_length(alloc, ary_type, true);
5504 }
5505 }
5506
5507 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5508 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5509 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5510 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5511 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5512 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5513 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5514 JVMState* saved_jvms_before_guards) {
5515 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5516 // There is at least one unrelated uncommon trap which needs to be replaced.
5517 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5518
5519 JVMState* saved_jvms = jvms();
5520 const int saved_reexecute_sp = _reexecute_sp;
5521 set_jvms(sfpt->jvms());
5522 _reexecute_sp = jvms()->sp();
5523
5524 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5525
5526 // Restore state
5527 set_jvms(saved_jvms);
5528 _reexecute_sp = saved_reexecute_sp;
5529 }
5530 }
5531
5532 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5533 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5534 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5535 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5536 while (if_proj->is_IfProj()) {
5537 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5538 if (uncommon_trap != nullptr) {
5539 create_new_uncommon_trap(uncommon_trap);
5540 }
5541 assert(if_proj->in(0)->is_If(), "must be If");
5542 if_proj = if_proj->in(0)->in(0);
5543 }
5544 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5545 "must have reached control projection of init node");
5546 }
5547
5548 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5549 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5550 assert(trap_request != 0, "no valid UCT trap request");
5551 PreserveJVMState pjvms(this);
5552 set_control(uncommon_trap_call->in(0));
5553 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5554 Deoptimization::trap_request_action(trap_request));
5555 assert(stopped(), "Should be stopped");
5556 _gvn.hash_delete(uncommon_trap_call);
5557 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5558 }
5559
5560 //------------------------------inline_arraycopy-----------------------
5561 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5562 // Object dest, int destPos,
5563 // int length);
5564 bool LibraryCallKit::inline_arraycopy() {
5565 // Get the arguments.
5566 Node* src = argument(0); // type: oop
5567 Node* src_offset = argument(1); // type: int
5568 Node* dest = argument(2); // type: oop
5569 Node* dest_offset = argument(3); // type: int
5570 Node* length = argument(4); // type: int
5571
5572 uint new_idx = C->unique();
5573
5574 // Check for allocation before we add nodes that would confuse
5575 // tightly_coupled_allocation()
5576 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5577
5578 int saved_reexecute_sp = -1;
5579 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5580 // See arraycopy_restore_alloc_state() comment
5581 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5582 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5583 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5584 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5585
5586 // The following tests must be performed
5587 // (1) src and dest are arrays.
5588 // (2) src and dest arrays must have elements of the same BasicType
5589 // (3) src and dest must not be null.
5590 // (4) src_offset must not be negative.
5591 // (5) dest_offset must not be negative.
5592 // (6) length must not be negative.
5593 // (7) src_offset + length must not exceed length of src.
5594 // (8) dest_offset + length must not exceed length of dest.
5595 // (9) each element of an oop array must be assignable
5596
5597 // (3) src and dest must not be null.
5598 // always do this here because we need the JVM state for uncommon traps
5599 Node* null_ctl = top();
5600 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5601 assert(null_ctl->is_top(), "no null control here");
5602 dest = null_check(dest, T_ARRAY);
5603
5604 if (!can_emit_guards) {
5605 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5606 // guards but the arraycopy node could still take advantage of a
5607 // tightly allocated allocation. tightly_coupled_allocation() is
5608 // called again to make sure it takes the null check above into
5609 // account: the null check is mandatory and if it caused an
5610 // uncommon trap to be emitted then the allocation can't be
5611 // considered tightly coupled in this context.
5612 alloc = tightly_coupled_allocation(dest);
5613 }
5614
5615 bool validated = false;
5616
5617 const Type* src_type = _gvn.type(src);
5618 const Type* dest_type = _gvn.type(dest);
5619 const TypeAryPtr* top_src = src_type->isa_aryptr();
5620 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5621
5622 // Do we have the type of src?
5623 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5624 // Do we have the type of dest?
5625 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5626 // Is the type for src from speculation?
5627 bool src_spec = false;
5628 // Is the type for dest from speculation?
5629 bool dest_spec = false;
5630
5631 if ((!has_src || !has_dest) && can_emit_guards) {
5632 // We don't have sufficient type information, let's see if
5633 // speculative types can help. We need to have types for both src
5634 // and dest so that it pays off.
5635
5636 // Do we already have or could we have type information for src
5637 bool could_have_src = has_src;
5638 // Do we already have or could we have type information for dest
5639 bool could_have_dest = has_dest;
5640
5641 ciKlass* src_k = nullptr;
5642 if (!has_src) {
5643 src_k = src_type->speculative_type_not_null();
5644 if (src_k != nullptr && src_k->is_array_klass()) {
5645 could_have_src = true;
5646 }
5647 }
5648
5649 ciKlass* dest_k = nullptr;
5650 if (!has_dest) {
5651 dest_k = dest_type->speculative_type_not_null();
5652 if (dest_k != nullptr && dest_k->is_array_klass()) {
5653 could_have_dest = true;
5654 }
5655 }
5656
5657 if (could_have_src && could_have_dest) {
5658 // This is going to pay off so emit the required guards
5659 if (!has_src) {
5660 src = maybe_cast_profiled_obj(src, src_k, true);
5661 src_type = _gvn.type(src);
5662 top_src = src_type->isa_aryptr();
5663 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5664 src_spec = true;
5665 }
5666 if (!has_dest) {
5667 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5668 dest_type = _gvn.type(dest);
5669 top_dest = dest_type->isa_aryptr();
5670 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5671 dest_spec = true;
5672 }
5673 }
5674 }
5675
5676 if (has_src && has_dest && can_emit_guards) {
5677 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5678 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5679 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5680 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5681
5682 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
5683 // If both arrays are object arrays then having the exact types
5684 // for both will remove the need for a subtype check at runtime
5685 // before the call and may make it possible to pick a faster copy
5686 // routine (without a subtype check on every element)
5687 // Do we have the exact type of src?
5688 bool could_have_src = src_spec;
5689 // Do we have the exact type of dest?
5690 bool could_have_dest = dest_spec;
5691 ciKlass* src_k = nullptr;
5692 ciKlass* dest_k = nullptr;
5693 if (!src_spec) {
5694 src_k = src_type->speculative_type_not_null();
5695 if (src_k != nullptr && src_k->is_array_klass()) {
5696 could_have_src = true;
5697 }
5698 }
5699 if (!dest_spec) {
5700 dest_k = dest_type->speculative_type_not_null();
5701 if (dest_k != nullptr && dest_k->is_array_klass()) {
5702 could_have_dest = true;
5703 }
5704 }
5705 if (could_have_src && could_have_dest) {
5706 // If we can have both exact types, emit the missing guards
5707 if (could_have_src && !src_spec) {
5708 src = maybe_cast_profiled_obj(src, src_k, true);
5709 src_type = _gvn.type(src);
5710 top_src = src_type->isa_aryptr();
5711 }
5712 if (could_have_dest && !dest_spec) {
5713 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5714 dest_type = _gvn.type(dest);
5715 top_dest = dest_type->isa_aryptr();
5716 }
5717 }
5718 }
5719 }
5720
5721 ciMethod* trap_method = method();
5722 int trap_bci = bci();
5723 if (saved_jvms_before_guards != nullptr) {
5724 trap_method = alloc->jvms()->method();
5725 trap_bci = alloc->jvms()->bci();
5726 }
5727
5728 bool negative_length_guard_generated = false;
5729
5730 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5731 can_emit_guards && !src->is_top() && !dest->is_top()) {
5732 // validate arguments: enables transformation the ArrayCopyNode
5733 validated = true;
5734
5735 RegionNode* slow_region = new RegionNode(1);
5736 record_for_igvn(slow_region);
5737
5738 // (1) src and dest are arrays.
5739 generate_non_array_guard(load_object_klass(src), slow_region);
5740 generate_non_array_guard(load_object_klass(dest), slow_region);
5741
5742 // (2) src and dest arrays must have elements of the same BasicType
5743 // done at macro expansion or at Ideal transformation time
5744
5745 // (4) src_offset must not be negative.
5746 generate_negative_guard(src_offset, slow_region);
5747
5748 // (5) dest_offset must not be negative.
5749 generate_negative_guard(dest_offset, slow_region);
5750
5751 // (7) src_offset + length must not exceed length of src.
5752 generate_limit_guard(src_offset, length,
5753 load_array_length(src),
5754 slow_region);
5755
5756 // (8) dest_offset + length must not exceed length of dest.
5757 generate_limit_guard(dest_offset, length,
5758 load_array_length(dest),
5759 slow_region);
5760
5761 // (6) length must not be negative.
5762 // This is also checked in generate_arraycopy() during macro expansion, but
5763 // we also have to check it here for the case where the ArrayCopyNode will
5764 // be eliminated by Escape Analysis.
5765 if (EliminateAllocations) {
5766 generate_negative_guard(length, slow_region);
5767 negative_length_guard_generated = true;
5768 }
5769
5770 // (9) each element of an oop array must be assignable
5771 Node* dest_klass = load_object_klass(dest);
5772 if (src != dest) {
5773 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5774 slow_region->add_req(not_subtype_ctrl);
5775 }
5776
5777 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5778 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5779 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5780 src_type = _gvn.type(src);
5781 top_src = src_type->isa_aryptr();
5782
5783 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
5784 if (!stopped() && UseFlatArray) {
5785 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
5786 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
5787 if (top_src != nullptr && top_src->is_flat()) {
5788 // Src is flat, check that dest is flat as well
5789 if (top_dest != nullptr && !top_dest->is_flat()) {
5790 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
5791 // Since dest is flat and src <: dest, dest must have the same type as src.
5792 top_dest = top_src->cast_to_exactness(false);
5793 assert(top_dest->is_flat(), "dest must be flat");
5794 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
5795 }
5796 } else if (top_src == nullptr || !top_src->is_not_flat()) {
5797 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5798 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5799 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
5800 generate_fair_guard(flat_array_test(src), slow_region);
5801 if (top_src != nullptr) {
5802 top_src = top_src->cast_to_not_flat();
5803 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
5804 }
5805 }
5806 }
5807
5808 {
5809 PreserveJVMState pjvms(this);
5810 set_control(_gvn.transform(slow_region));
5811 uncommon_trap(Deoptimization::Reason_intrinsic,
5812 Deoptimization::Action_make_not_entrant);
5813 assert(stopped(), "Should be stopped");
5814 }
5815 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5816 }
5817
5818 if (stopped()) {
5819 return true;
5820 }
5821
5822 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5823 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5824 // so the compiler has a chance to eliminate them: during macro expansion,
5825 // we have to set their control (CastPP nodes are eliminated).
5826 load_object_klass(src), load_object_klass(dest),
5827 load_array_length(src), load_array_length(dest));
5828
5829 ac->set_arraycopy(validated);
5830
5831 Node* n = _gvn.transform(ac);
5832 if (n == ac) {
5833 ac->connect_outputs(this);
5834 } else {
5835 assert(validated, "shouldn't transform if all arguments not validated");
5836 set_all_memory(n);
5837 }
5838 clear_upper_avx();
5839
5840
5841 return true;
5842 }
5843
5844
5845 // Helper function which determines if an arraycopy immediately follows
5846 // an allocation, with no intervening tests or other escapes for the object.
5847 AllocateArrayNode*
5848 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5849 if (stopped()) return nullptr; // no fast path
5850 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5851
5852 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr, &_gvn);
5853 if (alloc == nullptr) return nullptr;
5854
5855 Node* rawmem = memory(Compile::AliasIdxRaw);
5856 // Is the allocation's memory state untouched?
5857 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5858 // Bail out if there have been raw-memory effects since the allocation.
5859 // (Example: There might have been a call or safepoint.)
5860 return nullptr;
5861 }
5862 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5863 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5864 return nullptr;
5865 }
5866
5867 // There must be no unexpected observers of this allocation.
5868 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5869 Node* obs = ptr->fast_out(i);
5870 if (obs != this->map()) {
5871 return nullptr;
5872 }
5873 }
5874
5875 // This arraycopy must unconditionally follow the allocation of the ptr.
5876 Node* alloc_ctl = ptr->in(0);
5877 Node* ctl = control();
5878 while (ctl != alloc_ctl) {
5879 // There may be guards which feed into the slow_region.
5880 // Any other control flow means that we might not get a chance
5881 // to finish initializing the allocated object.
5882 // Various low-level checks bottom out in uncommon traps. These
5883 // are considered safe since we've already checked above that
5884 // there is no unexpected observer of this allocation.
5885 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5886 assert(ctl->in(0)->is_If(), "must be If");
5887 ctl = ctl->in(0)->in(0);
5888 } else {
5889 return nullptr;
5890 }
5891 }
5892
5893 // If we get this far, we have an allocation which immediately
5894 // precedes the arraycopy, and we can take over zeroing the new object.
5895 // The arraycopy will finish the initialization, and provide
5896 // a new control state to which we will anchor the destination pointer.
5897
5898 return alloc;
5899 }
5900
5901 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5902 if (node->is_IfProj()) {
5903 Node* other_proj = node->as_IfProj()->other_if_proj();
5904 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5905 Node* obs = other_proj->fast_out(j);
5906 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5907 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5908 return obs->as_CallStaticJava();
5909 }
5910 }
5911 }
5912 return nullptr;
5913 }
5914
5915 //-------------inline_encodeISOArray-----------------------------------
5916 // encode char[] to byte[] in ISO_8859_1 or ASCII
5917 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5918 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5919 // no receiver since it is static method
5920 Node *src = argument(0);
5921 Node *src_offset = argument(1);
5922 Node *dst = argument(2);
5923 Node *dst_offset = argument(3);
5924 Node *length = argument(4);
5925
5926 src = must_be_not_null(src, true);
5927 dst = must_be_not_null(dst, true);
5928
5929 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5930 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5931 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5932 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5933 // failed array check
5934 return false;
5935 }
5936
5937 // Figure out the size and type of the elements we will be copying.
5938 BasicType src_elem = src_type->elem()->array_element_basic_type();
5939 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5940 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5941 return false;
5942 }
5943
5944 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5945 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5946 // 'src_start' points to src array + scaled offset
5947 // 'dst_start' points to dst array + scaled offset
5948
5949 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5950 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5951 enc = _gvn.transform(enc);
5952 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5953 set_memory(res_mem, mtype);
5954 set_result(enc);
5955 clear_upper_avx();
5956
5957 return true;
5958 }
5959
5960 //-------------inline_multiplyToLen-----------------------------------
5961 bool LibraryCallKit::inline_multiplyToLen() {
5962 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5963
5964 address stubAddr = StubRoutines::multiplyToLen();
5965 if (stubAddr == nullptr) {
5966 return false; // Intrinsic's stub is not implemented on this platform
5967 }
5968 const char* stubName = "multiplyToLen";
5969
5970 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5971
5972 // no receiver because it is a static method
5973 Node* x = argument(0);
5974 Node* xlen = argument(1);
5975 Node* y = argument(2);
5976 Node* ylen = argument(3);
5977 Node* z = argument(4);
5978
5979 x = must_be_not_null(x, true);
5980 y = must_be_not_null(y, true);
5981
5982 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5983 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5984 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5985 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5986 // failed array check
5987 return false;
5988 }
5989
5990 BasicType x_elem = x_type->elem()->array_element_basic_type();
5991 BasicType y_elem = y_type->elem()->array_element_basic_type();
5992 if (x_elem != T_INT || y_elem != T_INT) {
5993 return false;
5994 }
5995
5996 // Set the original stack and the reexecute bit for the interpreter to reexecute
5997 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5998 // on the return from z array allocation in runtime.
5999 { PreserveReexecuteState preexecs(this);
6000 jvms()->set_should_reexecute(true);
6001
6002 Node* x_start = array_element_address(x, intcon(0), x_elem);
6003 Node* y_start = array_element_address(y, intcon(0), y_elem);
6004 // 'x_start' points to x array + scaled xlen
6005 // 'y_start' points to y array + scaled ylen
6006
6007 // Allocate the result array
6008 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
6009 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
6010 Node* klass_node = makecon(TypeKlassPtr::make(klass));
6011
6012 IdealKit ideal(this);
6013
6014 #define __ ideal.
6015 Node* one = __ ConI(1);
6016 Node* zero = __ ConI(0);
6017 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
6018 __ set(need_alloc, zero);
6019 __ set(z_alloc, z);
6020 __ if_then(z, BoolTest::eq, null()); {
6021 __ increment (need_alloc, one);
6022 } __ else_(); {
6023 // Update graphKit memory and control from IdealKit.
6024 sync_kit(ideal);
6025 Node *cast = new CastPPNode(z, TypePtr::NOTNULL);
6026 cast->init_req(0, control());
6027 _gvn.set_type(cast, cast->bottom_type());
6028 C->record_for_igvn(cast);
6029
6030 Node* zlen_arg = load_array_length(cast);
6031 // Update IdealKit memory and control from graphKit.
6032 __ sync_kit(this);
6033 __ if_then(zlen_arg, BoolTest::lt, zlen); {
6034 __ increment (need_alloc, one);
6035 } __ end_if();
6036 } __ end_if();
6037
6038 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
6039 // Update graphKit memory and control from IdealKit.
6040 sync_kit(ideal);
6041 Node * narr = new_array(klass_node, zlen, 1);
6042 // Update IdealKit memory and control from graphKit.
6043 __ sync_kit(this);
6044 __ set(z_alloc, narr);
6045 } __ end_if();
6046
6047 sync_kit(ideal);
6048 z = __ value(z_alloc);
6049 // Can't use TypeAryPtr::INTS which uses Bottom offset.
6050 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
6051 // Final sync IdealKit and GraphKit.
6052 final_sync(ideal);
6053 #undef __
6054
6055 Node* z_start = array_element_address(z, intcon(0), T_INT);
6056
6057 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6058 OptoRuntime::multiplyToLen_Type(),
6059 stubAddr, stubName, TypePtr::BOTTOM,
6060 x_start, xlen, y_start, ylen, z_start, zlen);
6061 } // original reexecute is set back here
6062
6063 C->set_has_split_ifs(true); // Has chance for split-if optimization
6064 set_result(z);
6065 return true;
6066 }
6067
6068 //-------------inline_squareToLen------------------------------------
6069 bool LibraryCallKit::inline_squareToLen() {
6070 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6071
6072 address stubAddr = StubRoutines::squareToLen();
6073 if (stubAddr == nullptr) {
6074 return false; // Intrinsic's stub is not implemented on this platform
6075 }
6076 const char* stubName = "squareToLen";
6077
6078 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6079
6080 Node* x = argument(0);
6081 Node* len = argument(1);
6082 Node* z = argument(2);
6083 Node* zlen = argument(3);
6084
6085 x = must_be_not_null(x, true);
6086 z = must_be_not_null(z, true);
6087
6088 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6089 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6090 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6091 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6092 // failed array check
6093 return false;
6094 }
6095
6096 BasicType x_elem = x_type->elem()->array_element_basic_type();
6097 BasicType z_elem = z_type->elem()->array_element_basic_type();
6098 if (x_elem != T_INT || z_elem != T_INT) {
6099 return false;
6100 }
6101
6102
6103 Node* x_start = array_element_address(x, intcon(0), x_elem);
6104 Node* z_start = array_element_address(z, intcon(0), z_elem);
6105
6106 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6107 OptoRuntime::squareToLen_Type(),
6108 stubAddr, stubName, TypePtr::BOTTOM,
6109 x_start, len, z_start, zlen);
6110
6111 set_result(z);
6112 return true;
6113 }
6114
6115 //-------------inline_mulAdd------------------------------------------
6116 bool LibraryCallKit::inline_mulAdd() {
6117 assert(UseMulAddIntrinsic, "not implemented on this platform");
6118
6119 address stubAddr = StubRoutines::mulAdd();
6120 if (stubAddr == nullptr) {
6121 return false; // Intrinsic's stub is not implemented on this platform
6122 }
6123 const char* stubName = "mulAdd";
6124
6125 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6126
6127 Node* out = argument(0);
6128 Node* in = argument(1);
6129 Node* offset = argument(2);
6130 Node* len = argument(3);
6131 Node* k = argument(4);
6132
6133 in = must_be_not_null(in, true);
6134 out = must_be_not_null(out, true);
6135
6136 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6137 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6138 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6139 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6140 // failed array check
6141 return false;
6142 }
6143
6144 BasicType out_elem = out_type->elem()->array_element_basic_type();
6145 BasicType in_elem = in_type->elem()->array_element_basic_type();
6146 if (out_elem != T_INT || in_elem != T_INT) {
6147 return false;
6148 }
6149
6150 Node* outlen = load_array_length(out);
6151 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6152 Node* out_start = array_element_address(out, intcon(0), out_elem);
6153 Node* in_start = array_element_address(in, intcon(0), in_elem);
6154
6155 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6156 OptoRuntime::mulAdd_Type(),
6157 stubAddr, stubName, TypePtr::BOTTOM,
6158 out_start,in_start, new_offset, len, k);
6159 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6160 set_result(result);
6161 return true;
6162 }
6163
6164 //-------------inline_montgomeryMultiply-----------------------------------
6165 bool LibraryCallKit::inline_montgomeryMultiply() {
6166 address stubAddr = StubRoutines::montgomeryMultiply();
6167 if (stubAddr == nullptr) {
6168 return false; // Intrinsic's stub is not implemented on this platform
6169 }
6170
6171 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6172 const char* stubName = "montgomery_multiply";
6173
6174 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6175
6176 Node* a = argument(0);
6177 Node* b = argument(1);
6178 Node* n = argument(2);
6179 Node* len = argument(3);
6180 Node* inv = argument(4);
6181 Node* m = argument(6);
6182
6183 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6184 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6185 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6186 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6187 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6188 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6189 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6190 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6191 // failed array check
6192 return false;
6193 }
6194
6195 BasicType a_elem = a_type->elem()->array_element_basic_type();
6196 BasicType b_elem = b_type->elem()->array_element_basic_type();
6197 BasicType n_elem = n_type->elem()->array_element_basic_type();
6198 BasicType m_elem = m_type->elem()->array_element_basic_type();
6199 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6200 return false;
6201 }
6202
6203 // Make the call
6204 {
6205 Node* a_start = array_element_address(a, intcon(0), a_elem);
6206 Node* b_start = array_element_address(b, intcon(0), b_elem);
6207 Node* n_start = array_element_address(n, intcon(0), n_elem);
6208 Node* m_start = array_element_address(m, intcon(0), m_elem);
6209
6210 Node* call = make_runtime_call(RC_LEAF,
6211 OptoRuntime::montgomeryMultiply_Type(),
6212 stubAddr, stubName, TypePtr::BOTTOM,
6213 a_start, b_start, n_start, len, inv, top(),
6214 m_start);
6215 set_result(m);
6216 }
6217
6218 return true;
6219 }
6220
6221 bool LibraryCallKit::inline_montgomerySquare() {
6222 address stubAddr = StubRoutines::montgomerySquare();
6223 if (stubAddr == nullptr) {
6224 return false; // Intrinsic's stub is not implemented on this platform
6225 }
6226
6227 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6228 const char* stubName = "montgomery_square";
6229
6230 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6231
6232 Node* a = argument(0);
6233 Node* n = argument(1);
6234 Node* len = argument(2);
6235 Node* inv = argument(3);
6236 Node* m = argument(5);
6237
6238 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6239 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6240 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6241 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6242 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6243 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6244 // failed array check
6245 return false;
6246 }
6247
6248 BasicType a_elem = a_type->elem()->array_element_basic_type();
6249 BasicType n_elem = n_type->elem()->array_element_basic_type();
6250 BasicType m_elem = m_type->elem()->array_element_basic_type();
6251 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6252 return false;
6253 }
6254
6255 // Make the call
6256 {
6257 Node* a_start = array_element_address(a, intcon(0), a_elem);
6258 Node* n_start = array_element_address(n, intcon(0), n_elem);
6259 Node* m_start = array_element_address(m, intcon(0), m_elem);
6260
6261 Node* call = make_runtime_call(RC_LEAF,
6262 OptoRuntime::montgomerySquare_Type(),
6263 stubAddr, stubName, TypePtr::BOTTOM,
6264 a_start, n_start, len, inv, top(),
6265 m_start);
6266 set_result(m);
6267 }
6268
6269 return true;
6270 }
6271
6272 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6273 address stubAddr = nullptr;
6274 const char* stubName = nullptr;
6275
6276 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6277 if (stubAddr == nullptr) {
6278 return false; // Intrinsic's stub is not implemented on this platform
6279 }
6280
6281 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6282
6283 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6284
6285 Node* newArr = argument(0);
6286 Node* oldArr = argument(1);
6287 Node* newIdx = argument(2);
6288 Node* shiftCount = argument(3);
6289 Node* numIter = argument(4);
6290
6291 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6292 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6293 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6294 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6295 return false;
6296 }
6297
6298 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6299 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6300 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6301 return false;
6302 }
6303
6304 // Make the call
6305 {
6306 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6307 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6308
6309 Node* call = make_runtime_call(RC_LEAF,
6310 OptoRuntime::bigIntegerShift_Type(),
6311 stubAddr,
6312 stubName,
6313 TypePtr::BOTTOM,
6314 newArr_start,
6315 oldArr_start,
6316 newIdx,
6317 shiftCount,
6318 numIter);
6319 }
6320
6321 return true;
6322 }
6323
6324 //-------------inline_vectorizedMismatch------------------------------
6325 bool LibraryCallKit::inline_vectorizedMismatch() {
6326 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6327
6328 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6329 Node* obja = argument(0); // Object
6330 Node* aoffset = argument(1); // long
6331 Node* objb = argument(3); // Object
6332 Node* boffset = argument(4); // long
6333 Node* length = argument(6); // int
6334 Node* scale = argument(7); // int
6335
6336 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6337 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6338 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6339 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6340 scale == top()) {
6341 return false; // failed input validation
6342 }
6343
6344 Node* obja_adr = make_unsafe_address(obja, aoffset);
6345 Node* objb_adr = make_unsafe_address(objb, boffset);
6346
6347 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6348 //
6349 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6350 // if (length <= inline_limit) {
6351 // inline_path:
6352 // vmask = VectorMaskGen length
6353 // vload1 = LoadVectorMasked obja, vmask
6354 // vload2 = LoadVectorMasked objb, vmask
6355 // result1 = VectorCmpMasked vload1, vload2, vmask
6356 // } else {
6357 // call_stub_path:
6358 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6359 // }
6360 // exit_block:
6361 // return Phi(result1, result2);
6362 //
6363 enum { inline_path = 1, // input is small enough to process it all at once
6364 stub_path = 2, // input is too large; call into the VM
6365 PATH_LIMIT = 3
6366 };
6367
6368 Node* exit_block = new RegionNode(PATH_LIMIT);
6369 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6370 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6371
6372 Node* call_stub_path = control();
6373
6374 BasicType elem_bt = T_ILLEGAL;
6375
6376 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6377 if (scale_t->is_con()) {
6378 switch (scale_t->get_con()) {
6379 case 0: elem_bt = T_BYTE; break;
6380 case 1: elem_bt = T_SHORT; break;
6381 case 2: elem_bt = T_INT; break;
6382 case 3: elem_bt = T_LONG; break;
6383
6384 default: elem_bt = T_ILLEGAL; break; // not supported
6385 }
6386 }
6387
6388 int inline_limit = 0;
6389 bool do_partial_inline = false;
6390
6391 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6392 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6393 do_partial_inline = inline_limit >= 16;
6394 }
6395
6396 if (do_partial_inline) {
6397 assert(elem_bt != T_ILLEGAL, "sanity");
6398
6399 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6400 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6401 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6402
6403 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6404 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6405 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6406
6407 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6408
6409 if (!stopped()) {
6410 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6411
6412 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6413 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6414 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6415 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6416
6417 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6418 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6419 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6420 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6421
6422 exit_block->init_req(inline_path, control());
6423 memory_phi->init_req(inline_path, map()->memory());
6424 result_phi->init_req(inline_path, result);
6425
6426 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6427 clear_upper_avx();
6428 }
6429 }
6430 }
6431
6432 if (call_stub_path != nullptr) {
6433 set_control(call_stub_path);
6434
6435 Node* call = make_runtime_call(RC_LEAF,
6436 OptoRuntime::vectorizedMismatch_Type(),
6437 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6438 obja_adr, objb_adr, length, scale);
6439
6440 exit_block->init_req(stub_path, control());
6441 memory_phi->init_req(stub_path, map()->memory());
6442 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6443 }
6444
6445 exit_block = _gvn.transform(exit_block);
6446 memory_phi = _gvn.transform(memory_phi);
6447 result_phi = _gvn.transform(result_phi);
6448
6449 set_control(exit_block);
6450 set_all_memory(memory_phi);
6451 set_result(result_phi);
6452
6453 return true;
6454 }
6455
6456 //------------------------------inline_vectorizedHashcode----------------------------
6457 bool LibraryCallKit::inline_vectorizedHashCode() {
6458 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6459
6460 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6461 Node* array = argument(0);
6462 Node* offset = argument(1);
6463 Node* length = argument(2);
6464 Node* initialValue = argument(3);
6465 Node* basic_type = argument(4);
6466
6467 if (basic_type == top()) {
6468 return false; // failed input validation
6469 }
6470
6471 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6472 if (!basic_type_t->is_con()) {
6473 return false; // Only intrinsify if mode argument is constant
6474 }
6475
6476 array = must_be_not_null(array, true);
6477
6478 BasicType bt = (BasicType)basic_type_t->get_con();
6479
6480 // Resolve address of first element
6481 Node* array_start = array_element_address(array, offset, bt);
6482
6483 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6484 array_start, length, initialValue, basic_type)));
6485 clear_upper_avx();
6486
6487 return true;
6488 }
6489
6490 /**
6491 * Calculate CRC32 for byte.
6492 * int java.util.zip.CRC32.update(int crc, int b)
6493 */
6494 bool LibraryCallKit::inline_updateCRC32() {
6495 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6496 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6497 // no receiver since it is static method
6498 Node* crc = argument(0); // type: int
6499 Node* b = argument(1); // type: int
6500
6501 /*
6502 * int c = ~ crc;
6503 * b = timesXtoThe32[(b ^ c) & 0xFF];
6504 * b = b ^ (c >>> 8);
6505 * crc = ~b;
6506 */
6507
6508 Node* M1 = intcon(-1);
6509 crc = _gvn.transform(new XorINode(crc, M1));
6510 Node* result = _gvn.transform(new XorINode(crc, b));
6511 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6512
6513 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6514 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6515 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6516 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6517
6518 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6519 result = _gvn.transform(new XorINode(crc, result));
6520 result = _gvn.transform(new XorINode(result, M1));
6521 set_result(result);
6522 return true;
6523 }
6524
6525 /**
6526 * Calculate CRC32 for byte[] array.
6527 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6528 */
6529 bool LibraryCallKit::inline_updateBytesCRC32() {
6530 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6531 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6532 // no receiver since it is static method
6533 Node* crc = argument(0); // type: int
6534 Node* src = argument(1); // type: oop
6535 Node* offset = argument(2); // type: int
6536 Node* length = argument(3); // type: int
6537
6538 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6539 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6540 // failed array check
6541 return false;
6542 }
6543
6544 // Figure out the size and type of the elements we will be copying.
6545 BasicType src_elem = src_type->elem()->array_element_basic_type();
6546 if (src_elem != T_BYTE) {
6547 return false;
6548 }
6549
6550 // 'src_start' points to src array + scaled offset
6551 src = must_be_not_null(src, true);
6552 Node* src_start = array_element_address(src, offset, src_elem);
6553
6554 // We assume that range check is done by caller.
6555 // TODO: generate range check (offset+length < src.length) in debug VM.
6556
6557 // Call the stub.
6558 address stubAddr = StubRoutines::updateBytesCRC32();
6559 const char *stubName = "updateBytesCRC32";
6560
6561 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6562 stubAddr, stubName, TypePtr::BOTTOM,
6563 crc, src_start, length);
6564 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6565 set_result(result);
6566 return true;
6567 }
6568
6569 /**
6570 * Calculate CRC32 for ByteBuffer.
6571 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6572 */
6573 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6574 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6575 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6576 // no receiver since it is static method
6577 Node* crc = argument(0); // type: int
6578 Node* src = argument(1); // type: long
6579 Node* offset = argument(3); // type: int
6580 Node* length = argument(4); // type: int
6581
6582 src = ConvL2X(src); // adjust Java long to machine word
6583 Node* base = _gvn.transform(new CastX2PNode(src));
6584 offset = ConvI2X(offset);
6585
6586 // 'src_start' points to src array + scaled offset
6587 Node* src_start = basic_plus_adr(top(), base, offset);
6588
6589 // Call the stub.
6590 address stubAddr = StubRoutines::updateBytesCRC32();
6591 const char *stubName = "updateBytesCRC32";
6592
6593 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6594 stubAddr, stubName, TypePtr::BOTTOM,
6595 crc, src_start, length);
6596 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6597 set_result(result);
6598 return true;
6599 }
6600
6601 //------------------------------get_table_from_crc32c_class-----------------------
6602 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6603 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6604 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6605
6606 return table;
6607 }
6608
6609 //------------------------------inline_updateBytesCRC32C-----------------------
6610 //
6611 // Calculate CRC32C for byte[] array.
6612 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6613 //
6614 bool LibraryCallKit::inline_updateBytesCRC32C() {
6615 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6616 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6617 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6618 // no receiver since it is a static method
6619 Node* crc = argument(0); // type: int
6620 Node* src = argument(1); // type: oop
6621 Node* offset = argument(2); // type: int
6622 Node* end = argument(3); // type: int
6623
6624 Node* length = _gvn.transform(new SubINode(end, offset));
6625
6626 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6627 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6628 // failed array check
6629 return false;
6630 }
6631
6632 // Figure out the size and type of the elements we will be copying.
6633 BasicType src_elem = src_type->elem()->array_element_basic_type();
6634 if (src_elem != T_BYTE) {
6635 return false;
6636 }
6637
6638 // 'src_start' points to src array + scaled offset
6639 src = must_be_not_null(src, true);
6640 Node* src_start = array_element_address(src, offset, src_elem);
6641
6642 // static final int[] byteTable in class CRC32C
6643 Node* table = get_table_from_crc32c_class(callee()->holder());
6644 table = must_be_not_null(table, true);
6645 Node* table_start = array_element_address(table, intcon(0), T_INT);
6646
6647 // We assume that range check is done by caller.
6648 // TODO: generate range check (offset+length < src.length) in debug VM.
6649
6650 // Call the stub.
6651 address stubAddr = StubRoutines::updateBytesCRC32C();
6652 const char *stubName = "updateBytesCRC32C";
6653
6654 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6655 stubAddr, stubName, TypePtr::BOTTOM,
6656 crc, src_start, length, table_start);
6657 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6658 set_result(result);
6659 return true;
6660 }
6661
6662 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6663 //
6664 // Calculate CRC32C for DirectByteBuffer.
6665 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6666 //
6667 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6668 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6669 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6670 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6671 // no receiver since it is a static method
6672 Node* crc = argument(0); // type: int
6673 Node* src = argument(1); // type: long
6674 Node* offset = argument(3); // type: int
6675 Node* end = argument(4); // type: int
6676
6677 Node* length = _gvn.transform(new SubINode(end, offset));
6678
6679 src = ConvL2X(src); // adjust Java long to machine word
6680 Node* base = _gvn.transform(new CastX2PNode(src));
6681 offset = ConvI2X(offset);
6682
6683 // 'src_start' points to src array + scaled offset
6684 Node* src_start = basic_plus_adr(top(), base, offset);
6685
6686 // static final int[] byteTable in class CRC32C
6687 Node* table = get_table_from_crc32c_class(callee()->holder());
6688 table = must_be_not_null(table, true);
6689 Node* table_start = array_element_address(table, intcon(0), T_INT);
6690
6691 // Call the stub.
6692 address stubAddr = StubRoutines::updateBytesCRC32C();
6693 const char *stubName = "updateBytesCRC32C";
6694
6695 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6696 stubAddr, stubName, TypePtr::BOTTOM,
6697 crc, src_start, length, table_start);
6698 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6699 set_result(result);
6700 return true;
6701 }
6702
6703 //------------------------------inline_updateBytesAdler32----------------------
6704 //
6705 // Calculate Adler32 checksum for byte[] array.
6706 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6707 //
6708 bool LibraryCallKit::inline_updateBytesAdler32() {
6709 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6710 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6711 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6712 // no receiver since it is static method
6713 Node* crc = argument(0); // type: int
6714 Node* src = argument(1); // type: oop
6715 Node* offset = argument(2); // type: int
6716 Node* length = argument(3); // type: int
6717
6718 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6719 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6720 // failed array check
6721 return false;
6722 }
6723
6724 // Figure out the size and type of the elements we will be copying.
6725 BasicType src_elem = src_type->elem()->array_element_basic_type();
6726 if (src_elem != T_BYTE) {
6727 return false;
6728 }
6729
6730 // 'src_start' points to src array + scaled offset
6731 Node* src_start = array_element_address(src, offset, src_elem);
6732
6733 // We assume that range check is done by caller.
6734 // TODO: generate range check (offset+length < src.length) in debug VM.
6735
6736 // Call the stub.
6737 address stubAddr = StubRoutines::updateBytesAdler32();
6738 const char *stubName = "updateBytesAdler32";
6739
6740 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6741 stubAddr, stubName, TypePtr::BOTTOM,
6742 crc, src_start, length);
6743 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6744 set_result(result);
6745 return true;
6746 }
6747
6748 //------------------------------inline_updateByteBufferAdler32---------------
6749 //
6750 // Calculate Adler32 checksum for DirectByteBuffer.
6751 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6752 //
6753 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6754 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6755 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6756 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6757 // no receiver since it is static method
6758 Node* crc = argument(0); // type: int
6759 Node* src = argument(1); // type: long
6760 Node* offset = argument(3); // type: int
6761 Node* length = argument(4); // type: int
6762
6763 src = ConvL2X(src); // adjust Java long to machine word
6764 Node* base = _gvn.transform(new CastX2PNode(src));
6765 offset = ConvI2X(offset);
6766
6767 // 'src_start' points to src array + scaled offset
6768 Node* src_start = basic_plus_adr(top(), base, offset);
6769
6770 // Call the stub.
6771 address stubAddr = StubRoutines::updateBytesAdler32();
6772 const char *stubName = "updateBytesAdler32";
6773
6774 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6775 stubAddr, stubName, TypePtr::BOTTOM,
6776 crc, src_start, length);
6777
6778 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6779 set_result(result);
6780 return true;
6781 }
6782
6783 //----------------------------inline_reference_get----------------------------
6784 // public T java.lang.ref.Reference.get();
6785 bool LibraryCallKit::inline_reference_get() {
6786 const int referent_offset = java_lang_ref_Reference::referent_offset();
6787
6788 // Get the argument:
6789 Node* reference_obj = null_check_receiver();
6790 if (stopped()) return true;
6791
6792 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6793 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6794 decorators, /*is_static*/ false, nullptr);
6795 if (result == nullptr) return false;
6796
6797 // Add memory barrier to prevent commoning reads from this field
6798 // across safepoint since GC can change its value.
6799 insert_mem_bar(Op_MemBarCPUOrder);
6800
6801 set_result(result);
6802 return true;
6803 }
6804
6805 //----------------------------inline_reference_refersTo0----------------------------
6806 // bool java.lang.ref.Reference.refersTo0();
6807 // bool java.lang.ref.PhantomReference.refersTo0();
6808 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6809 // Get arguments:
6810 Node* reference_obj = null_check_receiver();
6811 Node* other_obj = argument(1);
6812 if (stopped()) return true;
6813
6814 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6815 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6816 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6817 decorators, /*is_static*/ false, nullptr);
6818 if (referent == nullptr) return false;
6819
6820 // Add memory barrier to prevent commoning reads from this field
6821 // across safepoint since GC can change its value.
6822 insert_mem_bar(Op_MemBarCPUOrder);
6823
6824 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6825 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6826 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6827
6828 RegionNode* region = new RegionNode(3);
6829 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6830
6831 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6832 region->init_req(1, if_true);
6833 phi->init_req(1, intcon(1));
6834
6835 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6836 region->init_req(2, if_false);
6837 phi->init_req(2, intcon(0));
6838
6839 set_control(_gvn.transform(region));
6840 record_for_igvn(region);
6841 set_result(_gvn.transform(phi));
6842 return true;
6843 }
6844
6845
6846 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6847 DecoratorSet decorators, bool is_static,
6848 ciInstanceKlass* fromKls) {
6849 if (fromKls == nullptr) {
6850 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6851 assert(tinst != nullptr, "obj is null");
6852 assert(tinst->is_loaded(), "obj is not loaded");
6853 fromKls = tinst->instance_klass();
6854 } else {
6855 assert(is_static, "only for static field access");
6856 }
6857 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6858 ciSymbol::make(fieldTypeString),
6859 is_static);
6860
6861 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6862 if (field == nullptr) return (Node *) nullptr;
6863
6864 if (is_static) {
6865 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6866 fromObj = makecon(tip);
6867 }
6868
6869 // Next code copied from Parse::do_get_xxx():
6870
6871 // Compute address and memory type.
6872 int offset = field->offset_in_bytes();
6873 bool is_vol = field->is_volatile();
6874 ciType* field_klass = field->type();
6875 assert(field_klass->is_loaded(), "should be loaded");
6876 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6877 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6878 BasicType bt = field->layout_type();
6879
6880 // Build the resultant type of the load
6881 const Type *type;
6882 if (bt == T_OBJECT) {
6883 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6884 } else {
6885 type = Type::get_const_basic_type(bt);
6886 }
6887
6888 if (is_vol) {
6889 decorators |= MO_SEQ_CST;
6890 }
6891
6892 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6893 }
6894
6895 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6896 bool is_exact /* true */, bool is_static /* false */,
6897 ciInstanceKlass * fromKls /* nullptr */) {
6898 if (fromKls == nullptr) {
6899 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6900 assert(tinst != nullptr, "obj is null");
6901 assert(tinst->is_loaded(), "obj is not loaded");
6902 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6903 fromKls = tinst->instance_klass();
6904 }
6905 else {
6906 assert(is_static, "only for static field access");
6907 }
6908 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6909 ciSymbol::make(fieldTypeString),
6910 is_static);
6911
6912 assert(field != nullptr, "undefined field");
6913 assert(!field->is_volatile(), "not defined for volatile fields");
6914
6915 if (is_static) {
6916 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6917 fromObj = makecon(tip);
6918 }
6919
6920 // Next code copied from Parse::do_get_xxx():
6921
6922 // Compute address and memory type.
6923 int offset = field->offset_in_bytes();
6924 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6925
6926 return adr;
6927 }
6928
6929 //------------------------------inline_aescrypt_Block-----------------------
6930 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6931 address stubAddr = nullptr;
6932 const char *stubName;
6933 assert(UseAES, "need AES instruction support");
6934
6935 switch(id) {
6936 case vmIntrinsics::_aescrypt_encryptBlock:
6937 stubAddr = StubRoutines::aescrypt_encryptBlock();
6938 stubName = "aescrypt_encryptBlock";
6939 break;
6940 case vmIntrinsics::_aescrypt_decryptBlock:
6941 stubAddr = StubRoutines::aescrypt_decryptBlock();
6942 stubName = "aescrypt_decryptBlock";
6943 break;
6944 default:
6945 break;
6946 }
6947 if (stubAddr == nullptr) return false;
6948
6949 Node* aescrypt_object = argument(0);
6950 Node* src = argument(1);
6951 Node* src_offset = argument(2);
6952 Node* dest = argument(3);
6953 Node* dest_offset = argument(4);
6954
6955 src = must_be_not_null(src, true);
6956 dest = must_be_not_null(dest, true);
6957
6958 // (1) src and dest are arrays.
6959 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6960 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6961 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6962 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6963
6964 // for the quick and dirty code we will skip all the checks.
6965 // we are just trying to get the call to be generated.
6966 Node* src_start = src;
6967 Node* dest_start = dest;
6968 if (src_offset != nullptr || dest_offset != nullptr) {
6969 assert(src_offset != nullptr && dest_offset != nullptr, "");
6970 src_start = array_element_address(src, src_offset, T_BYTE);
6971 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6972 }
6973
6974 // now need to get the start of its expanded key array
6975 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6976 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6977 if (k_start == nullptr) return false;
6978
6979 // Call the stub.
6980 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6981 stubAddr, stubName, TypePtr::BOTTOM,
6982 src_start, dest_start, k_start);
6983
6984 return true;
6985 }
6986
6987 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6988 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6989 address stubAddr = nullptr;
6990 const char *stubName = nullptr;
6991
6992 assert(UseAES, "need AES instruction support");
6993
6994 switch(id) {
6995 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6996 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6997 stubName = "cipherBlockChaining_encryptAESCrypt";
6998 break;
6999 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7000 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7001 stubName = "cipherBlockChaining_decryptAESCrypt";
7002 break;
7003 default:
7004 break;
7005 }
7006 if (stubAddr == nullptr) return false;
7007
7008 Node* cipherBlockChaining_object = argument(0);
7009 Node* src = argument(1);
7010 Node* src_offset = argument(2);
7011 Node* len = argument(3);
7012 Node* dest = argument(4);
7013 Node* dest_offset = argument(5);
7014
7015 src = must_be_not_null(src, false);
7016 dest = must_be_not_null(dest, false);
7017
7018 // (1) src and dest are arrays.
7019 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7020 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7021 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7022 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7023
7024 // checks are the responsibility of the caller
7025 Node* src_start = src;
7026 Node* dest_start = dest;
7027 if (src_offset != nullptr || dest_offset != nullptr) {
7028 assert(src_offset != nullptr && dest_offset != nullptr, "");
7029 src_start = array_element_address(src, src_offset, T_BYTE);
7030 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7031 }
7032
7033 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7034 // (because of the predicated logic executed earlier).
7035 // so we cast it here safely.
7036 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7037
7038 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7039 if (embeddedCipherObj == nullptr) return false;
7040
7041 // cast it to what we know it will be at runtime
7042 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7043 assert(tinst != nullptr, "CBC obj is null");
7044 assert(tinst->is_loaded(), "CBC obj is not loaded");
7045 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7046 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7047
7048 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7049 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7050 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7051 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7052 aescrypt_object = _gvn.transform(aescrypt_object);
7053
7054 // we need to get the start of the aescrypt_object's expanded key array
7055 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7056 if (k_start == nullptr) return false;
7057
7058 // similarly, get the start address of the r vector
7059 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7060 if (objRvec == nullptr) return false;
7061 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7062
7063 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7064 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7065 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7066 stubAddr, stubName, TypePtr::BOTTOM,
7067 src_start, dest_start, k_start, r_start, len);
7068
7069 // return cipher length (int)
7070 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7071 set_result(retvalue);
7072 return true;
7073 }
7074
7075 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7076 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7077 address stubAddr = nullptr;
7078 const char *stubName = nullptr;
7079
7080 assert(UseAES, "need AES instruction support");
7081
7082 switch (id) {
7083 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7084 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7085 stubName = "electronicCodeBook_encryptAESCrypt";
7086 break;
7087 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7088 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7089 stubName = "electronicCodeBook_decryptAESCrypt";
7090 break;
7091 default:
7092 break;
7093 }
7094
7095 if (stubAddr == nullptr) return false;
7096
7097 Node* electronicCodeBook_object = argument(0);
7098 Node* src = argument(1);
7099 Node* src_offset = argument(2);
7100 Node* len = argument(3);
7101 Node* dest = argument(4);
7102 Node* dest_offset = argument(5);
7103
7104 // (1) src and dest are arrays.
7105 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7106 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7107 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7108 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7109
7110 // checks are the responsibility of the caller
7111 Node* src_start = src;
7112 Node* dest_start = dest;
7113 if (src_offset != nullptr || dest_offset != nullptr) {
7114 assert(src_offset != nullptr && dest_offset != nullptr, "");
7115 src_start = array_element_address(src, src_offset, T_BYTE);
7116 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7117 }
7118
7119 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7120 // (because of the predicated logic executed earlier).
7121 // so we cast it here safely.
7122 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7123
7124 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7125 if (embeddedCipherObj == nullptr) return false;
7126
7127 // cast it to what we know it will be at runtime
7128 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7129 assert(tinst != nullptr, "ECB obj is null");
7130 assert(tinst->is_loaded(), "ECB obj is not loaded");
7131 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7132 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7133
7134 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7135 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7136 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7137 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7138 aescrypt_object = _gvn.transform(aescrypt_object);
7139
7140 // we need to get the start of the aescrypt_object's expanded key array
7141 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7142 if (k_start == nullptr) return false;
7143
7144 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7145 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7146 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7147 stubAddr, stubName, TypePtr::BOTTOM,
7148 src_start, dest_start, k_start, len);
7149
7150 // return cipher length (int)
7151 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7152 set_result(retvalue);
7153 return true;
7154 }
7155
7156 //------------------------------inline_counterMode_AESCrypt-----------------------
7157 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7158 assert(UseAES, "need AES instruction support");
7159 if (!UseAESCTRIntrinsics) return false;
7160
7161 address stubAddr = nullptr;
7162 const char *stubName = nullptr;
7163 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7164 stubAddr = StubRoutines::counterMode_AESCrypt();
7165 stubName = "counterMode_AESCrypt";
7166 }
7167 if (stubAddr == nullptr) return false;
7168
7169 Node* counterMode_object = argument(0);
7170 Node* src = argument(1);
7171 Node* src_offset = argument(2);
7172 Node* len = argument(3);
7173 Node* dest = argument(4);
7174 Node* dest_offset = argument(5);
7175
7176 // (1) src and dest are arrays.
7177 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7178 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7179 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7180 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7181
7182 // checks are the responsibility of the caller
7183 Node* src_start = src;
7184 Node* dest_start = dest;
7185 if (src_offset != nullptr || dest_offset != nullptr) {
7186 assert(src_offset != nullptr && dest_offset != nullptr, "");
7187 src_start = array_element_address(src, src_offset, T_BYTE);
7188 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7189 }
7190
7191 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7192 // (because of the predicated logic executed earlier).
7193 // so we cast it here safely.
7194 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7195 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7196 if (embeddedCipherObj == nullptr) return false;
7197 // cast it to what we know it will be at runtime
7198 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7199 assert(tinst != nullptr, "CTR obj is null");
7200 assert(tinst->is_loaded(), "CTR obj is not loaded");
7201 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7202 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7203 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7204 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7205 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7206 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7207 aescrypt_object = _gvn.transform(aescrypt_object);
7208 // we need to get the start of the aescrypt_object's expanded key array
7209 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7210 if (k_start == nullptr) return false;
7211 // similarly, get the start address of the r vector
7212 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7213 if (obj_counter == nullptr) return false;
7214 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7215
7216 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7217 if (saved_encCounter == nullptr) return false;
7218 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7219 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7220
7221 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7222 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7223 OptoRuntime::counterMode_aescrypt_Type(),
7224 stubAddr, stubName, TypePtr::BOTTOM,
7225 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7226
7227 // return cipher length (int)
7228 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7229 set_result(retvalue);
7230 return true;
7231 }
7232
7233 //------------------------------get_key_start_from_aescrypt_object-----------------------
7234 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7235 #if defined(PPC64) || defined(S390)
7236 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7237 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7238 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7239 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7240 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7241 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7242 if (objSessionK == nullptr) {
7243 return (Node *) nullptr;
7244 }
7245 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7246 #else
7247 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7248 #endif // PPC64
7249 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7250 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7251
7252 // now have the array, need to get the start address of the K array
7253 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7254 return k_start;
7255 }
7256
7257 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7258 // Return node representing slow path of predicate check.
7259 // the pseudo code we want to emulate with this predicate is:
7260 // for encryption:
7261 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7262 // for decryption:
7263 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7264 // note cipher==plain is more conservative than the original java code but that's OK
7265 //
7266 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7267 // The receiver was checked for null already.
7268 Node* objCBC = argument(0);
7269
7270 Node* src = argument(1);
7271 Node* dest = argument(4);
7272
7273 // Load embeddedCipher field of CipherBlockChaining object.
7274 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7275
7276 // get AESCrypt klass for instanceOf check
7277 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7278 // will have same classloader as CipherBlockChaining object
7279 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7280 assert(tinst != nullptr, "CBCobj is null");
7281 assert(tinst->is_loaded(), "CBCobj is not loaded");
7282
7283 // we want to do an instanceof comparison against the AESCrypt class
7284 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7285 if (!klass_AESCrypt->is_loaded()) {
7286 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7287 Node* ctrl = control();
7288 set_control(top()); // no regular fast path
7289 return ctrl;
7290 }
7291
7292 src = must_be_not_null(src, true);
7293 dest = must_be_not_null(dest, true);
7294
7295 // Resolve oops to stable for CmpP below.
7296 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7297
7298 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7299 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7300 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7301
7302 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7303
7304 // for encryption, we are done
7305 if (!decrypting)
7306 return instof_false; // even if it is null
7307
7308 // for decryption, we need to add a further check to avoid
7309 // taking the intrinsic path when cipher and plain are the same
7310 // see the original java code for why.
7311 RegionNode* region = new RegionNode(3);
7312 region->init_req(1, instof_false);
7313
7314 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7315 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7316 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7317 region->init_req(2, src_dest_conjoint);
7318
7319 record_for_igvn(region);
7320 return _gvn.transform(region);
7321 }
7322
7323 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7324 // Return node representing slow path of predicate check.
7325 // the pseudo code we want to emulate with this predicate is:
7326 // for encryption:
7327 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7328 // for decryption:
7329 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7330 // note cipher==plain is more conservative than the original java code but that's OK
7331 //
7332 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7333 // The receiver was checked for null already.
7334 Node* objECB = argument(0);
7335
7336 // Load embeddedCipher field of ElectronicCodeBook object.
7337 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7338
7339 // get AESCrypt klass for instanceOf check
7340 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7341 // will have same classloader as ElectronicCodeBook object
7342 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7343 assert(tinst != nullptr, "ECBobj is null");
7344 assert(tinst->is_loaded(), "ECBobj is not loaded");
7345
7346 // we want to do an instanceof comparison against the AESCrypt class
7347 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7348 if (!klass_AESCrypt->is_loaded()) {
7349 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7350 Node* ctrl = control();
7351 set_control(top()); // no regular fast path
7352 return ctrl;
7353 }
7354 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7355
7356 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7357 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7358 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7359
7360 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7361
7362 // for encryption, we are done
7363 if (!decrypting)
7364 return instof_false; // even if it is null
7365
7366 // for decryption, we need to add a further check to avoid
7367 // taking the intrinsic path when cipher and plain are the same
7368 // see the original java code for why.
7369 RegionNode* region = new RegionNode(3);
7370 region->init_req(1, instof_false);
7371 Node* src = argument(1);
7372 Node* dest = argument(4);
7373 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7374 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7375 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7376 region->init_req(2, src_dest_conjoint);
7377
7378 record_for_igvn(region);
7379 return _gvn.transform(region);
7380 }
7381
7382 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7383 // Return node representing slow path of predicate check.
7384 // the pseudo code we want to emulate with this predicate is:
7385 // for encryption:
7386 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7387 // for decryption:
7388 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7389 // note cipher==plain is more conservative than the original java code but that's OK
7390 //
7391
7392 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7393 // The receiver was checked for null already.
7394 Node* objCTR = argument(0);
7395
7396 // Load embeddedCipher field of CipherBlockChaining object.
7397 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7398
7399 // get AESCrypt klass for instanceOf check
7400 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7401 // will have same classloader as CipherBlockChaining object
7402 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7403 assert(tinst != nullptr, "CTRobj is null");
7404 assert(tinst->is_loaded(), "CTRobj is not loaded");
7405
7406 // we want to do an instanceof comparison against the AESCrypt class
7407 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7408 if (!klass_AESCrypt->is_loaded()) {
7409 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7410 Node* ctrl = control();
7411 set_control(top()); // no regular fast path
7412 return ctrl;
7413 }
7414
7415 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7416 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7417 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7418 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7419 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7420
7421 return instof_false; // even if it is null
7422 }
7423
7424 //------------------------------inline_ghash_processBlocks
7425 bool LibraryCallKit::inline_ghash_processBlocks() {
7426 address stubAddr;
7427 const char *stubName;
7428 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7429
7430 stubAddr = StubRoutines::ghash_processBlocks();
7431 stubName = "ghash_processBlocks";
7432
7433 Node* data = argument(0);
7434 Node* offset = argument(1);
7435 Node* len = argument(2);
7436 Node* state = argument(3);
7437 Node* subkeyH = argument(4);
7438
7439 state = must_be_not_null(state, true);
7440 subkeyH = must_be_not_null(subkeyH, true);
7441 data = must_be_not_null(data, true);
7442
7443 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7444 assert(state_start, "state is null");
7445 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7446 assert(subkeyH_start, "subkeyH is null");
7447 Node* data_start = array_element_address(data, offset, T_BYTE);
7448 assert(data_start, "data is null");
7449
7450 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7451 OptoRuntime::ghash_processBlocks_Type(),
7452 stubAddr, stubName, TypePtr::BOTTOM,
7453 state_start, subkeyH_start, data_start, len);
7454 return true;
7455 }
7456
7457 //------------------------------inline_chacha20Block
7458 bool LibraryCallKit::inline_chacha20Block() {
7459 address stubAddr;
7460 const char *stubName;
7461 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7462
7463 stubAddr = StubRoutines::chacha20Block();
7464 stubName = "chacha20Block";
7465
7466 Node* state = argument(0);
7467 Node* result = argument(1);
7468
7469 state = must_be_not_null(state, true);
7470 result = must_be_not_null(result, true);
7471
7472 Node* state_start = array_element_address(state, intcon(0), T_INT);
7473 assert(state_start, "state is null");
7474 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7475 assert(result_start, "result is null");
7476
7477 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7478 OptoRuntime::chacha20Block_Type(),
7479 stubAddr, stubName, TypePtr::BOTTOM,
7480 state_start, result_start);
7481 // return key stream length (int)
7482 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7483 set_result(retvalue);
7484 return true;
7485 }
7486
7487 bool LibraryCallKit::inline_base64_encodeBlock() {
7488 address stubAddr;
7489 const char *stubName;
7490 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7491 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7492 stubAddr = StubRoutines::base64_encodeBlock();
7493 stubName = "encodeBlock";
7494
7495 if (!stubAddr) return false;
7496 Node* base64obj = argument(0);
7497 Node* src = argument(1);
7498 Node* offset = argument(2);
7499 Node* len = argument(3);
7500 Node* dest = argument(4);
7501 Node* dp = argument(5);
7502 Node* isURL = argument(6);
7503
7504 src = must_be_not_null(src, true);
7505 dest = must_be_not_null(dest, true);
7506
7507 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7508 assert(src_start, "source array is null");
7509 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7510 assert(dest_start, "destination array is null");
7511
7512 Node* base64 = make_runtime_call(RC_LEAF,
7513 OptoRuntime::base64_encodeBlock_Type(),
7514 stubAddr, stubName, TypePtr::BOTTOM,
7515 src_start, offset, len, dest_start, dp, isURL);
7516 return true;
7517 }
7518
7519 bool LibraryCallKit::inline_base64_decodeBlock() {
7520 address stubAddr;
7521 const char *stubName;
7522 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7523 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7524 stubAddr = StubRoutines::base64_decodeBlock();
7525 stubName = "decodeBlock";
7526
7527 if (!stubAddr) return false;
7528 Node* base64obj = argument(0);
7529 Node* src = argument(1);
7530 Node* src_offset = argument(2);
7531 Node* len = argument(3);
7532 Node* dest = argument(4);
7533 Node* dest_offset = argument(5);
7534 Node* isURL = argument(6);
7535 Node* isMIME = argument(7);
7536
7537 src = must_be_not_null(src, true);
7538 dest = must_be_not_null(dest, true);
7539
7540 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7541 assert(src_start, "source array is null");
7542 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7543 assert(dest_start, "destination array is null");
7544
7545 Node* call = make_runtime_call(RC_LEAF,
7546 OptoRuntime::base64_decodeBlock_Type(),
7547 stubAddr, stubName, TypePtr::BOTTOM,
7548 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7549 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7550 set_result(result);
7551 return true;
7552 }
7553
7554 bool LibraryCallKit::inline_poly1305_processBlocks() {
7555 address stubAddr;
7556 const char *stubName;
7557 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7558 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7559 stubAddr = StubRoutines::poly1305_processBlocks();
7560 stubName = "poly1305_processBlocks";
7561
7562 if (!stubAddr) return false;
7563 null_check_receiver(); // null-check receiver
7564 if (stopped()) return true;
7565
7566 Node* input = argument(1);
7567 Node* input_offset = argument(2);
7568 Node* len = argument(3);
7569 Node* alimbs = argument(4);
7570 Node* rlimbs = argument(5);
7571
7572 input = must_be_not_null(input, true);
7573 alimbs = must_be_not_null(alimbs, true);
7574 rlimbs = must_be_not_null(rlimbs, true);
7575
7576 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7577 assert(input_start, "input array is null");
7578 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7579 assert(acc_start, "acc array is null");
7580 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7581 assert(r_start, "r array is null");
7582
7583 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7584 OptoRuntime::poly1305_processBlocks_Type(),
7585 stubAddr, stubName, TypePtr::BOTTOM,
7586 input_start, len, acc_start, r_start);
7587 return true;
7588 }
7589
7590 //------------------------------inline_digestBase_implCompress-----------------------
7591 //
7592 // Calculate MD5 for single-block byte[] array.
7593 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7594 //
7595 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7596 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7597 //
7598 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7599 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7600 //
7601 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7602 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7603 //
7604 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7605 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7606 //
7607 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7608 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7609
7610 Node* digestBase_obj = argument(0);
7611 Node* src = argument(1); // type oop
7612 Node* ofs = argument(2); // type int
7613
7614 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7615 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7616 // failed array check
7617 return false;
7618 }
7619 // Figure out the size and type of the elements we will be copying.
7620 BasicType src_elem = src_type->elem()->array_element_basic_type();
7621 if (src_elem != T_BYTE) {
7622 return false;
7623 }
7624 // 'src_start' points to src array + offset
7625 src = must_be_not_null(src, true);
7626 Node* src_start = array_element_address(src, ofs, src_elem);
7627 Node* state = nullptr;
7628 Node* block_size = nullptr;
7629 address stubAddr;
7630 const char *stubName;
7631
7632 switch(id) {
7633 case vmIntrinsics::_md5_implCompress:
7634 assert(UseMD5Intrinsics, "need MD5 instruction support");
7635 state = get_state_from_digest_object(digestBase_obj, T_INT);
7636 stubAddr = StubRoutines::md5_implCompress();
7637 stubName = "md5_implCompress";
7638 break;
7639 case vmIntrinsics::_sha_implCompress:
7640 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7641 state = get_state_from_digest_object(digestBase_obj, T_INT);
7642 stubAddr = StubRoutines::sha1_implCompress();
7643 stubName = "sha1_implCompress";
7644 break;
7645 case vmIntrinsics::_sha2_implCompress:
7646 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7647 state = get_state_from_digest_object(digestBase_obj, T_INT);
7648 stubAddr = StubRoutines::sha256_implCompress();
7649 stubName = "sha256_implCompress";
7650 break;
7651 case vmIntrinsics::_sha5_implCompress:
7652 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7653 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7654 stubAddr = StubRoutines::sha512_implCompress();
7655 stubName = "sha512_implCompress";
7656 break;
7657 case vmIntrinsics::_sha3_implCompress:
7658 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7659 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7660 stubAddr = StubRoutines::sha3_implCompress();
7661 stubName = "sha3_implCompress";
7662 block_size = get_block_size_from_digest_object(digestBase_obj);
7663 if (block_size == nullptr) return false;
7664 break;
7665 default:
7666 fatal_unexpected_iid(id);
7667 return false;
7668 }
7669 if (state == nullptr) return false;
7670
7671 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7672 if (stubAddr == nullptr) return false;
7673
7674 // Call the stub.
7675 Node* call;
7676 if (block_size == nullptr) {
7677 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7678 stubAddr, stubName, TypePtr::BOTTOM,
7679 src_start, state);
7680 } else {
7681 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7682 stubAddr, stubName, TypePtr::BOTTOM,
7683 src_start, state, block_size);
7684 }
7685
7686 return true;
7687 }
7688
7689 //------------------------------inline_digestBase_implCompressMB-----------------------
7690 //
7691 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7692 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7693 //
7694 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7695 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7696 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7697 assert((uint)predicate < 5, "sanity");
7698 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7699
7700 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7701 Node* src = argument(1); // byte[] array
7702 Node* ofs = argument(2); // type int
7703 Node* limit = argument(3); // type int
7704
7705 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7706 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7707 // failed array check
7708 return false;
7709 }
7710 // Figure out the size and type of the elements we will be copying.
7711 BasicType src_elem = src_type->elem()->array_element_basic_type();
7712 if (src_elem != T_BYTE) {
7713 return false;
7714 }
7715 // 'src_start' points to src array + offset
7716 src = must_be_not_null(src, false);
7717 Node* src_start = array_element_address(src, ofs, src_elem);
7718
7719 const char* klass_digestBase_name = nullptr;
7720 const char* stub_name = nullptr;
7721 address stub_addr = nullptr;
7722 BasicType elem_type = T_INT;
7723
7724 switch (predicate) {
7725 case 0:
7726 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7727 klass_digestBase_name = "sun/security/provider/MD5";
7728 stub_name = "md5_implCompressMB";
7729 stub_addr = StubRoutines::md5_implCompressMB();
7730 }
7731 break;
7732 case 1:
7733 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7734 klass_digestBase_name = "sun/security/provider/SHA";
7735 stub_name = "sha1_implCompressMB";
7736 stub_addr = StubRoutines::sha1_implCompressMB();
7737 }
7738 break;
7739 case 2:
7740 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7741 klass_digestBase_name = "sun/security/provider/SHA2";
7742 stub_name = "sha256_implCompressMB";
7743 stub_addr = StubRoutines::sha256_implCompressMB();
7744 }
7745 break;
7746 case 3:
7747 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7748 klass_digestBase_name = "sun/security/provider/SHA5";
7749 stub_name = "sha512_implCompressMB";
7750 stub_addr = StubRoutines::sha512_implCompressMB();
7751 elem_type = T_LONG;
7752 }
7753 break;
7754 case 4:
7755 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7756 klass_digestBase_name = "sun/security/provider/SHA3";
7757 stub_name = "sha3_implCompressMB";
7758 stub_addr = StubRoutines::sha3_implCompressMB();
7759 elem_type = T_BYTE;
7760 }
7761 break;
7762 default:
7763 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7764 }
7765 if (klass_digestBase_name != nullptr) {
7766 assert(stub_addr != nullptr, "Stub is generated");
7767 if (stub_addr == nullptr) return false;
7768
7769 // get DigestBase klass to lookup for SHA klass
7770 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7771 assert(tinst != nullptr, "digestBase_obj is not instance???");
7772 assert(tinst->is_loaded(), "DigestBase is not loaded");
7773
7774 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7775 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7776 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7777 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7778 }
7779 return false;
7780 }
7781
7782 //------------------------------inline_digestBase_implCompressMB-----------------------
7783 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7784 BasicType elem_type, address stubAddr, const char *stubName,
7785 Node* src_start, Node* ofs, Node* limit) {
7786 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7787 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7788 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7789 digest_obj = _gvn.transform(digest_obj);
7790
7791 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7792 if (state == nullptr) return false;
7793
7794 Node* block_size = nullptr;
7795 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7796 block_size = get_block_size_from_digest_object(digest_obj);
7797 if (block_size == nullptr) return false;
7798 }
7799
7800 // Call the stub.
7801 Node* call;
7802 if (block_size == nullptr) {
7803 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7804 OptoRuntime::digestBase_implCompressMB_Type(false),
7805 stubAddr, stubName, TypePtr::BOTTOM,
7806 src_start, state, ofs, limit);
7807 } else {
7808 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7809 OptoRuntime::digestBase_implCompressMB_Type(true),
7810 stubAddr, stubName, TypePtr::BOTTOM,
7811 src_start, state, block_size, ofs, limit);
7812 }
7813
7814 // return ofs (int)
7815 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7816 set_result(result);
7817
7818 return true;
7819 }
7820
7821 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7822 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7823 assert(UseAES, "need AES instruction support");
7824 address stubAddr = nullptr;
7825 const char *stubName = nullptr;
7826 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7827 stubName = "galoisCounterMode_AESCrypt";
7828
7829 if (stubAddr == nullptr) return false;
7830
7831 Node* in = argument(0);
7832 Node* inOfs = argument(1);
7833 Node* len = argument(2);
7834 Node* ct = argument(3);
7835 Node* ctOfs = argument(4);
7836 Node* out = argument(5);
7837 Node* outOfs = argument(6);
7838 Node* gctr_object = argument(7);
7839 Node* ghash_object = argument(8);
7840
7841 // (1) in, ct and out are arrays.
7842 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7843 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7844 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7845 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7846 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7847 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7848
7849 // checks are the responsibility of the caller
7850 Node* in_start = in;
7851 Node* ct_start = ct;
7852 Node* out_start = out;
7853 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7854 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7855 in_start = array_element_address(in, inOfs, T_BYTE);
7856 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7857 out_start = array_element_address(out, outOfs, T_BYTE);
7858 }
7859
7860 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7861 // (because of the predicated logic executed earlier).
7862 // so we cast it here safely.
7863 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7864 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7865 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7866 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7867 Node* state = load_field_from_object(ghash_object, "state", "[J");
7868
7869 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7870 return false;
7871 }
7872 // cast it to what we know it will be at runtime
7873 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7874 assert(tinst != nullptr, "GCTR obj is null");
7875 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7876 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7877 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7878 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7879 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7880 const TypeOopPtr* xtype = aklass->as_instance_type();
7881 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7882 aescrypt_object = _gvn.transform(aescrypt_object);
7883 // we need to get the start of the aescrypt_object's expanded key array
7884 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7885 if (k_start == nullptr) return false;
7886 // similarly, get the start address of the r vector
7887 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7888 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7889 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7890
7891
7892 // Call the stub, passing params
7893 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7894 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7895 stubAddr, stubName, TypePtr::BOTTOM,
7896 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7897
7898 // return cipher length (int)
7899 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7900 set_result(retvalue);
7901
7902 return true;
7903 }
7904
7905 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7906 // Return node representing slow path of predicate check.
7907 // the pseudo code we want to emulate with this predicate is:
7908 // for encryption:
7909 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7910 // for decryption:
7911 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7912 // note cipher==plain is more conservative than the original java code but that's OK
7913 //
7914
7915 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7916 // The receiver was checked for null already.
7917 Node* objGCTR = argument(7);
7918 // Load embeddedCipher field of GCTR object.
7919 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7920 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7921
7922 // get AESCrypt klass for instanceOf check
7923 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7924 // will have same classloader as CipherBlockChaining object
7925 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7926 assert(tinst != nullptr, "GCTR obj is null");
7927 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7928
7929 // we want to do an instanceof comparison against the AESCrypt class
7930 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7931 if (!klass_AESCrypt->is_loaded()) {
7932 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7933 Node* ctrl = control();
7934 set_control(top()); // no regular fast path
7935 return ctrl;
7936 }
7937
7938 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7939 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7940 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7941 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7942 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7943
7944 return instof_false; // even if it is null
7945 }
7946
7947 //------------------------------get_state_from_digest_object-----------------------
7948 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7949 const char* state_type;
7950 switch (elem_type) {
7951 case T_BYTE: state_type = "[B"; break;
7952 case T_INT: state_type = "[I"; break;
7953 case T_LONG: state_type = "[J"; break;
7954 default: ShouldNotReachHere();
7955 }
7956 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7957 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7958 if (digest_state == nullptr) return (Node *) nullptr;
7959
7960 // now have the array, need to get the start address of the state array
7961 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7962 return state;
7963 }
7964
7965 //------------------------------get_block_size_from_sha3_object----------------------------------
7966 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7967 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7968 assert (block_size != nullptr, "sanity");
7969 return block_size;
7970 }
7971
7972 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7973 // Return node representing slow path of predicate check.
7974 // the pseudo code we want to emulate with this predicate is:
7975 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7976 //
7977 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7978 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7979 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7980 assert((uint)predicate < 5, "sanity");
7981
7982 // The receiver was checked for null already.
7983 Node* digestBaseObj = argument(0);
7984
7985 // get DigestBase klass for instanceOf check
7986 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7987 assert(tinst != nullptr, "digestBaseObj is null");
7988 assert(tinst->is_loaded(), "DigestBase is not loaded");
7989
7990 const char* klass_name = nullptr;
7991 switch (predicate) {
7992 case 0:
7993 if (UseMD5Intrinsics) {
7994 // we want to do an instanceof comparison against the MD5 class
7995 klass_name = "sun/security/provider/MD5";
7996 }
7997 break;
7998 case 1:
7999 if (UseSHA1Intrinsics) {
8000 // we want to do an instanceof comparison against the SHA class
8001 klass_name = "sun/security/provider/SHA";
8002 }
8003 break;
8004 case 2:
8005 if (UseSHA256Intrinsics) {
8006 // we want to do an instanceof comparison against the SHA2 class
8007 klass_name = "sun/security/provider/SHA2";
8008 }
8009 break;
8010 case 3:
8011 if (UseSHA512Intrinsics) {
8012 // we want to do an instanceof comparison against the SHA5 class
8013 klass_name = "sun/security/provider/SHA5";
8014 }
8015 break;
8016 case 4:
8017 if (UseSHA3Intrinsics) {
8018 // we want to do an instanceof comparison against the SHA3 class
8019 klass_name = "sun/security/provider/SHA3";
8020 }
8021 break;
8022 default:
8023 fatal("unknown SHA intrinsic predicate: %d", predicate);
8024 }
8025
8026 ciKlass* klass = nullptr;
8027 if (klass_name != nullptr) {
8028 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8029 }
8030 if ((klass == nullptr) || !klass->is_loaded()) {
8031 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8032 Node* ctrl = control();
8033 set_control(top()); // no intrinsic path
8034 return ctrl;
8035 }
8036 ciInstanceKlass* instklass = klass->as_instance_klass();
8037
8038 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8039 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8040 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8041 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8042
8043 return instof_false; // even if it is null
8044 }
8045
8046 //-------------inline_fma-----------------------------------
8047 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8048 Node *a = nullptr;
8049 Node *b = nullptr;
8050 Node *c = nullptr;
8051 Node* result = nullptr;
8052 switch (id) {
8053 case vmIntrinsics::_fmaD:
8054 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8055 // no receiver since it is static method
8056 a = round_double_node(argument(0));
8057 b = round_double_node(argument(2));
8058 c = round_double_node(argument(4));
8059 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8060 break;
8061 case vmIntrinsics::_fmaF:
8062 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8063 a = argument(0);
8064 b = argument(1);
8065 c = argument(2);
8066 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8067 break;
8068 default:
8069 fatal_unexpected_iid(id); break;
8070 }
8071 set_result(result);
8072 return true;
8073 }
8074
8075 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8076 // argument(0) is receiver
8077 Node* codePoint = argument(1);
8078 Node* n = nullptr;
8079
8080 switch (id) {
8081 case vmIntrinsics::_isDigit :
8082 n = new DigitNode(control(), codePoint);
8083 break;
8084 case vmIntrinsics::_isLowerCase :
8085 n = new LowerCaseNode(control(), codePoint);
8086 break;
8087 case vmIntrinsics::_isUpperCase :
8088 n = new UpperCaseNode(control(), codePoint);
8089 break;
8090 case vmIntrinsics::_isWhitespace :
8091 n = new WhitespaceNode(control(), codePoint);
8092 break;
8093 default:
8094 fatal_unexpected_iid(id);
8095 }
8096
8097 set_result(_gvn.transform(n));
8098 return true;
8099 }
8100
8101 //------------------------------inline_fp_min_max------------------------------
8102 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8103 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8104
8105 // The intrinsic should be used only when the API branches aren't predictable,
8106 // the last one performing the most important comparison. The following heuristic
8107 // uses the branch statistics to eventually bail out if necessary.
8108
8109 ciMethodData *md = callee()->method_data();
8110
8111 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8112 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8113
8114 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8115 // Bail out if the call-site didn't contribute enough to the statistics.
8116 return false;
8117 }
8118
8119 uint taken = 0, not_taken = 0;
8120
8121 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8122 if (p->is_BranchData()) {
8123 taken = ((ciBranchData*)p)->taken();
8124 not_taken = ((ciBranchData*)p)->not_taken();
8125 }
8126 }
8127
8128 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8129 balance = balance < 0 ? -balance : balance;
8130 if ( balance > 0.2 ) {
8131 // Bail out if the most important branch is predictable enough.
8132 return false;
8133 }
8134 }
8135 */
8136
8137 Node *a = nullptr;
8138 Node *b = nullptr;
8139 Node *n = nullptr;
8140 switch (id) {
8141 case vmIntrinsics::_maxF:
8142 case vmIntrinsics::_minF:
8143 case vmIntrinsics::_maxF_strict:
8144 case vmIntrinsics::_minF_strict:
8145 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8146 a = argument(0);
8147 b = argument(1);
8148 break;
8149 case vmIntrinsics::_maxD:
8150 case vmIntrinsics::_minD:
8151 case vmIntrinsics::_maxD_strict:
8152 case vmIntrinsics::_minD_strict:
8153 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8154 a = round_double_node(argument(0));
8155 b = round_double_node(argument(2));
8156 break;
8157 default:
8158 fatal_unexpected_iid(id);
8159 break;
8160 }
8161 switch (id) {
8162 case vmIntrinsics::_maxF:
8163 case vmIntrinsics::_maxF_strict:
8164 n = new MaxFNode(a, b);
8165 break;
8166 case vmIntrinsics::_minF:
8167 case vmIntrinsics::_minF_strict:
8168 n = new MinFNode(a, b);
8169 break;
8170 case vmIntrinsics::_maxD:
8171 case vmIntrinsics::_maxD_strict:
8172 n = new MaxDNode(a, b);
8173 break;
8174 case vmIntrinsics::_minD:
8175 case vmIntrinsics::_minD_strict:
8176 n = new MinDNode(a, b);
8177 break;
8178 default:
8179 fatal_unexpected_iid(id);
8180 break;
8181 }
8182 set_result(_gvn.transform(n));
8183 return true;
8184 }
8185
8186 bool LibraryCallKit::inline_profileBoolean() {
8187 Node* counts = argument(1);
8188 const TypeAryPtr* ary = nullptr;
8189 ciArray* aobj = nullptr;
8190 if (counts->is_Con()
8191 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8192 && (aobj = ary->const_oop()->as_array()) != nullptr
8193 && (aobj->length() == 2)) {
8194 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8195 jint false_cnt = aobj->element_value(0).as_int();
8196 jint true_cnt = aobj->element_value(1).as_int();
8197
8198 if (C->log() != nullptr) {
8199 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8200 false_cnt, true_cnt);
8201 }
8202
8203 if (false_cnt + true_cnt == 0) {
8204 // According to profile, never executed.
8205 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8206 Deoptimization::Action_reinterpret);
8207 return true;
8208 }
8209
8210 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8211 // is a number of each value occurrences.
8212 Node* result = argument(0);
8213 if (false_cnt == 0 || true_cnt == 0) {
8214 // According to profile, one value has been never seen.
8215 int expected_val = (false_cnt == 0) ? 1 : 0;
8216
8217 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8218 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8219
8220 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8221 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8222 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8223
8224 { // Slow path: uncommon trap for never seen value and then reexecute
8225 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8226 // the value has been seen at least once.
8227 PreserveJVMState pjvms(this);
8228 PreserveReexecuteState preexecs(this);
8229 jvms()->set_should_reexecute(true);
8230
8231 set_control(slow_path);
8232 set_i_o(i_o());
8233
8234 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8235 Deoptimization::Action_reinterpret);
8236 }
8237 // The guard for never seen value enables sharpening of the result and
8238 // returning a constant. It allows to eliminate branches on the same value
8239 // later on.
8240 set_control(fast_path);
8241 result = intcon(expected_val);
8242 }
8243 // Stop profiling.
8244 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8245 // By replacing method body with profile data (represented as ProfileBooleanNode
8246 // on IR level) we effectively disable profiling.
8247 // It enables full speed execution once optimized code is generated.
8248 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8249 C->record_for_igvn(profile);
8250 set_result(profile);
8251 return true;
8252 } else {
8253 // Continue profiling.
8254 // Profile data isn't available at the moment. So, execute method's bytecode version.
8255 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8256 // is compiled and counters aren't available since corresponding MethodHandle
8257 // isn't a compile-time constant.
8258 return false;
8259 }
8260 }
8261
8262 bool LibraryCallKit::inline_isCompileConstant() {
8263 Node* n = argument(0);
8264 set_result(n->is_Con() ? intcon(1) : intcon(0));
8265 return true;
8266 }
8267
8268 //------------------------------- inline_getObjectSize --------------------------------------
8269 //
8270 // Calculate the runtime size of the object/array.
8271 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8272 //
8273 bool LibraryCallKit::inline_getObjectSize() {
8274 Node* obj = argument(3);
8275 Node* klass_node = load_object_klass(obj);
8276
8277 jint layout_con = Klass::_lh_neutral_value;
8278 Node* layout_val = get_layout_helper(klass_node, layout_con);
8279 int layout_is_con = (layout_val == nullptr);
8280
8281 if (layout_is_con) {
8282 // Layout helper is constant, can figure out things at compile time.
8283
8284 if (Klass::layout_helper_is_instance(layout_con)) {
8285 // Instance case: layout_con contains the size itself.
8286 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8287 set_result(size);
8288 } else {
8289 // Array case: size is round(header + element_size*arraylength).
8290 // Since arraylength is different for every array instance, we have to
8291 // compute the whole thing at runtime.
8292
8293 Node* arr_length = load_array_length(obj);
8294
8295 int round_mask = MinObjAlignmentInBytes - 1;
8296 int hsize = Klass::layout_helper_header_size(layout_con);
8297 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8298
8299 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8300 round_mask = 0; // strength-reduce it if it goes away completely
8301 }
8302 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8303 Node* header_size = intcon(hsize + round_mask);
8304
8305 Node* lengthx = ConvI2X(arr_length);
8306 Node* headerx = ConvI2X(header_size);
8307
8308 Node* abody = lengthx;
8309 if (eshift != 0) {
8310 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8311 }
8312 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8313 if (round_mask != 0) {
8314 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8315 }
8316 size = ConvX2L(size);
8317 set_result(size);
8318 }
8319 } else {
8320 // Layout helper is not constant, need to test for array-ness at runtime.
8321
8322 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8323 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8324 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8325 record_for_igvn(result_reg);
8326
8327 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8328 if (array_ctl != nullptr) {
8329 // Array case: size is round(header + element_size*arraylength).
8330 // Since arraylength is different for every array instance, we have to
8331 // compute the whole thing at runtime.
8332
8333 PreserveJVMState pjvms(this);
8334 set_control(array_ctl);
8335 Node* arr_length = load_array_length(obj);
8336
8337 int round_mask = MinObjAlignmentInBytes - 1;
8338 Node* mask = intcon(round_mask);
8339
8340 Node* hss = intcon(Klass::_lh_header_size_shift);
8341 Node* hsm = intcon(Klass::_lh_header_size_mask);
8342 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8343 header_size = _gvn.transform(new AndINode(header_size, hsm));
8344 header_size = _gvn.transform(new AddINode(header_size, mask));
8345
8346 // There is no need to mask or shift this value.
8347 // The semantics of LShiftINode include an implicit mask to 0x1F.
8348 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8349 Node* elem_shift = layout_val;
8350
8351 Node* lengthx = ConvI2X(arr_length);
8352 Node* headerx = ConvI2X(header_size);
8353
8354 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8355 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8356 if (round_mask != 0) {
8357 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8358 }
8359 size = ConvX2L(size);
8360
8361 result_reg->init_req(_array_path, control());
8362 result_val->init_req(_array_path, size);
8363 }
8364
8365 if (!stopped()) {
8366 // Instance case: the layout helper gives us instance size almost directly,
8367 // but we need to mask out the _lh_instance_slow_path_bit.
8368 Node* size = ConvI2X(layout_val);
8369 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8370 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8371 size = _gvn.transform(new AndXNode(size, mask));
8372 size = ConvX2L(size);
8373
8374 result_reg->init_req(_instance_path, control());
8375 result_val->init_req(_instance_path, size);
8376 }
8377
8378 set_result(result_reg, result_val);
8379 }
8380
8381 return true;
8382 }
8383
8384 //------------------------------- inline_blackhole --------------------------------------
8385 //
8386 // Make sure all arguments to this node are alive.
8387 // This matches methods that were requested to be blackholed through compile commands.
8388 //
8389 bool LibraryCallKit::inline_blackhole() {
8390 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8391 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8392 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8393
8394 // Blackhole node pinches only the control, not memory. This allows
8395 // the blackhole to be pinned in the loop that computes blackholed
8396 // values, but have no other side effects, like breaking the optimizations
8397 // across the blackhole.
8398
8399 Node* bh = _gvn.transform(new BlackholeNode(control()));
8400 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8401
8402 // Bind call arguments as blackhole arguments to keep them alive
8403 uint nargs = callee()->arg_size();
8404 for (uint i = 0; i < nargs; i++) {
8405 bh->add_req(argument(i));
8406 }
8407
8408 return true;
8409 }