1 /*
2 * Copyright (c) 1999, 2024, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "prims/jvmtiExport.hpp"
54 #include "prims/jvmtiThreadState.hpp"
55 #include "prims/unsafe.hpp"
56 #include "runtime/jniHandles.inline.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/sharedRuntime.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "utilities/macros.hpp"
61 #include "utilities/powerOfTwo.hpp"
62
63 //---------------------------make_vm_intrinsic----------------------------
64 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
65 vmIntrinsicID id = m->intrinsic_id();
66 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
67
68 if (!m->is_loaded()) {
69 // Do not attempt to inline unloaded methods.
70 return nullptr;
71 }
72
73 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
74 bool is_available = false;
75
76 {
77 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
78 // the compiler must transition to '_thread_in_vm' state because both
79 // methods access VM-internal data.
80 VM_ENTRY_MARK;
81 methodHandle mh(THREAD, m->get_Method());
82 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
83 if (is_available && is_virtual) {
84 is_available = vmIntrinsics::does_virtual_dispatch(id);
85 }
86 }
87
88 if (is_available) {
89 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
90 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
91 return new LibraryIntrinsic(m, is_virtual,
92 vmIntrinsics::predicates_needed(id),
93 vmIntrinsics::does_virtual_dispatch(id),
94 id);
95 } else {
96 return nullptr;
97 }
98 }
99
100 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
101 LibraryCallKit kit(jvms, this);
102 Compile* C = kit.C;
103 int nodes = C->unique();
104 #ifndef PRODUCT
105 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
106 char buf[1000];
107 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
108 tty->print_cr("Intrinsic %s", str);
109 }
110 #endif
111 ciMethod* callee = kit.callee();
112 const int bci = kit.bci();
113 #ifdef ASSERT
114 Node* ctrl = kit.control();
115 #endif
116 // Try to inline the intrinsic.
117 if (callee->check_intrinsic_candidate() &&
118 kit.try_to_inline(_last_predicate)) {
119 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
120 : "(intrinsic)";
121 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
122 if (C->print_intrinsics() || C->print_inlining()) {
123 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 }
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 C->print_inlining_update(this);
135 return kit.transfer_exceptions_into_jvms();
136 }
137
138 // The intrinsic bailed out
139 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 if (C->print_intrinsics() || C->print_inlining()) {
151 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
152 }
153 } else {
154 // Root compile
155 ResourceMark rm;
156 stringStream msg_stream;
157 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
158 vmIntrinsics::name_at(intrinsic_id()),
159 is_virtual() ? " (virtual)" : "", bci);
160 const char *msg = msg_stream.freeze();
161 log_debug(jit, inlining)("%s", msg);
162 if (C->print_intrinsics() || C->print_inlining()) {
163 tty->print("%s", msg);
164 }
165 }
166 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
167 C->print_inlining_update(this);
168
169 return nullptr;
170 }
171
172 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
173 LibraryCallKit kit(jvms, this);
174 Compile* C = kit.C;
175 int nodes = C->unique();
176 _last_predicate = predicate;
177 #ifndef PRODUCT
178 assert(is_predicated() && predicate < predicates_count(), "sanity");
179 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
180 char buf[1000];
181 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
182 tty->print_cr("Predicate for intrinsic %s", str);
183 }
184 #endif
185 ciMethod* callee = kit.callee();
186 const int bci = kit.bci();
187
188 Node* slow_ctl = kit.try_to_predicate(predicate);
189 if (!kit.failing()) {
190 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
191 : "(intrinsic, predicate)";
192 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
193 if (C->print_intrinsics() || C->print_inlining()) {
194 C->print_inlining(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
195 }
196 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
197 if (C->log()) {
198 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
199 vmIntrinsics::name_at(intrinsic_id()),
200 (is_virtual() ? " virtual='1'" : ""),
201 C->unique() - nodes);
202 }
203 return slow_ctl; // Could be null if the check folds.
204 }
205
206 // The intrinsic bailed out
207 if (jvms->has_method()) {
208 // Not a root compile.
209 const char* msg = "failed to generate predicate for intrinsic";
210 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
211 if (C->print_intrinsics() || C->print_inlining()) {
212 C->print_inlining(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
213 }
214 } else {
215 // Root compile
216 ResourceMark rm;
217 stringStream msg_stream;
218 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
219 vmIntrinsics::name_at(intrinsic_id()),
220 is_virtual() ? " (virtual)" : "", bci);
221 const char *msg = msg_stream.freeze();
222 log_debug(jit, inlining)("%s", msg);
223 if (C->print_intrinsics() || C->print_inlining()) {
224 C->print_inlining_stream()->print("%s", msg);
225 }
226 }
227 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
228 return nullptr;
229 }
230
231 bool LibraryCallKit::try_to_inline(int predicate) {
232 // Handle symbolic names for otherwise undistinguished boolean switches:
233 const bool is_store = true;
234 const bool is_compress = true;
235 const bool is_static = true;
236 const bool is_volatile = true;
237
238 if (!jvms()->has_method()) {
239 // Root JVMState has a null method.
240 assert(map()->memory()->Opcode() == Op_Parm, "");
241 // Insert the memory aliasing node
242 set_all_memory(reset_memory());
243 }
244 assert(merged_memory(), "");
245
246 switch (intrinsic_id()) {
247 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
248 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
249 case vmIntrinsics::_getClass: return inline_native_getClass();
250
251 case vmIntrinsics::_ceil:
252 case vmIntrinsics::_floor:
253 case vmIntrinsics::_rint:
254 case vmIntrinsics::_dsin:
255 case vmIntrinsics::_dcos:
256 case vmIntrinsics::_dtan:
257 case vmIntrinsics::_dabs:
258 case vmIntrinsics::_fabs:
259 case vmIntrinsics::_iabs:
260 case vmIntrinsics::_labs:
261 case vmIntrinsics::_datan2:
262 case vmIntrinsics::_dsqrt:
263 case vmIntrinsics::_dsqrt_strict:
264 case vmIntrinsics::_dexp:
265 case vmIntrinsics::_dlog:
266 case vmIntrinsics::_dlog10:
267 case vmIntrinsics::_dpow:
268 case vmIntrinsics::_dcopySign:
269 case vmIntrinsics::_fcopySign:
270 case vmIntrinsics::_dsignum:
271 case vmIntrinsics::_roundF:
272 case vmIntrinsics::_roundD:
273 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
274
275 case vmIntrinsics::_notify:
276 case vmIntrinsics::_notifyAll:
277 return inline_notify(intrinsic_id());
278
279 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
280 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
281 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
282 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
283 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
284 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
285 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
286 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
287 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
288 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
289 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
290 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
291 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
292 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
293
294 case vmIntrinsics::_arraycopy: return inline_arraycopy();
295
296 case vmIntrinsics::_arraySort: return inline_array_sort();
297 case vmIntrinsics::_arrayPartition: return inline_array_partition();
298
299 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
300 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
301 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
302 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
303
304 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
305 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
306 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
307 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
308 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
309 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
310 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
311 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
312
313 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
314
315 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
316
317 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
318 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
319 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
320 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
321
322 case vmIntrinsics::_compressStringC:
323 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
324 case vmIntrinsics::_inflateStringC:
325 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
326
327 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
328 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
329 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
330 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
331 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
332 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
333 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
334 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
335 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
336
337 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
338 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
339 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
340 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
341 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
342 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
343 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
344 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
345 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
346
347 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
348 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
349 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
350 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
351 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
352 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
353 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
354 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
355 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
356
357 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
358 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
359 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
360 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
361 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
362 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
363 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
364 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
365 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
366
367 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
373 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
374 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
375 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
376
377 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
378 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
379 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
380 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
381 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
382 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
383 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
384 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
385 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
386
387 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
388 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
389 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
390 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
391 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
392 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
393 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
394 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
395 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
396
397 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
398 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
399 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
400 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
401 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
402 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
403 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
404 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
405 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
406
407 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
408 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
409 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
410 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
411 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
412 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
413 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
414 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
415 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
416
417 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
418 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
422
423 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
424 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
425 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
426 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
427 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
428 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
429 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
430 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
431 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
432 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
433 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
434 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
435 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
443
444 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
450 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
451 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
452 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
453 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
454 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
455 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
456 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
459
460 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
461 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
464
465 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
466 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
470
471 case vmIntrinsics::_loadFence:
472 case vmIntrinsics::_storeFence:
473 case vmIntrinsics::_storeStoreFence:
474 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
475
476 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
477
478 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
479 case vmIntrinsics::_currentThread: return inline_native_currentThread();
480 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
481
482 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
483 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
484
485 #if INCLUDE_JVMTI
486 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
487 "notifyJvmtiStart", true, false);
488 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
489 "notifyJvmtiEnd", false, true);
490 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
491 "notifyJvmtiMount", false, false);
492 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
493 "notifyJvmtiUnmount", false, false);
494 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
495 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
496 #endif
497
498 #ifdef JFR_HAVE_INTRINSICS
499 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
500 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
501 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
502 #endif
503 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
504 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
505 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
506 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
507 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
508 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
509 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
510 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
511 case vmIntrinsics::_getLength: return inline_native_getLength();
512 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
513 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
514 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
515 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
516 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
517 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
518 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
519
520 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
521 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
522
523 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
524
525 case vmIntrinsics::_isInstance:
526 case vmIntrinsics::_getModifiers:
527 case vmIntrinsics::_isInterface:
528 case vmIntrinsics::_isArray:
529 case vmIntrinsics::_isPrimitive:
530 case vmIntrinsics::_isHidden:
531 case vmIntrinsics::_getSuperclass:
532 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
533
534 case vmIntrinsics::_floatToRawIntBits:
535 case vmIntrinsics::_floatToIntBits:
536 case vmIntrinsics::_intBitsToFloat:
537 case vmIntrinsics::_doubleToRawLongBits:
538 case vmIntrinsics::_doubleToLongBits:
539 case vmIntrinsics::_longBitsToDouble:
540 case vmIntrinsics::_floatToFloat16:
541 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
542
543 case vmIntrinsics::_floatIsFinite:
544 case vmIntrinsics::_floatIsInfinite:
545 case vmIntrinsics::_doubleIsFinite:
546 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
547
548 case vmIntrinsics::_numberOfLeadingZeros_i:
549 case vmIntrinsics::_numberOfLeadingZeros_l:
550 case vmIntrinsics::_numberOfTrailingZeros_i:
551 case vmIntrinsics::_numberOfTrailingZeros_l:
552 case vmIntrinsics::_bitCount_i:
553 case vmIntrinsics::_bitCount_l:
554 case vmIntrinsics::_reverse_i:
555 case vmIntrinsics::_reverse_l:
556 case vmIntrinsics::_reverseBytes_i:
557 case vmIntrinsics::_reverseBytes_l:
558 case vmIntrinsics::_reverseBytes_s:
559 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
560
561 case vmIntrinsics::_compress_i:
562 case vmIntrinsics::_compress_l:
563 case vmIntrinsics::_expand_i:
564 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
565
566 case vmIntrinsics::_compareUnsigned_i:
567 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
568
569 case vmIntrinsics::_divideUnsigned_i:
570 case vmIntrinsics::_divideUnsigned_l:
571 case vmIntrinsics::_remainderUnsigned_i:
572 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
573
574 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
575
576 case vmIntrinsics::_Reference_get: return inline_reference_get();
577 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
578 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
579
580 case vmIntrinsics::_Class_cast: return inline_Class_cast();
581
582 case vmIntrinsics::_aescrypt_encryptBlock:
583 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
584
585 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
586 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
587 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
588
589 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
590 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
591 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
592
593 case vmIntrinsics::_counterMode_AESCrypt:
594 return inline_counterMode_AESCrypt(intrinsic_id());
595
596 case vmIntrinsics::_galoisCounterMode_AESCrypt:
597 return inline_galoisCounterMode_AESCrypt();
598
599 case vmIntrinsics::_md5_implCompress:
600 case vmIntrinsics::_sha_implCompress:
601 case vmIntrinsics::_sha2_implCompress:
602 case vmIntrinsics::_sha5_implCompress:
603 case vmIntrinsics::_sha3_implCompress:
604 return inline_digestBase_implCompress(intrinsic_id());
605
606 case vmIntrinsics::_digestBase_implCompressMB:
607 return inline_digestBase_implCompressMB(predicate);
608
609 case vmIntrinsics::_multiplyToLen:
610 return inline_multiplyToLen();
611
612 case vmIntrinsics::_squareToLen:
613 return inline_squareToLen();
614
615 case vmIntrinsics::_mulAdd:
616 return inline_mulAdd();
617
618 case vmIntrinsics::_montgomeryMultiply:
619 return inline_montgomeryMultiply();
620 case vmIntrinsics::_montgomerySquare:
621 return inline_montgomerySquare();
622
623 case vmIntrinsics::_bigIntegerRightShiftWorker:
624 return inline_bigIntegerShift(true);
625 case vmIntrinsics::_bigIntegerLeftShiftWorker:
626 return inline_bigIntegerShift(false);
627
628 case vmIntrinsics::_vectorizedMismatch:
629 return inline_vectorizedMismatch();
630
631 case vmIntrinsics::_ghash_processBlocks:
632 return inline_ghash_processBlocks();
633 case vmIntrinsics::_chacha20Block:
634 return inline_chacha20Block();
635 case vmIntrinsics::_base64_encodeBlock:
636 return inline_base64_encodeBlock();
637 case vmIntrinsics::_base64_decodeBlock:
638 return inline_base64_decodeBlock();
639 case vmIntrinsics::_poly1305_processBlocks:
640 return inline_poly1305_processBlocks();
641
642 case vmIntrinsics::_encodeISOArray:
643 case vmIntrinsics::_encodeByteISOArray:
644 return inline_encodeISOArray(false);
645 case vmIntrinsics::_encodeAsciiArray:
646 return inline_encodeISOArray(true);
647
648 case vmIntrinsics::_updateCRC32:
649 return inline_updateCRC32();
650 case vmIntrinsics::_updateBytesCRC32:
651 return inline_updateBytesCRC32();
652 case vmIntrinsics::_updateByteBufferCRC32:
653 return inline_updateByteBufferCRC32();
654
655 case vmIntrinsics::_updateBytesCRC32C:
656 return inline_updateBytesCRC32C();
657 case vmIntrinsics::_updateDirectByteBufferCRC32C:
658 return inline_updateDirectByteBufferCRC32C();
659
660 case vmIntrinsics::_updateBytesAdler32:
661 return inline_updateBytesAdler32();
662 case vmIntrinsics::_updateByteBufferAdler32:
663 return inline_updateByteBufferAdler32();
664
665 case vmIntrinsics::_profileBoolean:
666 return inline_profileBoolean();
667 case vmIntrinsics::_isCompileConstant:
668 return inline_isCompileConstant();
669
670 case vmIntrinsics::_countPositives:
671 return inline_countPositives();
672
673 case vmIntrinsics::_fmaD:
674 case vmIntrinsics::_fmaF:
675 return inline_fma(intrinsic_id());
676
677 case vmIntrinsics::_isDigit:
678 case vmIntrinsics::_isLowerCase:
679 case vmIntrinsics::_isUpperCase:
680 case vmIntrinsics::_isWhitespace:
681 return inline_character_compare(intrinsic_id());
682
683 case vmIntrinsics::_min:
684 case vmIntrinsics::_max:
685 case vmIntrinsics::_min_strict:
686 case vmIntrinsics::_max_strict:
687 return inline_min_max(intrinsic_id());
688
689 case vmIntrinsics::_maxF:
690 case vmIntrinsics::_minF:
691 case vmIntrinsics::_maxD:
692 case vmIntrinsics::_minD:
693 case vmIntrinsics::_maxF_strict:
694 case vmIntrinsics::_minF_strict:
695 case vmIntrinsics::_maxD_strict:
696 case vmIntrinsics::_minD_strict:
697 return inline_fp_min_max(intrinsic_id());
698
699 case vmIntrinsics::_VectorUnaryOp:
700 return inline_vector_nary_operation(1);
701 case vmIntrinsics::_VectorBinaryOp:
702 return inline_vector_nary_operation(2);
703 case vmIntrinsics::_VectorTernaryOp:
704 return inline_vector_nary_operation(3);
705 case vmIntrinsics::_VectorFromBitsCoerced:
706 return inline_vector_frombits_coerced();
707 case vmIntrinsics::_VectorShuffleIota:
708 return inline_vector_shuffle_iota();
709 case vmIntrinsics::_VectorMaskOp:
710 return inline_vector_mask_operation();
711 case vmIntrinsics::_VectorShuffleToVector:
712 return inline_vector_shuffle_to_vector();
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(control(), index, TypeInt::POS);
877 (*pos_index) = _gvn.transform(ccast);
878 }
879 return is_neg;
880 }
881
882 // Make sure that 'position' is a valid limit index, in [0..length].
883 // There are two equivalent plans for checking this:
884 // A. (offset + copyLength) unsigned<= arrayLength
885 // B. offset <= (arrayLength - copyLength)
886 // We require that all of the values above, except for the sum and
887 // difference, are already known to be non-negative.
888 // Plan A is robust in the face of overflow, if offset and copyLength
889 // are both hugely positive.
890 //
891 // Plan B is less direct and intuitive, but it does not overflow at
892 // all, since the difference of two non-negatives is always
893 // representable. Whenever Java methods must perform the equivalent
894 // check they generally use Plan B instead of Plan A.
895 // For the moment we use Plan A.
896 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
897 Node* subseq_length,
898 Node* array_length,
899 RegionNode* region) {
900 if (stopped())
901 return nullptr; // already stopped
902 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
903 if (zero_offset && subseq_length->eqv_uncast(array_length))
904 return nullptr; // common case of whole-array copy
905 Node* last = subseq_length;
906 if (!zero_offset) // last += offset
907 last = _gvn.transform(new AddINode(last, offset));
908 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
909 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
910 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
911 return is_over;
912 }
913
914 // Emit range checks for the given String.value byte array
915 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
916 if (stopped()) {
917 return; // already stopped
918 }
919 RegionNode* bailout = new RegionNode(1);
920 record_for_igvn(bailout);
921 if (char_count) {
922 // Convert char count to byte count
923 count = _gvn.transform(new LShiftINode(count, intcon(1)));
924 }
925
926 // Offset and count must not be negative
927 generate_negative_guard(offset, bailout);
928 generate_negative_guard(count, bailout);
929 // Offset + count must not exceed length of array
930 generate_limit_guard(offset, count, load_array_length(array), bailout);
931
932 if (bailout->req() > 1) {
933 PreserveJVMState pjvms(this);
934 set_control(_gvn.transform(bailout));
935 uncommon_trap(Deoptimization::Reason_intrinsic,
936 Deoptimization::Action_maybe_recompile);
937 }
938 }
939
940 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
941 bool is_immutable) {
942 ciKlass* thread_klass = env()->Thread_klass();
943 const Type* thread_type
944 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
945
946 Node* thread = _gvn.transform(new ThreadLocalNode());
947 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
948 tls_output = thread;
949
950 Node* thread_obj_handle
951 = (is_immutable
952 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
953 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
954 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
955 thread_obj_handle = _gvn.transform(thread_obj_handle);
956
957 DecoratorSet decorators = IN_NATIVE;
958 if (is_immutable) {
959 decorators |= C2_IMMUTABLE_MEMORY;
960 }
961 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
962 }
963
964 //--------------------------generate_current_thread--------------------
965 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
966 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
967 /*is_immutable*/false);
968 }
969
970 //--------------------------generate_virtual_thread--------------------
971 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
972 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
973 !C->method()->changes_current_thread());
974 }
975
976 //------------------------------make_string_method_node------------------------
977 // Helper method for String intrinsic functions. This version is called with
978 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
979 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
980 // containing the lengths of str1 and str2.
981 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
982 Node* result = nullptr;
983 switch (opcode) {
984 case Op_StrIndexOf:
985 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
986 str1_start, cnt1, str2_start, cnt2, ae);
987 break;
988 case Op_StrComp:
989 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
990 str1_start, cnt1, str2_start, cnt2, ae);
991 break;
992 case Op_StrEquals:
993 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
994 // Use the constant length if there is one because optimized match rule may exist.
995 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
996 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
997 break;
998 default:
999 ShouldNotReachHere();
1000 return nullptr;
1001 }
1002
1003 // All these intrinsics have checks.
1004 C->set_has_split_ifs(true); // Has chance for split-if optimization
1005 clear_upper_avx();
1006
1007 return _gvn.transform(result);
1008 }
1009
1010 //------------------------------inline_string_compareTo------------------------
1011 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1012 Node* arg1 = argument(0);
1013 Node* arg2 = argument(1);
1014
1015 arg1 = must_be_not_null(arg1, true);
1016 arg2 = must_be_not_null(arg2, true);
1017
1018 // Get start addr and length of first argument
1019 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1020 Node* arg1_cnt = load_array_length(arg1);
1021
1022 // Get start addr and length of second argument
1023 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1024 Node* arg2_cnt = load_array_length(arg2);
1025
1026 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1027 set_result(result);
1028 return true;
1029 }
1030
1031 //------------------------------inline_string_equals------------------------
1032 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1033 Node* arg1 = argument(0);
1034 Node* arg2 = argument(1);
1035
1036 // paths (plus control) merge
1037 RegionNode* region = new RegionNode(3);
1038 Node* phi = new PhiNode(region, TypeInt::BOOL);
1039
1040 if (!stopped()) {
1041
1042 arg1 = must_be_not_null(arg1, true);
1043 arg2 = must_be_not_null(arg2, true);
1044
1045 // Get start addr and length of first argument
1046 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1047 Node* arg1_cnt = load_array_length(arg1);
1048
1049 // Get start addr and length of second argument
1050 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1051 Node* arg2_cnt = load_array_length(arg2);
1052
1053 // Check for arg1_cnt != arg2_cnt
1054 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1055 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1056 Node* if_ne = generate_slow_guard(bol, nullptr);
1057 if (if_ne != nullptr) {
1058 phi->init_req(2, intcon(0));
1059 region->init_req(2, if_ne);
1060 }
1061
1062 // Check for count == 0 is done by assembler code for StrEquals.
1063
1064 if (!stopped()) {
1065 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1066 phi->init_req(1, equals);
1067 region->init_req(1, control());
1068 }
1069 }
1070
1071 // post merge
1072 set_control(_gvn.transform(region));
1073 record_for_igvn(region);
1074
1075 set_result(_gvn.transform(phi));
1076 return true;
1077 }
1078
1079 //------------------------------inline_array_equals----------------------------
1080 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1081 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1082 Node* arg1 = argument(0);
1083 Node* arg2 = argument(1);
1084
1085 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1086 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1087 clear_upper_avx();
1088
1089 return true;
1090 }
1091
1092
1093 //------------------------------inline_countPositives------------------------------
1094 bool LibraryCallKit::inline_countPositives() {
1095 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1096 return false;
1097 }
1098
1099 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1100 // no receiver since it is static method
1101 Node* ba = argument(0);
1102 Node* offset = argument(1);
1103 Node* len = argument(2);
1104
1105 ba = must_be_not_null(ba, true);
1106
1107 // Range checks
1108 generate_string_range_check(ba, offset, len, false);
1109 if (stopped()) {
1110 return true;
1111 }
1112 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1113 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1114 set_result(_gvn.transform(result));
1115 clear_upper_avx();
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_cast_for_basic_type(
1144 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1145 ConstraintCastNode::RegularDependency, bt);
1146 casted_length = _gvn.transform(casted_length);
1147 replace_in_map(length, casted_length);
1148 length = casted_length;
1149
1150 // Use an unsigned comparison for the range check itself
1151 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1152 BoolTest::mask btest = BoolTest::lt;
1153 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1154 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1155 _gvn.set_type(rc, rc->Value(&_gvn));
1156 if (!rc_bool->is_Con()) {
1157 record_for_igvn(rc);
1158 }
1159 set_control(_gvn.transform(new IfTrueNode(rc)));
1160 {
1161 PreserveJVMState pjvms(this);
1162 set_control(_gvn.transform(new IfFalseNode(rc)));
1163 uncommon_trap(Deoptimization::Reason_range_check,
1164 Deoptimization::Action_make_not_entrant);
1165 }
1166
1167 if (stopped()) {
1168 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1169 return true;
1170 }
1171
1172 // index is now known to be >= 0 and < length, cast it
1173 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1174 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1175 ConstraintCastNode::RegularDependency, bt);
1176 result = _gvn.transform(result);
1177 set_result(result);
1178 replace_in_map(index, result);
1179 return true;
1180 }
1181
1182 //------------------------------inline_string_indexOf------------------------
1183 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1184 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1185 return false;
1186 }
1187 Node* src = argument(0);
1188 Node* tgt = argument(1);
1189
1190 // Make the merge point
1191 RegionNode* result_rgn = new RegionNode(4);
1192 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1193
1194 src = must_be_not_null(src, true);
1195 tgt = must_be_not_null(tgt, true);
1196
1197 // Get start addr and length of source string
1198 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1199 Node* src_count = load_array_length(src);
1200
1201 // Get start addr and length of substring
1202 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1203 Node* tgt_count = load_array_length(tgt);
1204
1205 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1206 // Divide src size by 2 if String is UTF16 encoded
1207 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1208 }
1209 if (ae == StrIntrinsicNode::UU) {
1210 // Divide substring size by 2 if String is UTF16 encoded
1211 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1212 }
1213
1214 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, result_rgn, result_phi, ae);
1215 if (result != nullptr) {
1216 result_phi->init_req(3, result);
1217 result_rgn->init_req(3, control());
1218 }
1219 set_control(_gvn.transform(result_rgn));
1220 record_for_igvn(result_rgn);
1221 set_result(_gvn.transform(result_phi));
1222
1223 return true;
1224 }
1225
1226 //-----------------------------inline_string_indexOf-----------------------
1227 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1228 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1229 return false;
1230 }
1231 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1232 return false;
1233 }
1234 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1235 Node* src = argument(0); // byte[]
1236 Node* src_count = argument(1); // char count
1237 Node* tgt = argument(2); // byte[]
1238 Node* tgt_count = argument(3); // char count
1239 Node* from_index = argument(4); // char index
1240
1241 src = must_be_not_null(src, true);
1242 tgt = must_be_not_null(tgt, true);
1243
1244 // Multiply byte array index by 2 if String is UTF16 encoded
1245 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1246 src_count = _gvn.transform(new SubINode(src_count, from_index));
1247 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1248 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1249
1250 // Range checks
1251 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1252 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1253 if (stopped()) {
1254 return true;
1255 }
1256
1257 RegionNode* region = new RegionNode(5);
1258 Node* phi = new PhiNode(region, TypeInt::INT);
1259
1260 Node* result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count, region, phi, ae);
1261 if (result != nullptr) {
1262 // The result is index relative to from_index if substring was found, -1 otherwise.
1263 // Generate code which will fold into cmove.
1264 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1265 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1266
1267 Node* if_lt = generate_slow_guard(bol, nullptr);
1268 if (if_lt != nullptr) {
1269 // result == -1
1270 phi->init_req(3, result);
1271 region->init_req(3, if_lt);
1272 }
1273 if (!stopped()) {
1274 result = _gvn.transform(new AddINode(result, from_index));
1275 phi->init_req(4, result);
1276 region->init_req(4, control());
1277 }
1278 }
1279
1280 set_control(_gvn.transform(region));
1281 record_for_igvn(region);
1282 set_result(_gvn.transform(phi));
1283 clear_upper_avx();
1284
1285 return true;
1286 }
1287
1288 // Create StrIndexOfNode with fast path checks
1289 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1290 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1291 // Check for substr count > string count
1292 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1293 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1294 Node* if_gt = generate_slow_guard(bol, nullptr);
1295 if (if_gt != nullptr) {
1296 phi->init_req(1, intcon(-1));
1297 region->init_req(1, if_gt);
1298 }
1299 if (!stopped()) {
1300 // Check for substr count == 0
1301 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1302 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1303 Node* if_zero = generate_slow_guard(bol, nullptr);
1304 if (if_zero != nullptr) {
1305 phi->init_req(2, intcon(0));
1306 region->init_req(2, if_zero);
1307 }
1308 }
1309 if (!stopped()) {
1310 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1311 }
1312 return nullptr;
1313 }
1314
1315 //-----------------------------inline_string_indexOfChar-----------------------
1316 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1317 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1318 return false;
1319 }
1320 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1321 return false;
1322 }
1323 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1324 Node* src = argument(0); // byte[]
1325 Node* int_ch = argument(1);
1326 Node* from_index = argument(2);
1327 Node* max = argument(3);
1328
1329 src = must_be_not_null(src, true);
1330
1331 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1332 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1333 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1334
1335 // Range checks
1336 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1337
1338 // Check for int_ch >= 0
1339 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1340 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1341 {
1342 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1343 uncommon_trap(Deoptimization::Reason_intrinsic,
1344 Deoptimization::Action_maybe_recompile);
1345 }
1346 if (stopped()) {
1347 return true;
1348 }
1349
1350 RegionNode* region = new RegionNode(3);
1351 Node* phi = new PhiNode(region, TypeInt::INT);
1352
1353 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1354 C->set_has_split_ifs(true); // Has chance for split-if optimization
1355 _gvn.transform(result);
1356
1357 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1359
1360 Node* if_lt = generate_slow_guard(bol, nullptr);
1361 if (if_lt != nullptr) {
1362 // result == -1
1363 phi->init_req(2, result);
1364 region->init_req(2, if_lt);
1365 }
1366 if (!stopped()) {
1367 result = _gvn.transform(new AddINode(result, from_index));
1368 phi->init_req(1, result);
1369 region->init_req(1, control());
1370 }
1371 set_control(_gvn.transform(region));
1372 record_for_igvn(region);
1373 set_result(_gvn.transform(phi));
1374 clear_upper_avx();
1375
1376 return true;
1377 }
1378 //---------------------------inline_string_copy---------------------
1379 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1380 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1381 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1382 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1383 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1384 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1385 bool LibraryCallKit::inline_string_copy(bool compress) {
1386 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1387 return false;
1388 }
1389 int nargs = 5; // 2 oops, 3 ints
1390 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1391
1392 Node* src = argument(0);
1393 Node* src_offset = argument(1);
1394 Node* dst = argument(2);
1395 Node* dst_offset = argument(3);
1396 Node* length = argument(4);
1397
1398 // Check for allocation before we add nodes that would confuse
1399 // tightly_coupled_allocation()
1400 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1401
1402 // Figure out the size and type of the elements we will be copying.
1403 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1404 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1405 if (src_type == nullptr || dst_type == nullptr) {
1406 return false;
1407 }
1408 BasicType src_elem = src_type->elem()->array_element_basic_type();
1409 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1410 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1411 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1412 "Unsupported array types for inline_string_copy");
1413
1414 src = must_be_not_null(src, true);
1415 dst = must_be_not_null(dst, true);
1416
1417 // Convert char[] offsets to byte[] offsets
1418 bool convert_src = (compress && src_elem == T_BYTE);
1419 bool convert_dst = (!compress && dst_elem == T_BYTE);
1420 if (convert_src) {
1421 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1422 } else if (convert_dst) {
1423 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1424 }
1425
1426 // Range checks
1427 generate_string_range_check(src, src_offset, length, convert_src);
1428 generate_string_range_check(dst, dst_offset, length, convert_dst);
1429 if (stopped()) {
1430 return true;
1431 }
1432
1433 Node* src_start = array_element_address(src, src_offset, src_elem);
1434 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1435 // 'src_start' points to src array + scaled offset
1436 // 'dst_start' points to dst array + scaled offset
1437 Node* count = nullptr;
1438 if (compress) {
1439 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1440 } else {
1441 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1442 }
1443
1444 if (alloc != nullptr) {
1445 if (alloc->maybe_set_complete(&_gvn)) {
1446 // "You break it, you buy it."
1447 InitializeNode* init = alloc->initialization();
1448 assert(init->is_complete(), "we just did this");
1449 init->set_complete_with_arraycopy();
1450 assert(dst->is_CheckCastPP(), "sanity");
1451 assert(dst->in(0)->in(0) == init, "dest pinned");
1452 }
1453 // Do not let stores that initialize this object be reordered with
1454 // a subsequent store that would make this object accessible by
1455 // other threads.
1456 // Record what AllocateNode this StoreStore protects so that
1457 // escape analysis can go from the MemBarStoreStoreNode to the
1458 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1459 // based on the escape status of the AllocateNode.
1460 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1461 }
1462 if (compress) {
1463 set_result(_gvn.transform(count));
1464 }
1465 clear_upper_avx();
1466
1467 return true;
1468 }
1469
1470 #ifdef _LP64
1471 #define XTOP ,top() /*additional argument*/
1472 #else //_LP64
1473 #define XTOP /*no additional argument*/
1474 #endif //_LP64
1475
1476 //------------------------inline_string_toBytesU--------------------------
1477 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1478 bool LibraryCallKit::inline_string_toBytesU() {
1479 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1480 return false;
1481 }
1482 // Get the arguments.
1483 Node* value = argument(0);
1484 Node* offset = argument(1);
1485 Node* length = argument(2);
1486
1487 Node* newcopy = nullptr;
1488
1489 // Set the original stack and the reexecute bit for the interpreter to reexecute
1490 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1491 { PreserveReexecuteState preexecs(this);
1492 jvms()->set_should_reexecute(true);
1493
1494 // Check if a null path was taken unconditionally.
1495 value = null_check(value);
1496
1497 RegionNode* bailout = new RegionNode(1);
1498 record_for_igvn(bailout);
1499
1500 // Range checks
1501 generate_negative_guard(offset, bailout);
1502 generate_negative_guard(length, bailout);
1503 generate_limit_guard(offset, length, load_array_length(value), bailout);
1504 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1505 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1506
1507 if (bailout->req() > 1) {
1508 PreserveJVMState pjvms(this);
1509 set_control(_gvn.transform(bailout));
1510 uncommon_trap(Deoptimization::Reason_intrinsic,
1511 Deoptimization::Action_maybe_recompile);
1512 }
1513 if (stopped()) {
1514 return true;
1515 }
1516
1517 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1518 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1519 newcopy = new_array(klass_node, size, 0); // no arguments to push
1520 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1521 guarantee(alloc != nullptr, "created above");
1522
1523 // Calculate starting addresses.
1524 Node* src_start = array_element_address(value, offset, T_CHAR);
1525 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1526
1527 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1528 const TypeInt* toffset = gvn().type(offset)->is_int();
1529 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1530
1531 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1532 const char* copyfunc_name = "arraycopy";
1533 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1534 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1535 OptoRuntime::fast_arraycopy_Type(),
1536 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1537 src_start, dst_start, ConvI2X(length) XTOP);
1538 // Do not let reads from the cloned object float above the arraycopy.
1539 if (alloc->maybe_set_complete(&_gvn)) {
1540 // "You break it, you buy it."
1541 InitializeNode* init = alloc->initialization();
1542 assert(init->is_complete(), "we just did this");
1543 init->set_complete_with_arraycopy();
1544 assert(newcopy->is_CheckCastPP(), "sanity");
1545 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1546 }
1547 // Do not let stores that initialize this object be reordered with
1548 // a subsequent store that would make this object accessible by
1549 // other threads.
1550 // Record what AllocateNode this StoreStore protects so that
1551 // escape analysis can go from the MemBarStoreStoreNode to the
1552 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1553 // based on the escape status of the AllocateNode.
1554 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1555 } // original reexecute is set back here
1556
1557 C->set_has_split_ifs(true); // Has chance for split-if optimization
1558 if (!stopped()) {
1559 set_result(newcopy);
1560 }
1561 clear_upper_avx();
1562
1563 return true;
1564 }
1565
1566 //------------------------inline_string_getCharsU--------------------------
1567 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1568 bool LibraryCallKit::inline_string_getCharsU() {
1569 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1570 return false;
1571 }
1572
1573 // Get the arguments.
1574 Node* src = argument(0);
1575 Node* src_begin = argument(1);
1576 Node* src_end = argument(2); // exclusive offset (i < src_end)
1577 Node* dst = argument(3);
1578 Node* dst_begin = argument(4);
1579
1580 // Check for allocation before we add nodes that would confuse
1581 // tightly_coupled_allocation()
1582 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1583
1584 // Check if a null path was taken unconditionally.
1585 src = null_check(src);
1586 dst = null_check(dst);
1587 if (stopped()) {
1588 return true;
1589 }
1590
1591 // Get length and convert char[] offset to byte[] offset
1592 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1593 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1594
1595 // Range checks
1596 generate_string_range_check(src, src_begin, length, true);
1597 generate_string_range_check(dst, dst_begin, length, false);
1598 if (stopped()) {
1599 return true;
1600 }
1601
1602 if (!stopped()) {
1603 // Calculate starting addresses.
1604 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1605 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1606
1607 // Check if array addresses are aligned to HeapWordSize
1608 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1609 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1610 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1611 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1612
1613 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1614 const char* copyfunc_name = "arraycopy";
1615 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1616 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1617 OptoRuntime::fast_arraycopy_Type(),
1618 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1619 src_start, dst_start, ConvI2X(length) XTOP);
1620 // Do not let reads from the cloned object float above the arraycopy.
1621 if (alloc != nullptr) {
1622 if (alloc->maybe_set_complete(&_gvn)) {
1623 // "You break it, you buy it."
1624 InitializeNode* init = alloc->initialization();
1625 assert(init->is_complete(), "we just did this");
1626 init->set_complete_with_arraycopy();
1627 assert(dst->is_CheckCastPP(), "sanity");
1628 assert(dst->in(0)->in(0) == init, "dest pinned");
1629 }
1630 // Do not let stores that initialize this object be reordered with
1631 // a subsequent store that would make this object accessible by
1632 // other threads.
1633 // Record what AllocateNode this StoreStore protects so that
1634 // escape analysis can go from the MemBarStoreStoreNode to the
1635 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1636 // based on the escape status of the AllocateNode.
1637 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1638 } else {
1639 insert_mem_bar(Op_MemBarCPUOrder);
1640 }
1641 }
1642
1643 C->set_has_split_ifs(true); // Has chance for split-if optimization
1644 return true;
1645 }
1646
1647 //----------------------inline_string_char_access----------------------------
1648 // Store/Load char to/from byte[] array.
1649 // static void StringUTF16.putChar(byte[] val, int index, int c)
1650 // static char StringUTF16.getChar(byte[] val, int index)
1651 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1652 Node* value = argument(0);
1653 Node* index = argument(1);
1654 Node* ch = is_store ? argument(2) : nullptr;
1655
1656 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1657 // correctly requires matched array shapes.
1658 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1659 "sanity: byte[] and char[] bases agree");
1660 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1661 "sanity: byte[] and char[] scales agree");
1662
1663 // Bail when getChar over constants is requested: constant folding would
1664 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1665 // Java method would constant fold nicely instead.
1666 if (!is_store && value->is_Con() && index->is_Con()) {
1667 return false;
1668 }
1669
1670 // Save state and restore on bailout
1671 uint old_sp = sp();
1672 SafePointNode* old_map = clone_map();
1673
1674 value = must_be_not_null(value, true);
1675
1676 Node* adr = array_element_address(value, index, T_CHAR);
1677 if (adr->is_top()) {
1678 set_map(old_map);
1679 set_sp(old_sp);
1680 return false;
1681 }
1682 destruct_map_clone(old_map);
1683 if (is_store) {
1684 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1685 } else {
1686 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);
1687 set_result(ch);
1688 }
1689 return true;
1690 }
1691
1692 //--------------------------round_double_node--------------------------------
1693 // Round a double node if necessary.
1694 Node* LibraryCallKit::round_double_node(Node* n) {
1695 if (Matcher::strict_fp_requires_explicit_rounding) {
1696 #ifdef IA32
1697 if (UseSSE < 2) {
1698 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1699 }
1700 #else
1701 Unimplemented();
1702 #endif // IA32
1703 }
1704 return n;
1705 }
1706
1707 //------------------------------inline_math-----------------------------------
1708 // public static double Math.abs(double)
1709 // public static double Math.sqrt(double)
1710 // public static double Math.log(double)
1711 // public static double Math.log10(double)
1712 // public static double Math.round(double)
1713 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1714 Node* arg = round_double_node(argument(0));
1715 Node* n = nullptr;
1716 switch (id) {
1717 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1718 case vmIntrinsics::_dsqrt:
1719 case vmIntrinsics::_dsqrt_strict:
1720 n = new SqrtDNode(C, control(), arg); break;
1721 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1722 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1723 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1724 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1725 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1726 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1727 default: fatal_unexpected_iid(id); break;
1728 }
1729 set_result(_gvn.transform(n));
1730 return true;
1731 }
1732
1733 //------------------------------inline_math-----------------------------------
1734 // public static float Math.abs(float)
1735 // public static int Math.abs(int)
1736 // public static long Math.abs(long)
1737 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1738 Node* arg = argument(0);
1739 Node* n = nullptr;
1740 switch (id) {
1741 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1742 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1743 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1744 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1745 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1746 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1747 default: fatal_unexpected_iid(id); break;
1748 }
1749 set_result(_gvn.transform(n));
1750 return true;
1751 }
1752
1753 //------------------------------runtime_math-----------------------------
1754 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1755 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1756 "must be (DD)D or (D)D type");
1757
1758 // Inputs
1759 Node* a = round_double_node(argument(0));
1760 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1761
1762 const TypePtr* no_memory_effects = nullptr;
1763 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1764 no_memory_effects,
1765 a, top(), b, b ? top() : nullptr);
1766 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1767 #ifdef ASSERT
1768 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1769 assert(value_top == top(), "second value must be top");
1770 #endif
1771
1772 set_result(value);
1773 return true;
1774 }
1775
1776 //------------------------------inline_math_pow-----------------------------
1777 bool LibraryCallKit::inline_math_pow() {
1778 Node* exp = round_double_node(argument(2));
1779 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1780 if (d != nullptr) {
1781 if (d->getd() == 2.0) {
1782 // Special case: pow(x, 2.0) => x * x
1783 Node* base = round_double_node(argument(0));
1784 set_result(_gvn.transform(new MulDNode(base, base)));
1785 return true;
1786 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1787 // Special case: pow(x, 0.5) => sqrt(x)
1788 Node* base = round_double_node(argument(0));
1789 Node* zero = _gvn.zerocon(T_DOUBLE);
1790
1791 RegionNode* region = new RegionNode(3);
1792 Node* phi = new PhiNode(region, Type::DOUBLE);
1793
1794 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1795 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1796 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1797 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1798 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1799
1800 Node* if_pow = generate_slow_guard(test, nullptr);
1801 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1802 phi->init_req(1, value_sqrt);
1803 region->init_req(1, control());
1804
1805 if (if_pow != nullptr) {
1806 set_control(if_pow);
1807 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1808 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1809 const TypePtr* no_memory_effects = nullptr;
1810 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1811 no_memory_effects, base, top(), exp, top());
1812 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1813 #ifdef ASSERT
1814 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1815 assert(value_top == top(), "second value must be top");
1816 #endif
1817 phi->init_req(2, value_pow);
1818 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1819 }
1820
1821 C->set_has_split_ifs(true); // Has chance for split-if optimization
1822 set_control(_gvn.transform(region));
1823 record_for_igvn(region);
1824 set_result(_gvn.transform(phi));
1825
1826 return true;
1827 }
1828 }
1829
1830 return StubRoutines::dpow() != nullptr ?
1831 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1832 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1833 }
1834
1835 //------------------------------inline_math_native-----------------------------
1836 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1837 switch (id) {
1838 case vmIntrinsics::_dsin:
1839 return StubRoutines::dsin() != nullptr ?
1840 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1841 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1842 case vmIntrinsics::_dcos:
1843 return StubRoutines::dcos() != nullptr ?
1844 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1845 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1846 case vmIntrinsics::_dtan:
1847 return StubRoutines::dtan() != nullptr ?
1848 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1849 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1850 case vmIntrinsics::_dexp:
1851 return StubRoutines::dexp() != nullptr ?
1852 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1853 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1854 case vmIntrinsics::_dlog:
1855 return StubRoutines::dlog() != nullptr ?
1856 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1857 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1858 case vmIntrinsics::_dlog10:
1859 return StubRoutines::dlog10() != nullptr ?
1860 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1861 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1862
1863 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1864 case vmIntrinsics::_ceil:
1865 case vmIntrinsics::_floor:
1866 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1867
1868 case vmIntrinsics::_dsqrt:
1869 case vmIntrinsics::_dsqrt_strict:
1870 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1871 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1872 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1873 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1874 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1875
1876 case vmIntrinsics::_dpow: return inline_math_pow();
1877 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1878 case vmIntrinsics::_fcopySign: return inline_math(id);
1879 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1880 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1881 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1882
1883 // These intrinsics are not yet correctly implemented
1884 case vmIntrinsics::_datan2:
1885 return false;
1886
1887 default:
1888 fatal_unexpected_iid(id);
1889 return false;
1890 }
1891 }
1892
1893 //----------------------------inline_notify-----------------------------------*
1894 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1895 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1896 address func;
1897 if (id == vmIntrinsics::_notify) {
1898 func = OptoRuntime::monitor_notify_Java();
1899 } else {
1900 func = OptoRuntime::monitor_notifyAll_Java();
1901 }
1902 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1903 make_slow_call_ex(call, env()->Throwable_klass(), false);
1904 return true;
1905 }
1906
1907
1908 //----------------------------inline_min_max-----------------------------------
1909 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1910 set_result(generate_min_max(id, argument(0), argument(1)));
1911 return true;
1912 }
1913
1914 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1915 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1916 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1917 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1918 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1919
1920 {
1921 PreserveJVMState pjvms(this);
1922 PreserveReexecuteState preexecs(this);
1923 jvms()->set_should_reexecute(true);
1924
1925 set_control(slow_path);
1926 set_i_o(i_o());
1927
1928 uncommon_trap(Deoptimization::Reason_intrinsic,
1929 Deoptimization::Action_none);
1930 }
1931
1932 set_control(fast_path);
1933 set_result(math);
1934 }
1935
1936 template <typename OverflowOp>
1937 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1938 typedef typename OverflowOp::MathOp MathOp;
1939
1940 MathOp* mathOp = new MathOp(arg1, arg2);
1941 Node* operation = _gvn.transform( mathOp );
1942 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1943 inline_math_mathExact(operation, ofcheck);
1944 return true;
1945 }
1946
1947 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1948 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1949 }
1950
1951 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1952 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1953 }
1954
1955 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1956 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1957 }
1958
1959 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
1960 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
1961 }
1962
1963 bool LibraryCallKit::inline_math_negateExactI() {
1964 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
1965 }
1966
1967 bool LibraryCallKit::inline_math_negateExactL() {
1968 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
1969 }
1970
1971 bool LibraryCallKit::inline_math_multiplyExactI() {
1972 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
1973 }
1974
1975 bool LibraryCallKit::inline_math_multiplyExactL() {
1976 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
1977 }
1978
1979 bool LibraryCallKit::inline_math_multiplyHigh() {
1980 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
1981 return true;
1982 }
1983
1984 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
1985 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
1986 return true;
1987 }
1988
1989 Node*
1990 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
1991 Node* result_val = nullptr;
1992 switch (id) {
1993 case vmIntrinsics::_min:
1994 case vmIntrinsics::_min_strict:
1995 result_val = _gvn.transform(new MinINode(x0, y0));
1996 break;
1997 case vmIntrinsics::_max:
1998 case vmIntrinsics::_max_strict:
1999 result_val = _gvn.transform(new MaxINode(x0, y0));
2000 break;
2001 default:
2002 fatal_unexpected_iid(id);
2003 break;
2004 }
2005 return result_val;
2006 }
2007
2008 inline int
2009 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2010 const TypePtr* base_type = TypePtr::NULL_PTR;
2011 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2012 if (base_type == nullptr) {
2013 // Unknown type.
2014 return Type::AnyPtr;
2015 } else if (base_type == TypePtr::NULL_PTR) {
2016 // Since this is a null+long form, we have to switch to a rawptr.
2017 base = _gvn.transform(new CastX2PNode(offset));
2018 offset = MakeConX(0);
2019 return Type::RawPtr;
2020 } else if (base_type->base() == Type::RawPtr) {
2021 return Type::RawPtr;
2022 } else if (base_type->isa_oopptr()) {
2023 // Base is never null => always a heap address.
2024 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2025 return Type::OopPtr;
2026 }
2027 // Offset is small => always a heap address.
2028 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2029 if (offset_type != nullptr &&
2030 base_type->offset() == 0 && // (should always be?)
2031 offset_type->_lo >= 0 &&
2032 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2033 return Type::OopPtr;
2034 } else if (type == T_OBJECT) {
2035 // off heap access to an oop doesn't make any sense. Has to be on
2036 // heap.
2037 return Type::OopPtr;
2038 }
2039 // Otherwise, it might either be oop+off or null+addr.
2040 return Type::AnyPtr;
2041 } else {
2042 // No information:
2043 return Type::AnyPtr;
2044 }
2045 }
2046
2047 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2048 Node* uncasted_base = base;
2049 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2050 if (kind == Type::RawPtr) {
2051 return basic_plus_adr(top(), uncasted_base, offset);
2052 } else if (kind == Type::AnyPtr) {
2053 assert(base == uncasted_base, "unexpected base change");
2054 if (can_cast) {
2055 if (!_gvn.type(base)->speculative_maybe_null() &&
2056 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2057 // According to profiling, this access is always on
2058 // heap. Casting the base to not null and thus avoiding membars
2059 // around the access should allow better optimizations
2060 Node* null_ctl = top();
2061 base = null_check_oop(base, &null_ctl, true, true, true);
2062 assert(null_ctl->is_top(), "no null control here");
2063 return basic_plus_adr(base, offset);
2064 } else if (_gvn.type(base)->speculative_always_null() &&
2065 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2066 // According to profiling, this access is always off
2067 // heap.
2068 base = null_assert(base);
2069 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2070 offset = MakeConX(0);
2071 return basic_plus_adr(top(), raw_base, offset);
2072 }
2073 }
2074 // We don't know if it's an on heap or off heap access. Fall back
2075 // to raw memory access.
2076 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2077 return basic_plus_adr(top(), raw, offset);
2078 } else {
2079 assert(base == uncasted_base, "unexpected base change");
2080 // We know it's an on heap access so base can't be null
2081 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2082 base = must_be_not_null(base, true);
2083 }
2084 return basic_plus_adr(base, offset);
2085 }
2086 }
2087
2088 //--------------------------inline_number_methods-----------------------------
2089 // inline int Integer.numberOfLeadingZeros(int)
2090 // inline int Long.numberOfLeadingZeros(long)
2091 //
2092 // inline int Integer.numberOfTrailingZeros(int)
2093 // inline int Long.numberOfTrailingZeros(long)
2094 //
2095 // inline int Integer.bitCount(int)
2096 // inline int Long.bitCount(long)
2097 //
2098 // inline char Character.reverseBytes(char)
2099 // inline short Short.reverseBytes(short)
2100 // inline int Integer.reverseBytes(int)
2101 // inline long Long.reverseBytes(long)
2102 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2103 Node* arg = argument(0);
2104 Node* n = nullptr;
2105 switch (id) {
2106 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2107 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2108 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2109 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2110 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2111 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2112 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(0, arg); break;
2113 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( 0, arg); break;
2114 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( 0, arg); break;
2115 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( 0, arg); break;
2116 case vmIntrinsics::_reverse_i: n = new ReverseINode(0, arg); break;
2117 case vmIntrinsics::_reverse_l: n = new ReverseLNode(0, arg); break;
2118 default: fatal_unexpected_iid(id); break;
2119 }
2120 set_result(_gvn.transform(n));
2121 return true;
2122 }
2123
2124 //--------------------------inline_bitshuffle_methods-----------------------------
2125 // inline int Integer.compress(int, int)
2126 // inline int Integer.expand(int, int)
2127 // inline long Long.compress(long, long)
2128 // inline long Long.expand(long, long)
2129 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2130 Node* n = nullptr;
2131 switch (id) {
2132 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2133 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2134 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2135 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2136 default: fatal_unexpected_iid(id); break;
2137 }
2138 set_result(_gvn.transform(n));
2139 return true;
2140 }
2141
2142 //--------------------------inline_number_methods-----------------------------
2143 // inline int Integer.compareUnsigned(int, int)
2144 // inline int Long.compareUnsigned(long, long)
2145 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2146 Node* arg1 = argument(0);
2147 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2148 Node* n = nullptr;
2149 switch (id) {
2150 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2151 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2152 default: fatal_unexpected_iid(id); break;
2153 }
2154 set_result(_gvn.transform(n));
2155 return true;
2156 }
2157
2158 //--------------------------inline_unsigned_divmod_methods-----------------------------
2159 // inline int Integer.divideUnsigned(int, int)
2160 // inline int Integer.remainderUnsigned(int, int)
2161 // inline long Long.divideUnsigned(long, long)
2162 // inline long Long.remainderUnsigned(long, long)
2163 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2164 Node* n = nullptr;
2165 switch (id) {
2166 case vmIntrinsics::_divideUnsigned_i: {
2167 zero_check_int(argument(1));
2168 // Compile-time detect of null-exception
2169 if (stopped()) {
2170 return true; // keep the graph constructed so far
2171 }
2172 n = new UDivINode(control(), argument(0), argument(1));
2173 break;
2174 }
2175 case vmIntrinsics::_divideUnsigned_l: {
2176 zero_check_long(argument(2));
2177 // Compile-time detect of null-exception
2178 if (stopped()) {
2179 return true; // keep the graph constructed so far
2180 }
2181 n = new UDivLNode(control(), argument(0), argument(2));
2182 break;
2183 }
2184 case vmIntrinsics::_remainderUnsigned_i: {
2185 zero_check_int(argument(1));
2186 // Compile-time detect of null-exception
2187 if (stopped()) {
2188 return true; // keep the graph constructed so far
2189 }
2190 n = new UModINode(control(), argument(0), argument(1));
2191 break;
2192 }
2193 case vmIntrinsics::_remainderUnsigned_l: {
2194 zero_check_long(argument(2));
2195 // Compile-time detect of null-exception
2196 if (stopped()) {
2197 return true; // keep the graph constructed so far
2198 }
2199 n = new UModLNode(control(), argument(0), argument(2));
2200 break;
2201 }
2202 default: fatal_unexpected_iid(id); break;
2203 }
2204 set_result(_gvn.transform(n));
2205 return true;
2206 }
2207
2208 //----------------------------inline_unsafe_access----------------------------
2209
2210 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2211 // Attempt to infer a sharper value type from the offset and base type.
2212 ciKlass* sharpened_klass = nullptr;
2213
2214 // See if it is an instance field, with an object type.
2215 if (alias_type->field() != nullptr) {
2216 if (alias_type->field()->type()->is_klass()) {
2217 sharpened_klass = alias_type->field()->type()->as_klass();
2218 }
2219 }
2220
2221 const TypeOopPtr* result = nullptr;
2222 // See if it is a narrow oop array.
2223 if (adr_type->isa_aryptr()) {
2224 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2225 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2226 if (elem_type != nullptr && elem_type->is_loaded()) {
2227 // Sharpen the value type.
2228 result = elem_type;
2229 }
2230 }
2231 }
2232
2233 // The sharpened class might be unloaded if there is no class loader
2234 // contraint in place.
2235 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2236 // Sharpen the value type.
2237 result = TypeOopPtr::make_from_klass(sharpened_klass);
2238 }
2239 if (result != nullptr) {
2240 #ifndef PRODUCT
2241 if (C->print_intrinsics() || C->print_inlining()) {
2242 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2243 tty->print(" sharpened value: "); result->dump(); tty->cr();
2244 }
2245 #endif
2246 }
2247 return result;
2248 }
2249
2250 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2251 switch (kind) {
2252 case Relaxed:
2253 return MO_UNORDERED;
2254 case Opaque:
2255 return MO_RELAXED;
2256 case Acquire:
2257 return MO_ACQUIRE;
2258 case Release:
2259 return MO_RELEASE;
2260 case Volatile:
2261 return MO_SEQ_CST;
2262 default:
2263 ShouldNotReachHere();
2264 return 0;
2265 }
2266 }
2267
2268 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2269 if (callee()->is_static()) return false; // caller must have the capability!
2270 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2271 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2272 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2273 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2274
2275 if (is_reference_type(type)) {
2276 decorators |= ON_UNKNOWN_OOP_REF;
2277 }
2278
2279 if (unaligned) {
2280 decorators |= C2_UNALIGNED;
2281 }
2282
2283 #ifndef PRODUCT
2284 {
2285 ResourceMark rm;
2286 // Check the signatures.
2287 ciSignature* sig = callee()->signature();
2288 #ifdef ASSERT
2289 if (!is_store) {
2290 // Object getReference(Object base, int/long offset), etc.
2291 BasicType rtype = sig->return_type()->basic_type();
2292 assert(rtype == type, "getter must return the expected value");
2293 assert(sig->count() == 2, "oop getter has 2 arguments");
2294 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2295 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2296 } else {
2297 // void putReference(Object base, int/long offset, Object x), etc.
2298 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2299 assert(sig->count() == 3, "oop putter has 3 arguments");
2300 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2301 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2302 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2303 assert(vtype == type, "putter must accept the expected value");
2304 }
2305 #endif // ASSERT
2306 }
2307 #endif //PRODUCT
2308
2309 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2310
2311 Node* receiver = argument(0); // type: oop
2312
2313 // Build address expression.
2314 Node* heap_base_oop = top();
2315
2316 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2317 Node* base = argument(1); // type: oop
2318 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2319 Node* offset = argument(2); // type: long
2320 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2321 // to be plain byte offsets, which are also the same as those accepted
2322 // by oopDesc::field_addr.
2323 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2324 "fieldOffset must be byte-scaled");
2325 // 32-bit machines ignore the high half!
2326 offset = ConvL2X(offset);
2327
2328 // Save state and restore on bailout
2329 uint old_sp = sp();
2330 SafePointNode* old_map = clone_map();
2331
2332 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2333
2334 if (_gvn.type(base)->isa_ptr() == TypePtr::NULL_PTR) {
2335 if (type != T_OBJECT) {
2336 decorators |= IN_NATIVE; // off-heap primitive access
2337 } else {
2338 set_map(old_map);
2339 set_sp(old_sp);
2340 return false; // off-heap oop accesses are not supported
2341 }
2342 } else {
2343 heap_base_oop = base; // on-heap or mixed access
2344 }
2345
2346 // Can base be null? Otherwise, always on-heap access.
2347 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2348
2349 if (!can_access_non_heap) {
2350 decorators |= IN_HEAP;
2351 }
2352
2353 Node* val = is_store ? argument(4) : nullptr;
2354
2355 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2356 if (adr_type == TypePtr::NULL_PTR) {
2357 set_map(old_map);
2358 set_sp(old_sp);
2359 return false; // off-heap access with zero address
2360 }
2361
2362 // Try to categorize the address.
2363 Compile::AliasType* alias_type = C->alias_type(adr_type);
2364 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2365
2366 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2367 alias_type->adr_type() == TypeAryPtr::RANGE) {
2368 set_map(old_map);
2369 set_sp(old_sp);
2370 return false; // not supported
2371 }
2372
2373 bool mismatched = false;
2374 BasicType bt = alias_type->basic_type();
2375 if (bt != T_ILLEGAL) {
2376 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2377 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2378 // Alias type doesn't differentiate between byte[] and boolean[]).
2379 // Use address type to get the element type.
2380 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2381 }
2382 if (is_reference_type(bt, true)) {
2383 // accessing an array field with getReference is not a mismatch
2384 bt = T_OBJECT;
2385 }
2386 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2387 // Don't intrinsify mismatched object accesses
2388 set_map(old_map);
2389 set_sp(old_sp);
2390 return false;
2391 }
2392 mismatched = (bt != type);
2393 } else if (alias_type->adr_type()->isa_oopptr()) {
2394 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2395 }
2396
2397 destruct_map_clone(old_map);
2398 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2399
2400 if (mismatched) {
2401 decorators |= C2_MISMATCHED;
2402 }
2403
2404 // First guess at the value type.
2405 const Type *value_type = Type::get_const_basic_type(type);
2406
2407 // Figure out the memory ordering.
2408 decorators |= mo_decorator_for_access_kind(kind);
2409
2410 if (!is_store && type == T_OBJECT) {
2411 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2412 if (tjp != nullptr) {
2413 value_type = tjp;
2414 }
2415 }
2416
2417 receiver = null_check(receiver);
2418 if (stopped()) {
2419 return true;
2420 }
2421 // Heap pointers get a null-check from the interpreter,
2422 // as a courtesy. However, this is not guaranteed by Unsafe,
2423 // and it is not possible to fully distinguish unintended nulls
2424 // from intended ones in this API.
2425
2426 if (!is_store) {
2427 Node* p = nullptr;
2428 // Try to constant fold a load from a constant field
2429 ciField* field = alias_type->field();
2430 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2431 // final or stable field
2432 p = make_constant_from_field(field, heap_base_oop);
2433 }
2434
2435 if (p == nullptr) { // Could not constant fold the load
2436 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2437 // Normalize the value returned by getBoolean in the following cases
2438 if (type == T_BOOLEAN &&
2439 (mismatched ||
2440 heap_base_oop == top() || // - heap_base_oop is null or
2441 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2442 // and the unsafe access is made to large offset
2443 // (i.e., larger than the maximum offset necessary for any
2444 // field access)
2445 ) {
2446 IdealKit ideal = IdealKit(this);
2447 #define __ ideal.
2448 IdealVariable normalized_result(ideal);
2449 __ declarations_done();
2450 __ set(normalized_result, p);
2451 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2452 __ set(normalized_result, ideal.ConI(1));
2453 ideal.end_if();
2454 final_sync(ideal);
2455 p = __ value(normalized_result);
2456 #undef __
2457 }
2458 }
2459 if (type == T_ADDRESS) {
2460 p = gvn().transform(new CastP2XNode(nullptr, p));
2461 p = ConvX2UL(p);
2462 }
2463 // The load node has the control of the preceding MemBarCPUOrder. All
2464 // following nodes will have the control of the MemBarCPUOrder inserted at
2465 // the end of this method. So, pushing the load onto the stack at a later
2466 // point is fine.
2467 set_result(p);
2468 } else {
2469 if (bt == T_ADDRESS) {
2470 // Repackage the long as a pointer.
2471 val = ConvL2X(val);
2472 val = gvn().transform(new CastX2PNode(val));
2473 }
2474 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2475 }
2476
2477 return true;
2478 }
2479
2480 //----------------------------inline_unsafe_load_store----------------------------
2481 // This method serves a couple of different customers (depending on LoadStoreKind):
2482 //
2483 // LS_cmp_swap:
2484 //
2485 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2486 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2487 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2488 //
2489 // LS_cmp_swap_weak:
2490 //
2491 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2492 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2493 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2494 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2495 //
2496 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2497 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2498 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2499 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2500 //
2501 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2502 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2503 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2504 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2505 //
2506 // LS_cmp_exchange:
2507 //
2508 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2509 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2510 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2511 //
2512 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2513 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2514 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2515 //
2516 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2517 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2518 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2519 //
2520 // LS_get_add:
2521 //
2522 // int getAndAddInt( Object o, long offset, int delta)
2523 // long getAndAddLong(Object o, long offset, long delta)
2524 //
2525 // LS_get_set:
2526 //
2527 // int getAndSet(Object o, long offset, int newValue)
2528 // long getAndSet(Object o, long offset, long newValue)
2529 // Object getAndSet(Object o, long offset, Object newValue)
2530 //
2531 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2532 // This basic scheme here is the same as inline_unsafe_access, but
2533 // differs in enough details that combining them would make the code
2534 // overly confusing. (This is a true fact! I originally combined
2535 // them, but even I was confused by it!) As much code/comments as
2536 // possible are retained from inline_unsafe_access though to make
2537 // the correspondences clearer. - dl
2538
2539 if (callee()->is_static()) return false; // caller must have the capability!
2540
2541 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2542 decorators |= mo_decorator_for_access_kind(access_kind);
2543
2544 #ifndef PRODUCT
2545 BasicType rtype;
2546 {
2547 ResourceMark rm;
2548 // Check the signatures.
2549 ciSignature* sig = callee()->signature();
2550 rtype = sig->return_type()->basic_type();
2551 switch(kind) {
2552 case LS_get_add:
2553 case LS_get_set: {
2554 // Check the signatures.
2555 #ifdef ASSERT
2556 assert(rtype == type, "get and set must return the expected type");
2557 assert(sig->count() == 3, "get and set has 3 arguments");
2558 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2559 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2560 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2561 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2562 #endif // ASSERT
2563 break;
2564 }
2565 case LS_cmp_swap:
2566 case LS_cmp_swap_weak: {
2567 // Check the signatures.
2568 #ifdef ASSERT
2569 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2570 assert(sig->count() == 4, "CAS has 4 arguments");
2571 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2572 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2573 #endif // ASSERT
2574 break;
2575 }
2576 case LS_cmp_exchange: {
2577 // Check the signatures.
2578 #ifdef ASSERT
2579 assert(rtype == type, "CAS must return the expected type");
2580 assert(sig->count() == 4, "CAS has 4 arguments");
2581 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2582 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2583 #endif // ASSERT
2584 break;
2585 }
2586 default:
2587 ShouldNotReachHere();
2588 }
2589 }
2590 #endif //PRODUCT
2591
2592 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2593
2594 // Get arguments:
2595 Node* receiver = nullptr;
2596 Node* base = nullptr;
2597 Node* offset = nullptr;
2598 Node* oldval = nullptr;
2599 Node* newval = nullptr;
2600 switch(kind) {
2601 case LS_cmp_swap:
2602 case LS_cmp_swap_weak:
2603 case LS_cmp_exchange: {
2604 const bool two_slot_type = type2size[type] == 2;
2605 receiver = argument(0); // type: oop
2606 base = argument(1); // type: oop
2607 offset = argument(2); // type: long
2608 oldval = argument(4); // type: oop, int, or long
2609 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2610 break;
2611 }
2612 case LS_get_add:
2613 case LS_get_set: {
2614 receiver = argument(0); // type: oop
2615 base = argument(1); // type: oop
2616 offset = argument(2); // type: long
2617 oldval = nullptr;
2618 newval = argument(4); // type: oop, int, or long
2619 break;
2620 }
2621 default:
2622 ShouldNotReachHere();
2623 }
2624
2625 // Build field offset expression.
2626 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2627 // to be plain byte offsets, which are also the same as those accepted
2628 // by oopDesc::field_addr.
2629 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2630 // 32-bit machines ignore the high half of long offsets
2631 offset = ConvL2X(offset);
2632 // Save state and restore on bailout
2633 uint old_sp = sp();
2634 SafePointNode* old_map = clone_map();
2635 Node* adr = make_unsafe_address(base, offset,type, false);
2636 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2637
2638 Compile::AliasType* alias_type = C->alias_type(adr_type);
2639 BasicType bt = alias_type->basic_type();
2640 if (bt != T_ILLEGAL &&
2641 (is_reference_type(bt) != (type == T_OBJECT))) {
2642 // Don't intrinsify mismatched object accesses.
2643 set_map(old_map);
2644 set_sp(old_sp);
2645 return false;
2646 }
2647
2648 destruct_map_clone(old_map);
2649
2650 // For CAS, unlike inline_unsafe_access, there seems no point in
2651 // trying to refine types. Just use the coarse types here.
2652 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2653 const Type *value_type = Type::get_const_basic_type(type);
2654
2655 switch (kind) {
2656 case LS_get_set:
2657 case LS_cmp_exchange: {
2658 if (type == T_OBJECT) {
2659 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2660 if (tjp != nullptr) {
2661 value_type = tjp;
2662 }
2663 }
2664 break;
2665 }
2666 case LS_cmp_swap:
2667 case LS_cmp_swap_weak:
2668 case LS_get_add:
2669 break;
2670 default:
2671 ShouldNotReachHere();
2672 }
2673
2674 // Null check receiver.
2675 receiver = null_check(receiver);
2676 if (stopped()) {
2677 return true;
2678 }
2679
2680 int alias_idx = C->get_alias_index(adr_type);
2681
2682 if (is_reference_type(type)) {
2683 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2684
2685 // Transformation of a value which could be null pointer (CastPP #null)
2686 // could be delayed during Parse (for example, in adjust_map_after_if()).
2687 // Execute transformation here to avoid barrier generation in such case.
2688 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2689 newval = _gvn.makecon(TypePtr::NULL_PTR);
2690
2691 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2692 // Refine the value to a null constant, when it is known to be null
2693 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2694 }
2695 }
2696
2697 Node* result = nullptr;
2698 switch (kind) {
2699 case LS_cmp_exchange: {
2700 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2701 oldval, newval, value_type, type, decorators);
2702 break;
2703 }
2704 case LS_cmp_swap_weak:
2705 decorators |= C2_WEAK_CMPXCHG;
2706 case LS_cmp_swap: {
2707 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2708 oldval, newval, value_type, type, decorators);
2709 break;
2710 }
2711 case LS_get_set: {
2712 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2713 newval, value_type, type, decorators);
2714 break;
2715 }
2716 case LS_get_add: {
2717 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2718 newval, value_type, type, decorators);
2719 break;
2720 }
2721 default:
2722 ShouldNotReachHere();
2723 }
2724
2725 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2726 set_result(result);
2727 return true;
2728 }
2729
2730 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2731 // Regardless of form, don't allow previous ld/st to move down,
2732 // then issue acquire, release, or volatile mem_bar.
2733 insert_mem_bar(Op_MemBarCPUOrder);
2734 switch(id) {
2735 case vmIntrinsics::_loadFence:
2736 insert_mem_bar(Op_LoadFence);
2737 return true;
2738 case vmIntrinsics::_storeFence:
2739 insert_mem_bar(Op_StoreFence);
2740 return true;
2741 case vmIntrinsics::_storeStoreFence:
2742 insert_mem_bar(Op_StoreStoreFence);
2743 return true;
2744 case vmIntrinsics::_fullFence:
2745 insert_mem_bar(Op_MemBarVolatile);
2746 return true;
2747 default:
2748 fatal_unexpected_iid(id);
2749 return false;
2750 }
2751 }
2752
2753 bool LibraryCallKit::inline_onspinwait() {
2754 insert_mem_bar(Op_OnSpinWait);
2755 return true;
2756 }
2757
2758 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2759 if (!kls->is_Con()) {
2760 return true;
2761 }
2762 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2763 if (klsptr == nullptr) {
2764 return true;
2765 }
2766 ciInstanceKlass* ik = klsptr->instance_klass();
2767 // don't need a guard for a klass that is already initialized
2768 return !ik->is_initialized();
2769 }
2770
2771 //----------------------------inline_unsafe_writeback0-------------------------
2772 // public native void Unsafe.writeback0(long address)
2773 bool LibraryCallKit::inline_unsafe_writeback0() {
2774 if (!Matcher::has_match_rule(Op_CacheWB)) {
2775 return false;
2776 }
2777 #ifndef PRODUCT
2778 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2779 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2780 ciSignature* sig = callee()->signature();
2781 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2782 #endif
2783 null_check_receiver(); // null-check, then ignore
2784 Node *addr = argument(1);
2785 addr = new CastX2PNode(addr);
2786 addr = _gvn.transform(addr);
2787 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2788 flush = _gvn.transform(flush);
2789 set_memory(flush, TypeRawPtr::BOTTOM);
2790 return true;
2791 }
2792
2793 //----------------------------inline_unsafe_writeback0-------------------------
2794 // public native void Unsafe.writeback0(long address)
2795 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2796 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2797 return false;
2798 }
2799 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2800 return false;
2801 }
2802 #ifndef PRODUCT
2803 assert(Matcher::has_match_rule(Op_CacheWB),
2804 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2805 : "found match rule for CacheWBPostSync but not CacheWB"));
2806
2807 #endif
2808 null_check_receiver(); // null-check, then ignore
2809 Node *sync;
2810 if (is_pre) {
2811 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2812 } else {
2813 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2814 }
2815 sync = _gvn.transform(sync);
2816 set_memory(sync, TypeRawPtr::BOTTOM);
2817 return true;
2818 }
2819
2820 //----------------------------inline_unsafe_allocate---------------------------
2821 // public native Object Unsafe.allocateInstance(Class<?> cls);
2822 bool LibraryCallKit::inline_unsafe_allocate() {
2823
2824 #if INCLUDE_JVMTI
2825 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2826 return false;
2827 }
2828 #endif //INCLUDE_JVMTI
2829
2830 if (callee()->is_static()) return false; // caller must have the capability!
2831
2832 null_check_receiver(); // null-check, then ignore
2833 Node* cls = null_check(argument(1));
2834 if (stopped()) return true;
2835
2836 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2837 kls = null_check(kls);
2838 if (stopped()) return true; // argument was like int.class
2839
2840 #if INCLUDE_JVMTI
2841 // Don't try to access new allocated obj in the intrinsic.
2842 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2843 // Deoptimize and allocate in interpreter instead.
2844 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2845 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2846 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2847 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2848 {
2849 BuildCutout unless(this, tst, PROB_MAX);
2850 uncommon_trap(Deoptimization::Reason_intrinsic,
2851 Deoptimization::Action_make_not_entrant);
2852 }
2853 if (stopped()) {
2854 return true;
2855 }
2856 #endif //INCLUDE_JVMTI
2857
2858 Node* test = nullptr;
2859 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2860 // Note: The argument might still be an illegal value like
2861 // Serializable.class or Object[].class. The runtime will handle it.
2862 // But we must make an explicit check for initialization.
2863 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2864 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2865 // can generate code to load it as unsigned byte.
2866 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::unordered);
2867 Node* bits = intcon(InstanceKlass::fully_initialized);
2868 test = _gvn.transform(new SubINode(inst, bits));
2869 // The 'test' is non-zero if we need to take a slow path.
2870 }
2871
2872 Node* obj = new_instance(kls, test);
2873 set_result(obj);
2874 return true;
2875 }
2876
2877 //------------------------inline_native_time_funcs--------------
2878 // inline code for System.currentTimeMillis() and System.nanoTime()
2879 // these have the same type and signature
2880 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2881 const TypeFunc* tf = OptoRuntime::void_long_Type();
2882 const TypePtr* no_memory_effects = nullptr;
2883 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2884 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2885 #ifdef ASSERT
2886 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2887 assert(value_top == top(), "second value must be top");
2888 #endif
2889 set_result(value);
2890 return true;
2891 }
2892
2893
2894 #if INCLUDE_JVMTI
2895
2896 // When notifications are disabled then just update the VTMS transition bit and return.
2897 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2898 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2899 if (!DoJVMTIVirtualThreadTransitions) {
2900 return true;
2901 }
2902 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2903 IdealKit ideal(this);
2904
2905 Node* ONE = ideal.ConI(1);
2906 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2907 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2908 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2909
2910 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2911 sync_kit(ideal);
2912 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2913 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2914 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2915 ideal.sync_kit(this);
2916 } ideal.else_(); {
2917 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2918 Node* thread = ideal.thread();
2919 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2920 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2921 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2922
2923 sync_kit(ideal);
2924 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2925 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2926
2927 ideal.sync_kit(this);
2928 } ideal.end_if();
2929 final_sync(ideal);
2930
2931 return true;
2932 }
2933
2934 // Always update the temporary VTMS transition bit.
2935 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2936 if (!DoJVMTIVirtualThreadTransitions) {
2937 return true;
2938 }
2939 IdealKit ideal(this);
2940
2941 {
2942 // unconditionally update the temporary VTMS transition bit in current JavaThread
2943 Node* thread = ideal.thread();
2944 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
2945 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2946 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2947
2948 sync_kit(ideal);
2949 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2950 ideal.sync_kit(this);
2951 }
2952 final_sync(ideal);
2953
2954 return true;
2955 }
2956
2957 // Always update the is_disable_suspend bit.
2958 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
2959 if (!DoJVMTIVirtualThreadTransitions) {
2960 return true;
2961 }
2962 IdealKit ideal(this);
2963
2964 {
2965 // unconditionally update the is_disable_suspend bit in current JavaThread
2966 Node* thread = ideal.thread();
2967 Node* arg = _gvn.transform(argument(0)); // argument for notification
2968 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
2969 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2970
2971 sync_kit(ideal);
2972 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2973 ideal.sync_kit(this);
2974 }
2975 final_sync(ideal);
2976
2977 return true;
2978 }
2979
2980 #endif // INCLUDE_JVMTI
2981
2982 #ifdef JFR_HAVE_INTRINSICS
2983
2984 /**
2985 * if oop->klass != null
2986 * // normal class
2987 * epoch = _epoch_state ? 2 : 1
2988 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
2989 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
2990 * }
2991 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
2992 * else
2993 * // primitive class
2994 * if oop->array_klass != null
2995 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
2996 * else
2997 * id = LAST_TYPE_ID + 1 // void class path
2998 * if (!signaled)
2999 * signaled = true
3000 */
3001 bool LibraryCallKit::inline_native_classID() {
3002 Node* cls = argument(0);
3003
3004 IdealKit ideal(this);
3005 #define __ ideal.
3006 IdealVariable result(ideal); __ declarations_done();
3007 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3008 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3009 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3010
3011
3012 __ if_then(kls, BoolTest::ne, null()); {
3013 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3014 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3015
3016 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3017 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3018 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3019 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3020 mask = _gvn.transform(new OrLNode(mask, epoch));
3021 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3022
3023 float unlikely = PROB_UNLIKELY(0.999);
3024 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3025 sync_kit(ideal);
3026 make_runtime_call(RC_LEAF,
3027 OptoRuntime::class_id_load_barrier_Type(),
3028 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3029 "class id load barrier",
3030 TypePtr::BOTTOM,
3031 kls);
3032 ideal.sync_kit(this);
3033 } __ end_if();
3034
3035 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3036 } __ else_(); {
3037 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3038 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3039 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3040 __ if_then(array_kls, BoolTest::ne, null()); {
3041 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3042 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3043 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3044 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3045 } __ else_(); {
3046 // void class case
3047 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3048 } __ end_if();
3049
3050 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3051 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3052 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3053 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3054 } __ end_if();
3055 } __ end_if();
3056
3057 final_sync(ideal);
3058 set_result(ideal.value(result));
3059 #undef __
3060 return true;
3061 }
3062
3063 //------------------------inline_native_jvm_commit------------------
3064 bool LibraryCallKit::inline_native_jvm_commit() {
3065 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3066
3067 // Save input memory and i_o state.
3068 Node* input_memory_state = reset_memory();
3069 set_all_memory(input_memory_state);
3070 Node* input_io_state = i_o();
3071
3072 // TLS.
3073 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3074 // Jfr java buffer.
3075 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3076 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3077 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3078
3079 // Load the current value of the notified field in the JfrThreadLocal.
3080 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3081 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3082
3083 // Test for notification.
3084 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3085 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3086 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3087
3088 // True branch, is notified.
3089 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3090 set_control(is_notified);
3091
3092 // Reset notified state.
3093 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3094
3095 // Iff notified, the return address of the commit method is the current position of the backing java buffer. This is used to reset the event writer.
3096 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3097 // Convert the machine-word to a long.
3098 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3099
3100 // False branch, not notified.
3101 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3102 set_control(not_notified);
3103 set_all_memory(input_memory_state);
3104
3105 // Arg is the next position as a long.
3106 Node* arg = argument(0);
3107 // Convert long to machine-word.
3108 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3109
3110 // Store the next_position to the underlying jfr java buffer.
3111 Node* commit_memory;
3112 #ifdef _LP64
3113 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3114 #else
3115 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3116 #endif
3117
3118 // Now load the flags from off the java buffer and decide if the buffer is a lease. If so, it needs to be returned post-commit.
3119 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3120 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3121 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3122
3123 // And flags with lease constant.
3124 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3125
3126 // Branch on lease to conditionalize returning the leased java buffer.
3127 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3128 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3129 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3130
3131 // False branch, not a lease.
3132 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3133
3134 // True branch, is lease.
3135 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3136 set_control(is_lease);
3137
3138 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3139 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3140 OptoRuntime::void_void_Type(),
3141 StubRoutines::jfr_return_lease(),
3142 "return_lease", TypePtr::BOTTOM);
3143 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3144
3145 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3146 record_for_igvn(lease_compare_rgn);
3147 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3148 record_for_igvn(lease_compare_mem);
3149 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3150 record_for_igvn(lease_compare_io);
3151 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3152 record_for_igvn(lease_result_value);
3153
3154 // Update control and phi nodes.
3155 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3156 lease_compare_rgn->init_req(_false_path, not_lease);
3157
3158 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3159 lease_compare_mem->init_req(_false_path, commit_memory);
3160
3161 lease_compare_io->init_req(_true_path, i_o());
3162 lease_compare_io->init_req(_false_path, input_io_state);
3163
3164 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3165 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3166
3167 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3168 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3169 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3170 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3171
3172 // Update control and phi nodes.
3173 result_rgn->init_req(_true_path, is_notified);
3174 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3175
3176 result_mem->init_req(_true_path, notified_reset_memory);
3177 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3178
3179 result_io->init_req(_true_path, input_io_state);
3180 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3181
3182 result_value->init_req(_true_path, current_pos);
3183 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3184
3185 // Set output state.
3186 set_control(_gvn.transform(result_rgn));
3187 set_all_memory(_gvn.transform(result_mem));
3188 set_i_o(_gvn.transform(result_io));
3189 set_result(result_rgn, result_value);
3190 return true;
3191 }
3192
3193 /*
3194 * The intrinsic is a model of this pseudo-code:
3195 *
3196 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3197 * jobject h_event_writer = tl->java_event_writer();
3198 * if (h_event_writer == nullptr) {
3199 * return nullptr;
3200 * }
3201 * oop threadObj = Thread::threadObj();
3202 * oop vthread = java_lang_Thread::vthread(threadObj);
3203 * traceid tid;
3204 * bool excluded;
3205 * if (vthread != threadObj) { // i.e. current thread is virtual
3206 * tid = java_lang_Thread::tid(vthread);
3207 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3208 * excluded = vthread_epoch_raw & excluded_mask;
3209 * if (!excluded) {
3210 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3211 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3212 * if (vthread_epoch != current_epoch) {
3213 * write_checkpoint();
3214 * }
3215 * }
3216 * } else {
3217 * tid = java_lang_Thread::tid(threadObj);
3218 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3219 * excluded = thread_epoch_raw & excluded_mask;
3220 * }
3221 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3222 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3223 * if (tid_in_event_writer != tid) {
3224 * setField(event_writer, "threadID", tid);
3225 * setField(event_writer, "excluded", excluded);
3226 * }
3227 * return event_writer
3228 */
3229 bool LibraryCallKit::inline_native_getEventWriter() {
3230 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3231
3232 // Save input memory and i_o state.
3233 Node* input_memory_state = reset_memory();
3234 set_all_memory(input_memory_state);
3235 Node* input_io_state = i_o();
3236
3237 Node* excluded_mask = _gvn.intcon(32768);
3238 Node* epoch_mask = _gvn.intcon(32767);
3239
3240 // TLS
3241 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3242
3243 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3244 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3245
3246 // Load the eventwriter jobject handle.
3247 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3248
3249 // Null check the jobject handle.
3250 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3251 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3252 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3253
3254 // False path, jobj is null.
3255 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3256
3257 // True path, jobj is not null.
3258 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3259
3260 set_control(jobj_is_not_null);
3261
3262 // Load the threadObj for the CarrierThread.
3263 Node* threadObj = generate_current_thread(tls_ptr);
3264
3265 // Load the vthread.
3266 Node* vthread = generate_virtual_thread(tls_ptr);
3267
3268 // If vthread != threadObj, this is a virtual thread.
3269 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3270 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3271 IfNode* iff_vthread_not_equal_threadObj =
3272 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3273
3274 // False branch, fallback to threadObj.
3275 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3276 set_control(vthread_equal_threadObj);
3277
3278 // Load the tid field from the vthread object.
3279 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3280
3281 // Load the raw epoch value from the threadObj.
3282 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3283 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3284 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3285
3286 // Mask off the excluded information from the epoch.
3287 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3288
3289 // True branch, this is a virtual thread.
3290 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3291 set_control(vthread_not_equal_threadObj);
3292
3293 // Load the tid field from the vthread object.
3294 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3295
3296 // Load the raw epoch value from the vthread.
3297 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3298 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3299 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3300
3301 // Mask off the excluded information from the epoch.
3302 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3303
3304 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3305 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3306 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3307 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3308
3309 // False branch, vthread is excluded, no need to write epoch info.
3310 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3311
3312 // True branch, vthread is included, update epoch info.
3313 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3314 set_control(included);
3315
3316 // Get epoch value.
3317 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3318
3319 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3320 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3321 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3322
3323 // Compare the epoch in the vthread to the current epoch generation.
3324 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3325 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3326 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3327
3328 // False path, epoch is equal, checkpoint information is valid.
3329 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3330
3331 // True path, epoch is not equal, write a checkpoint for the vthread.
3332 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3333
3334 set_control(epoch_is_not_equal);
3335
3336 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3337 // The call also updates the native thread local thread id and the vthread with the current epoch.
3338 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3339 OptoRuntime::jfr_write_checkpoint_Type(),
3340 StubRoutines::jfr_write_checkpoint(),
3341 "write_checkpoint", TypePtr::BOTTOM);
3342 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3343
3344 // vthread epoch != current epoch
3345 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3346 record_for_igvn(epoch_compare_rgn);
3347 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3348 record_for_igvn(epoch_compare_mem);
3349 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3350 record_for_igvn(epoch_compare_io);
3351
3352 // Update control and phi nodes.
3353 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3354 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3355 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3356 epoch_compare_mem->init_req(_false_path, input_memory_state);
3357 epoch_compare_io->init_req(_true_path, i_o());
3358 epoch_compare_io->init_req(_false_path, input_io_state);
3359
3360 // excluded != true
3361 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3362 record_for_igvn(exclude_compare_rgn);
3363 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3364 record_for_igvn(exclude_compare_mem);
3365 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3366 record_for_igvn(exclude_compare_io);
3367
3368 // Update control and phi nodes.
3369 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3370 exclude_compare_rgn->init_req(_false_path, excluded);
3371 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3372 exclude_compare_mem->init_req(_false_path, input_memory_state);
3373 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3374 exclude_compare_io->init_req(_false_path, input_io_state);
3375
3376 // vthread != threadObj
3377 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3378 record_for_igvn(vthread_compare_rgn);
3379 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3380 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3381 record_for_igvn(vthread_compare_io);
3382 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3383 record_for_igvn(tid);
3384 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3385 record_for_igvn(exclusion);
3386
3387 // Update control and phi nodes.
3388 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3389 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3390 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3391 vthread_compare_mem->init_req(_false_path, input_memory_state);
3392 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3393 vthread_compare_io->init_req(_false_path, input_io_state);
3394 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3395 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3396 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3397 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3398
3399 // Update branch state.
3400 set_control(_gvn.transform(vthread_compare_rgn));
3401 set_all_memory(_gvn.transform(vthread_compare_mem));
3402 set_i_o(_gvn.transform(vthread_compare_io));
3403
3404 // Load the event writer oop by dereferencing the jobject handle.
3405 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3406 assert(klass_EventWriter->is_loaded(), "invariant");
3407 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3408 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3409 const TypeOopPtr* const xtype = aklass->as_instance_type();
3410 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3411 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3412
3413 // Load the current thread id from the event writer object.
3414 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3415 // Get the field offset to, conditionally, store an updated tid value later.
3416 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3417 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3418 // Get the field offset to, conditionally, store an updated exclusion value later.
3419 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3420 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3421
3422 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3423 record_for_igvn(event_writer_tid_compare_rgn);
3424 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3425 record_for_igvn(event_writer_tid_compare_mem);
3426 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3427 record_for_igvn(event_writer_tid_compare_io);
3428
3429 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3430 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3431 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3432 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3433
3434 // False path, tids are the same.
3435 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3436
3437 // True path, tid is not equal, need to update the tid in the event writer.
3438 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3439 record_for_igvn(tid_is_not_equal);
3440
3441 // Store the exclusion state to the event writer.
3442 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3443
3444 // Store the tid to the event writer.
3445 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3446
3447 // Update control and phi nodes.
3448 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3449 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3450 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3451 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3452 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3453 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3454
3455 // Result of top level CFG, Memory, IO and Value.
3456 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3457 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3458 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3459 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3460
3461 // Result control.
3462 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3463 result_rgn->init_req(_false_path, jobj_is_null);
3464
3465 // Result memory.
3466 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3467 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3468
3469 // Result IO.
3470 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3471 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3472
3473 // Result value.
3474 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3475 result_value->init_req(_false_path, null()); // return null
3476
3477 // Set output state.
3478 set_control(_gvn.transform(result_rgn));
3479 set_all_memory(_gvn.transform(result_mem));
3480 set_i_o(_gvn.transform(result_io));
3481 set_result(result_rgn, result_value);
3482 return true;
3483 }
3484
3485 /*
3486 * The intrinsic is a model of this pseudo-code:
3487 *
3488 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3489 * if (carrierThread != thread) { // is virtual thread
3490 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3491 * bool excluded = vthread_epoch_raw & excluded_mask;
3492 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3493 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3494 * if (!excluded) {
3495 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3496 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3497 * }
3498 * Atomic::release_store(&tl->_vthread, true);
3499 * return;
3500 * }
3501 * Atomic::release_store(&tl->_vthread, false);
3502 */
3503 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3504 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3505
3506 Node* input_memory_state = reset_memory();
3507 set_all_memory(input_memory_state);
3508
3509 Node* excluded_mask = _gvn.intcon(32768);
3510 Node* epoch_mask = _gvn.intcon(32767);
3511
3512 Node* const carrierThread = generate_current_thread(jt);
3513 // If thread != carrierThread, this is a virtual thread.
3514 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3515 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3516 IfNode* iff_thread_not_equal_carrierThread =
3517 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3518
3519 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3520
3521 // False branch, is carrierThread.
3522 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3523 // Store release
3524 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3525
3526 set_all_memory(input_memory_state);
3527
3528 // True branch, is virtual thread.
3529 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3530 set_control(thread_not_equal_carrierThread);
3531
3532 // Load the raw epoch value from the vthread.
3533 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3534 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3535 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3536
3537 // Mask off the excluded information from the epoch.
3538 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3539
3540 // Load the tid field from the thread.
3541 Node* tid = load_field_from_object(thread, "tid", "J");
3542
3543 // Store the vthread tid to the jfr thread local.
3544 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3545 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3546
3547 // Branch is_excluded to conditionalize updating the epoch .
3548 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3549 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3550 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3551
3552 // True branch, vthread is excluded, no need to write epoch info.
3553 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3554 set_control(excluded);
3555 Node* vthread_is_excluded = _gvn.intcon(1);
3556
3557 // False branch, vthread is included, update epoch info.
3558 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3559 set_control(included);
3560 Node* vthread_is_included = _gvn.intcon(0);
3561
3562 // Get epoch value.
3563 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3564
3565 // Store the vthread epoch to the jfr thread local.
3566 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3567 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3568
3569 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3570 record_for_igvn(excluded_rgn);
3571 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3572 record_for_igvn(excluded_mem);
3573 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3574 record_for_igvn(exclusion);
3575
3576 // Merge the excluded control and memory.
3577 excluded_rgn->init_req(_true_path, excluded);
3578 excluded_rgn->init_req(_false_path, included);
3579 excluded_mem->init_req(_true_path, tid_memory);
3580 excluded_mem->init_req(_false_path, included_memory);
3581 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3582 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3583
3584 // Set intermediate state.
3585 set_control(_gvn.transform(excluded_rgn));
3586 set_all_memory(excluded_mem);
3587
3588 // Store the vthread exclusion state to the jfr thread local.
3589 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3590 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3591
3592 // Store release
3593 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3594
3595 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3596 record_for_igvn(thread_compare_rgn);
3597 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3598 record_for_igvn(thread_compare_mem);
3599 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3600 record_for_igvn(vthread);
3601
3602 // Merge the thread_compare control and memory.
3603 thread_compare_rgn->init_req(_true_path, control());
3604 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3605 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3606 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3607
3608 // Set output state.
3609 set_control(_gvn.transform(thread_compare_rgn));
3610 set_all_memory(_gvn.transform(thread_compare_mem));
3611 }
3612
3613 #endif // JFR_HAVE_INTRINSICS
3614
3615 //------------------------inline_native_currentCarrierThread------------------
3616 bool LibraryCallKit::inline_native_currentCarrierThread() {
3617 Node* junk = nullptr;
3618 set_result(generate_current_thread(junk));
3619 return true;
3620 }
3621
3622 //------------------------inline_native_currentThread------------------
3623 bool LibraryCallKit::inline_native_currentThread() {
3624 Node* junk = nullptr;
3625 set_result(generate_virtual_thread(junk));
3626 return true;
3627 }
3628
3629 //------------------------inline_native_setVthread------------------
3630 bool LibraryCallKit::inline_native_setCurrentThread() {
3631 assert(C->method()->changes_current_thread(),
3632 "method changes current Thread but is not annotated ChangesCurrentThread");
3633 Node* arr = argument(1);
3634 Node* thread = _gvn.transform(new ThreadLocalNode());
3635 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3636 Node* thread_obj_handle
3637 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3638 thread_obj_handle = _gvn.transform(thread_obj_handle);
3639 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3640 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3641 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3642 return true;
3643 }
3644
3645 const Type* LibraryCallKit::scopedValueCache_type() {
3646 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3647 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3648 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3649
3650 // Because we create the scopedValue cache lazily we have to make the
3651 // type of the result BotPTR.
3652 bool xk = etype->klass_is_exact();
3653 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3654 return objects_type;
3655 }
3656
3657 Node* LibraryCallKit::scopedValueCache_helper() {
3658 Node* thread = _gvn.transform(new ThreadLocalNode());
3659 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3660 // We cannot use immutable_memory() because we might flip onto a
3661 // different carrier thread, at which point we'll need to use that
3662 // carrier thread's cache.
3663 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3664 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3665 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3666 }
3667
3668 //------------------------inline_native_scopedValueCache------------------
3669 bool LibraryCallKit::inline_native_scopedValueCache() {
3670 Node* cache_obj_handle = scopedValueCache_helper();
3671 const Type* objects_type = scopedValueCache_type();
3672 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3673
3674 return true;
3675 }
3676
3677 //------------------------inline_native_setScopedValueCache------------------
3678 bool LibraryCallKit::inline_native_setScopedValueCache() {
3679 Node* arr = argument(0);
3680 Node* cache_obj_handle = scopedValueCache_helper();
3681 const Type* objects_type = scopedValueCache_type();
3682
3683 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3684 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3685
3686 return true;
3687 }
3688
3689 //---------------------------load_mirror_from_klass----------------------------
3690 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3691 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3692 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3693 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3694 // mirror = ((OopHandle)mirror)->resolve();
3695 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3696 }
3697
3698 //-----------------------load_klass_from_mirror_common-------------------------
3699 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3700 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3701 // and branch to the given path on the region.
3702 // If never_see_null, take an uncommon trap on null, so we can optimistically
3703 // compile for the non-null case.
3704 // If the region is null, force never_see_null = true.
3705 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3706 bool never_see_null,
3707 RegionNode* region,
3708 int null_path,
3709 int offset) {
3710 if (region == nullptr) never_see_null = true;
3711 Node* p = basic_plus_adr(mirror, offset);
3712 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3713 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3714 Node* null_ctl = top();
3715 kls = null_check_oop(kls, &null_ctl, never_see_null);
3716 if (region != nullptr) {
3717 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3718 region->init_req(null_path, null_ctl);
3719 } else {
3720 assert(null_ctl == top(), "no loose ends");
3721 }
3722 return kls;
3723 }
3724
3725 //--------------------(inline_native_Class_query helpers)---------------------
3726 // Use this for JVM_ACC_INTERFACE, JVM_ACC_IS_CLONEABLE_FAST, JVM_ACC_HAS_FINALIZER.
3727 // Fall through if (mods & mask) == bits, take the guard otherwise.
3728 Node* LibraryCallKit::generate_access_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3729 // Branch around if the given klass has the given modifier bit set.
3730 // Like generate_guard, adds a new path onto the region.
3731 Node* modp = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3732 Node* mods = make_load(nullptr, modp, TypeInt::INT, T_INT, MemNode::unordered);
3733 Node* mask = intcon(modifier_mask);
3734 Node* bits = intcon(modifier_bits);
3735 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3736 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3737 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3738 return generate_fair_guard(bol, region);
3739 }
3740 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3741 return generate_access_flags_guard(kls, JVM_ACC_INTERFACE, 0, region);
3742 }
3743 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3744 return generate_access_flags_guard(kls, JVM_ACC_IS_HIDDEN_CLASS, 0, region);
3745 }
3746
3747 //-------------------------inline_native_Class_query-------------------
3748 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3749 const Type* return_type = TypeInt::BOOL;
3750 Node* prim_return_value = top(); // what happens if it's a primitive class?
3751 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3752 bool expect_prim = false; // most of these guys expect to work on refs
3753
3754 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3755
3756 Node* mirror = argument(0);
3757 Node* obj = top();
3758
3759 switch (id) {
3760 case vmIntrinsics::_isInstance:
3761 // nothing is an instance of a primitive type
3762 prim_return_value = intcon(0);
3763 obj = argument(1);
3764 break;
3765 case vmIntrinsics::_getModifiers:
3766 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3767 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3768 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3769 break;
3770 case vmIntrinsics::_isInterface:
3771 prim_return_value = intcon(0);
3772 break;
3773 case vmIntrinsics::_isArray:
3774 prim_return_value = intcon(0);
3775 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3776 break;
3777 case vmIntrinsics::_isPrimitive:
3778 prim_return_value = intcon(1);
3779 expect_prim = true; // obviously
3780 break;
3781 case vmIntrinsics::_isHidden:
3782 prim_return_value = intcon(0);
3783 break;
3784 case vmIntrinsics::_getSuperclass:
3785 prim_return_value = null();
3786 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3787 break;
3788 case vmIntrinsics::_getClassAccessFlags:
3789 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3790 return_type = TypeInt::INT; // not bool! 6297094
3791 break;
3792 default:
3793 fatal_unexpected_iid(id);
3794 break;
3795 }
3796
3797 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3798 if (mirror_con == nullptr) return false; // cannot happen?
3799
3800 #ifndef PRODUCT
3801 if (C->print_intrinsics() || C->print_inlining()) {
3802 ciType* k = mirror_con->java_mirror_type();
3803 if (k) {
3804 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3805 k->print_name();
3806 tty->cr();
3807 }
3808 }
3809 #endif
3810
3811 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3812 RegionNode* region = new RegionNode(PATH_LIMIT);
3813 record_for_igvn(region);
3814 PhiNode* phi = new PhiNode(region, return_type);
3815
3816 // The mirror will never be null of Reflection.getClassAccessFlags, however
3817 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3818 // if it is. See bug 4774291.
3819
3820 // For Reflection.getClassAccessFlags(), the null check occurs in
3821 // the wrong place; see inline_unsafe_access(), above, for a similar
3822 // situation.
3823 mirror = null_check(mirror);
3824 // If mirror or obj is dead, only null-path is taken.
3825 if (stopped()) return true;
3826
3827 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3828
3829 // Now load the mirror's klass metaobject, and null-check it.
3830 // Side-effects region with the control path if the klass is null.
3831 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3832 // If kls is null, we have a primitive mirror.
3833 phi->init_req(_prim_path, prim_return_value);
3834 if (stopped()) { set_result(region, phi); return true; }
3835 bool safe_for_replace = (region->in(_prim_path) == top());
3836
3837 Node* p; // handy temp
3838 Node* null_ctl;
3839
3840 // Now that we have the non-null klass, we can perform the real query.
3841 // For constant classes, the query will constant-fold in LoadNode::Value.
3842 Node* query_value = top();
3843 switch (id) {
3844 case vmIntrinsics::_isInstance:
3845 // nothing is an instance of a primitive type
3846 query_value = gen_instanceof(obj, kls, safe_for_replace);
3847 break;
3848
3849 case vmIntrinsics::_getModifiers:
3850 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
3851 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3852 break;
3853
3854 case vmIntrinsics::_isInterface:
3855 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3856 if (generate_interface_guard(kls, region) != nullptr)
3857 // A guard was added. If the guard is taken, it was an interface.
3858 phi->add_req(intcon(1));
3859 // If we fall through, it's a plain class.
3860 query_value = intcon(0);
3861 break;
3862
3863 case vmIntrinsics::_isArray:
3864 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
3865 if (generate_array_guard(kls, region) != nullptr)
3866 // A guard was added. If the guard is taken, it was an array.
3867 phi->add_req(intcon(1));
3868 // If we fall through, it's a plain class.
3869 query_value = intcon(0);
3870 break;
3871
3872 case vmIntrinsics::_isPrimitive:
3873 query_value = intcon(0); // "normal" path produces false
3874 break;
3875
3876 case vmIntrinsics::_isHidden:
3877 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
3878 if (generate_hidden_class_guard(kls, region) != nullptr)
3879 // A guard was added. If the guard is taken, it was an hidden class.
3880 phi->add_req(intcon(1));
3881 // If we fall through, it's a plain class.
3882 query_value = intcon(0);
3883 break;
3884
3885
3886 case vmIntrinsics::_getSuperclass:
3887 // The rules here are somewhat unfortunate, but we can still do better
3888 // with random logic than with a JNI call.
3889 // Interfaces store null or Object as _super, but must report null.
3890 // Arrays store an intermediate super as _super, but must report Object.
3891 // Other types can report the actual _super.
3892 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
3893 if (generate_interface_guard(kls, region) != nullptr)
3894 // A guard was added. If the guard is taken, it was an interface.
3895 phi->add_req(null());
3896 if (generate_array_guard(kls, region) != nullptr)
3897 // A guard was added. If the guard is taken, it was an array.
3898 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
3899 // If we fall through, it's a plain class. Get its _super.
3900 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
3901 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3902 null_ctl = top();
3903 kls = null_check_oop(kls, &null_ctl);
3904 if (null_ctl != top()) {
3905 // If the guard is taken, Object.superClass is null (both klass and mirror).
3906 region->add_req(null_ctl);
3907 phi ->add_req(null());
3908 }
3909 if (!stopped()) {
3910 query_value = load_mirror_from_klass(kls);
3911 }
3912 break;
3913
3914 case vmIntrinsics::_getClassAccessFlags:
3915 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
3916 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
3917 break;
3918
3919 default:
3920 fatal_unexpected_iid(id);
3921 break;
3922 }
3923
3924 // Fall-through is the normal case of a query to a real class.
3925 phi->init_req(1, query_value);
3926 region->init_req(1, control());
3927
3928 C->set_has_split_ifs(true); // Has chance for split-if optimization
3929 set_result(region, phi);
3930 return true;
3931 }
3932
3933 //-------------------------inline_Class_cast-------------------
3934 bool LibraryCallKit::inline_Class_cast() {
3935 Node* mirror = argument(0); // Class
3936 Node* obj = argument(1);
3937 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3938 if (mirror_con == nullptr) {
3939 return false; // dead path (mirror->is_top()).
3940 }
3941 if (obj == nullptr || obj->is_top()) {
3942 return false; // dead path
3943 }
3944 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
3945
3946 // First, see if Class.cast() can be folded statically.
3947 // java_mirror_type() returns non-null for compile-time Class constants.
3948 ciType* tm = mirror_con->java_mirror_type();
3949 if (tm != nullptr && tm->is_klass() &&
3950 tp != nullptr) {
3951 if (!tp->is_loaded()) {
3952 // Don't use intrinsic when class is not loaded.
3953 return false;
3954 } else {
3955 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
3956 if (static_res == Compile::SSC_always_true) {
3957 // isInstance() is true - fold the code.
3958 set_result(obj);
3959 return true;
3960 } else if (static_res == Compile::SSC_always_false) {
3961 // Don't use intrinsic, have to throw ClassCastException.
3962 // If the reference is null, the non-intrinsic bytecode will
3963 // be optimized appropriately.
3964 return false;
3965 }
3966 }
3967 }
3968
3969 // Bailout intrinsic and do normal inlining if exception path is frequent.
3970 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3971 return false;
3972 }
3973
3974 // Generate dynamic checks.
3975 // Class.cast() is java implementation of _checkcast bytecode.
3976 // Do checkcast (Parse::do_checkcast()) optimizations here.
3977
3978 mirror = null_check(mirror);
3979 // If mirror is dead, only null-path is taken.
3980 if (stopped()) {
3981 return true;
3982 }
3983
3984 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
3985 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
3986 RegionNode* region = new RegionNode(PATH_LIMIT);
3987 record_for_igvn(region);
3988
3989 // Now load the mirror's klass metaobject, and null-check it.
3990 // If kls is null, we have a primitive mirror and
3991 // nothing is an instance of a primitive type.
3992 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
3993
3994 Node* res = top();
3995 if (!stopped()) {
3996 Node* bad_type_ctrl = top();
3997 // Do checkcast optimizations.
3998 res = gen_checkcast(obj, kls, &bad_type_ctrl);
3999 region->init_req(_bad_type_path, bad_type_ctrl);
4000 }
4001 if (region->in(_prim_path) != top() ||
4002 region->in(_bad_type_path) != top()) {
4003 // Let Interpreter throw ClassCastException.
4004 PreserveJVMState pjvms(this);
4005 set_control(_gvn.transform(region));
4006 uncommon_trap(Deoptimization::Reason_intrinsic,
4007 Deoptimization::Action_maybe_recompile);
4008 }
4009 if (!stopped()) {
4010 set_result(res);
4011 }
4012 return true;
4013 }
4014
4015
4016 //--------------------------inline_native_subtype_check------------------------
4017 // This intrinsic takes the JNI calls out of the heart of
4018 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4019 bool LibraryCallKit::inline_native_subtype_check() {
4020 // Pull both arguments off the stack.
4021 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4022 args[0] = argument(0);
4023 args[1] = argument(1);
4024 Node* klasses[2]; // corresponding Klasses: superk, subk
4025 klasses[0] = klasses[1] = top();
4026
4027 enum {
4028 // A full decision tree on {superc is prim, subc is prim}:
4029 _prim_0_path = 1, // {P,N} => false
4030 // {P,P} & superc!=subc => false
4031 _prim_same_path, // {P,P} & superc==subc => true
4032 _prim_1_path, // {N,P} => false
4033 _ref_subtype_path, // {N,N} & subtype check wins => true
4034 _both_ref_path, // {N,N} & subtype check loses => false
4035 PATH_LIMIT
4036 };
4037
4038 RegionNode* region = new RegionNode(PATH_LIMIT);
4039 Node* phi = new PhiNode(region, TypeInt::BOOL);
4040 record_for_igvn(region);
4041
4042 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4043 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4044 int class_klass_offset = java_lang_Class::klass_offset();
4045
4046 // First null-check both mirrors and load each mirror's klass metaobject.
4047 int which_arg;
4048 for (which_arg = 0; which_arg <= 1; which_arg++) {
4049 Node* arg = args[which_arg];
4050 arg = null_check(arg);
4051 if (stopped()) break;
4052 args[which_arg] = arg;
4053
4054 Node* p = basic_plus_adr(arg, class_klass_offset);
4055 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4056 klasses[which_arg] = _gvn.transform(kls);
4057 }
4058
4059 // Having loaded both klasses, test each for null.
4060 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4061 for (which_arg = 0; which_arg <= 1; which_arg++) {
4062 Node* kls = klasses[which_arg];
4063 Node* null_ctl = top();
4064 kls = null_check_oop(kls, &null_ctl, never_see_null);
4065 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4066 region->init_req(prim_path, null_ctl);
4067 if (stopped()) break;
4068 klasses[which_arg] = kls;
4069 }
4070
4071 if (!stopped()) {
4072 // now we have two reference types, in klasses[0..1]
4073 Node* subk = klasses[1]; // the argument to isAssignableFrom
4074 Node* superk = klasses[0]; // the receiver
4075 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4076 // now we have a successful reference subtype check
4077 region->set_req(_ref_subtype_path, control());
4078 }
4079
4080 // If both operands are primitive (both klasses null), then
4081 // we must return true when they are identical primitives.
4082 // It is convenient to test this after the first null klass check.
4083 set_control(region->in(_prim_0_path)); // go back to first null check
4084 if (!stopped()) {
4085 // Since superc is primitive, make a guard for the superc==subc case.
4086 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4087 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4088 generate_guard(bol_eq, region, PROB_FAIR);
4089 if (region->req() == PATH_LIMIT+1) {
4090 // A guard was added. If the added guard is taken, superc==subc.
4091 region->swap_edges(PATH_LIMIT, _prim_same_path);
4092 region->del_req(PATH_LIMIT);
4093 }
4094 region->set_req(_prim_0_path, control()); // Not equal after all.
4095 }
4096
4097 // these are the only paths that produce 'true':
4098 phi->set_req(_prim_same_path, intcon(1));
4099 phi->set_req(_ref_subtype_path, intcon(1));
4100
4101 // pull together the cases:
4102 assert(region->req() == PATH_LIMIT, "sane region");
4103 for (uint i = 1; i < region->req(); i++) {
4104 Node* ctl = region->in(i);
4105 if (ctl == nullptr || ctl == top()) {
4106 region->set_req(i, top());
4107 phi ->set_req(i, top());
4108 } else if (phi->in(i) == nullptr) {
4109 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4110 }
4111 }
4112
4113 set_control(_gvn.transform(region));
4114 set_result(_gvn.transform(phi));
4115 return true;
4116 }
4117
4118 //---------------------generate_array_guard_common------------------------
4119 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4120 bool obj_array, bool not_array) {
4121
4122 if (stopped()) {
4123 return nullptr;
4124 }
4125
4126 // If obj_array/non_array==false/false:
4127 // Branch around if the given klass is in fact an array (either obj or prim).
4128 // If obj_array/non_array==false/true:
4129 // Branch around if the given klass is not an array klass of any kind.
4130 // If obj_array/non_array==true/true:
4131 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4132 // If obj_array/non_array==true/false:
4133 // Branch around if the kls is an oop array (Object[] or subtype)
4134 //
4135 // Like generate_guard, adds a new path onto the region.
4136 jint layout_con = 0;
4137 Node* layout_val = get_layout_helper(kls, layout_con);
4138 if (layout_val == nullptr) {
4139 bool query = (obj_array
4140 ? Klass::layout_helper_is_objArray(layout_con)
4141 : Klass::layout_helper_is_array(layout_con));
4142 if (query == not_array) {
4143 return nullptr; // never a branch
4144 } else { // always a branch
4145 Node* always_branch = control();
4146 if (region != nullptr)
4147 region->add_req(always_branch);
4148 set_control(top());
4149 return always_branch;
4150 }
4151 }
4152 // Now test the correct condition.
4153 jint nval = (obj_array
4154 ? (jint)(Klass::_lh_array_tag_type_value
4155 << Klass::_lh_array_tag_shift)
4156 : Klass::_lh_neutral_value);
4157 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4158 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4159 // invert the test if we are looking for a non-array
4160 if (not_array) btest = BoolTest(btest).negate();
4161 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4162 return generate_fair_guard(bol, region);
4163 }
4164
4165
4166 //-----------------------inline_native_newArray--------------------------
4167 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4168 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4169 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4170 Node* mirror;
4171 Node* count_val;
4172 if (uninitialized) {
4173 null_check_receiver();
4174 mirror = argument(1);
4175 count_val = argument(2);
4176 } else {
4177 mirror = argument(0);
4178 count_val = argument(1);
4179 }
4180
4181 mirror = null_check(mirror);
4182 // If mirror or obj is dead, only null-path is taken.
4183 if (stopped()) return true;
4184
4185 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4186 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4187 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4188 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4189 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4190
4191 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4192 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4193 result_reg, _slow_path);
4194 Node* normal_ctl = control();
4195 Node* no_array_ctl = result_reg->in(_slow_path);
4196
4197 // Generate code for the slow case. We make a call to newArray().
4198 set_control(no_array_ctl);
4199 if (!stopped()) {
4200 // Either the input type is void.class, or else the
4201 // array klass has not yet been cached. Either the
4202 // ensuing call will throw an exception, or else it
4203 // will cache the array klass for next time.
4204 PreserveJVMState pjvms(this);
4205 CallJavaNode* slow_call = nullptr;
4206 if (uninitialized) {
4207 // Generate optimized virtual call (holder class 'Unsafe' is final)
4208 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4209 } else {
4210 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4211 }
4212 Node* slow_result = set_results_for_java_call(slow_call);
4213 // this->control() comes from set_results_for_java_call
4214 result_reg->set_req(_slow_path, control());
4215 result_val->set_req(_slow_path, slow_result);
4216 result_io ->set_req(_slow_path, i_o());
4217 result_mem->set_req(_slow_path, reset_memory());
4218 }
4219
4220 set_control(normal_ctl);
4221 if (!stopped()) {
4222 // Normal case: The array type has been cached in the java.lang.Class.
4223 // The following call works fine even if the array type is polymorphic.
4224 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4225 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4226 result_reg->init_req(_normal_path, control());
4227 result_val->init_req(_normal_path, obj);
4228 result_io ->init_req(_normal_path, i_o());
4229 result_mem->init_req(_normal_path, reset_memory());
4230
4231 if (uninitialized) {
4232 // Mark the allocation so that zeroing is skipped
4233 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4234 alloc->maybe_set_complete(&_gvn);
4235 }
4236 }
4237
4238 // Return the combined state.
4239 set_i_o( _gvn.transform(result_io) );
4240 set_all_memory( _gvn.transform(result_mem));
4241
4242 C->set_has_split_ifs(true); // Has chance for split-if optimization
4243 set_result(result_reg, result_val);
4244 return true;
4245 }
4246
4247 //----------------------inline_native_getLength--------------------------
4248 // public static native int java.lang.reflect.Array.getLength(Object array);
4249 bool LibraryCallKit::inline_native_getLength() {
4250 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4251
4252 Node* array = null_check(argument(0));
4253 // If array is dead, only null-path is taken.
4254 if (stopped()) return true;
4255
4256 // Deoptimize if it is a non-array.
4257 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4258
4259 if (non_array != nullptr) {
4260 PreserveJVMState pjvms(this);
4261 set_control(non_array);
4262 uncommon_trap(Deoptimization::Reason_intrinsic,
4263 Deoptimization::Action_maybe_recompile);
4264 }
4265
4266 // If control is dead, only non-array-path is taken.
4267 if (stopped()) return true;
4268
4269 // The works fine even if the array type is polymorphic.
4270 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4271 Node* result = load_array_length(array);
4272
4273 C->set_has_split_ifs(true); // Has chance for split-if optimization
4274 set_result(result);
4275 return true;
4276 }
4277
4278 //------------------------inline_array_copyOf----------------------------
4279 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4280 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4281 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4282 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4283
4284 // Get the arguments.
4285 Node* original = argument(0);
4286 Node* start = is_copyOfRange? argument(1): intcon(0);
4287 Node* end = is_copyOfRange? argument(2): argument(1);
4288 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4289
4290 Node* newcopy = nullptr;
4291
4292 // Set the original stack and the reexecute bit for the interpreter to reexecute
4293 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4294 { PreserveReexecuteState preexecs(this);
4295 jvms()->set_should_reexecute(true);
4296
4297 array_type_mirror = null_check(array_type_mirror);
4298 original = null_check(original);
4299
4300 // Check if a null path was taken unconditionally.
4301 if (stopped()) return true;
4302
4303 Node* orig_length = load_array_length(original);
4304
4305 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4306 klass_node = null_check(klass_node);
4307
4308 RegionNode* bailout = new RegionNode(1);
4309 record_for_igvn(bailout);
4310
4311 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4312 // Bail out if that is so.
4313 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4314 if (not_objArray != nullptr) {
4315 // Improve the klass node's type from the new optimistic assumption:
4316 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4317 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4318 Node* cast = new CastPPNode(control(), klass_node, akls);
4319 klass_node = _gvn.transform(cast);
4320 }
4321
4322 // Bail out if either start or end is negative.
4323 generate_negative_guard(start, bailout, &start);
4324 generate_negative_guard(end, bailout, &end);
4325
4326 Node* length = end;
4327 if (_gvn.type(start) != TypeInt::ZERO) {
4328 length = _gvn.transform(new SubINode(end, start));
4329 }
4330
4331 // Bail out if length is negative (i.e., if start > end).
4332 // Without this the new_array would throw
4333 // NegativeArraySizeException but IllegalArgumentException is what
4334 // should be thrown
4335 generate_negative_guard(length, bailout, &length);
4336
4337 // Bail out if start is larger than the original length
4338 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4339 generate_negative_guard(orig_tail, bailout, &orig_tail);
4340
4341 if (bailout->req() > 1) {
4342 PreserveJVMState pjvms(this);
4343 set_control(_gvn.transform(bailout));
4344 uncommon_trap(Deoptimization::Reason_intrinsic,
4345 Deoptimization::Action_maybe_recompile);
4346 }
4347
4348 if (!stopped()) {
4349 // How many elements will we copy from the original?
4350 // The answer is MinI(orig_tail, length).
4351 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4352
4353 // Generate a direct call to the right arraycopy function(s).
4354 // We know the copy is disjoint but we might not know if the
4355 // oop stores need checking.
4356 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4357 // This will fail a store-check if x contains any non-nulls.
4358
4359 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4360 // loads/stores but it is legal only if we're sure the
4361 // Arrays.copyOf would succeed. So we need all input arguments
4362 // to the copyOf to be validated, including that the copy to the
4363 // new array won't trigger an ArrayStoreException. That subtype
4364 // check can be optimized if we know something on the type of
4365 // the input array from type speculation.
4366 if (_gvn.type(klass_node)->singleton()) {
4367 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4368 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4369
4370 int test = C->static_subtype_check(superk, subk);
4371 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4372 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4373 if (t_original->speculative_type() != nullptr) {
4374 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4375 }
4376 }
4377 }
4378
4379 bool validated = false;
4380 // Reason_class_check rather than Reason_intrinsic because we
4381 // want to intrinsify even if this traps.
4382 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4383 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4384
4385 if (not_subtype_ctrl != top()) {
4386 PreserveJVMState pjvms(this);
4387 set_control(not_subtype_ctrl);
4388 uncommon_trap(Deoptimization::Reason_class_check,
4389 Deoptimization::Action_make_not_entrant);
4390 assert(stopped(), "Should be stopped");
4391 }
4392 validated = true;
4393 }
4394
4395 if (!stopped()) {
4396 newcopy = new_array(klass_node, length, 0); // no arguments to push
4397
4398 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4399 load_object_klass(original), klass_node);
4400 if (!is_copyOfRange) {
4401 ac->set_copyof(validated);
4402 } else {
4403 ac->set_copyofrange(validated);
4404 }
4405 Node* n = _gvn.transform(ac);
4406 if (n == ac) {
4407 ac->connect_outputs(this);
4408 } else {
4409 assert(validated, "shouldn't transform if all arguments not validated");
4410 set_all_memory(n);
4411 }
4412 }
4413 }
4414 } // original reexecute is set back here
4415
4416 C->set_has_split_ifs(true); // Has chance for split-if optimization
4417 if (!stopped()) {
4418 set_result(newcopy);
4419 }
4420 return true;
4421 }
4422
4423
4424 //----------------------generate_virtual_guard---------------------------
4425 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4426 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4427 RegionNode* slow_region) {
4428 ciMethod* method = callee();
4429 int vtable_index = method->vtable_index();
4430 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4431 "bad index %d", vtable_index);
4432 // Get the Method* out of the appropriate vtable entry.
4433 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4434 vtable_index*vtableEntry::size_in_bytes() +
4435 in_bytes(vtableEntry::method_offset());
4436 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4437 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4438
4439 // Compare the target method with the expected method (e.g., Object.hashCode).
4440 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4441
4442 Node* native_call = makecon(native_call_addr);
4443 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4444 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4445
4446 return generate_slow_guard(test_native, slow_region);
4447 }
4448
4449 //-----------------------generate_method_call----------------------------
4450 // Use generate_method_call to make a slow-call to the real
4451 // method if the fast path fails. An alternative would be to
4452 // use a stub like OptoRuntime::slow_arraycopy_Java.
4453 // This only works for expanding the current library call,
4454 // not another intrinsic. (E.g., don't use this for making an
4455 // arraycopy call inside of the copyOf intrinsic.)
4456 CallJavaNode*
4457 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4458 // When compiling the intrinsic method itself, do not use this technique.
4459 guarantee(callee() != C->method(), "cannot make slow-call to self");
4460
4461 ciMethod* method = callee();
4462 // ensure the JVMS we have will be correct for this call
4463 guarantee(method_id == method->intrinsic_id(), "must match");
4464
4465 const TypeFunc* tf = TypeFunc::make(method);
4466 if (res_not_null) {
4467 assert(tf->return_type() == T_OBJECT, "");
4468 const TypeTuple* range = tf->range();
4469 const Type** fields = TypeTuple::fields(range->cnt());
4470 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4471 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4472 tf = TypeFunc::make(tf->domain(), new_range);
4473 }
4474 CallJavaNode* slow_call;
4475 if (is_static) {
4476 assert(!is_virtual, "");
4477 slow_call = new CallStaticJavaNode(C, tf,
4478 SharedRuntime::get_resolve_static_call_stub(), method);
4479 } else if (is_virtual) {
4480 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4481 int vtable_index = Method::invalid_vtable_index;
4482 if (UseInlineCaches) {
4483 // Suppress the vtable call
4484 } else {
4485 // hashCode and clone are not a miranda methods,
4486 // so the vtable index is fixed.
4487 // No need to use the linkResolver to get it.
4488 vtable_index = method->vtable_index();
4489 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4490 "bad index %d", vtable_index);
4491 }
4492 slow_call = new CallDynamicJavaNode(tf,
4493 SharedRuntime::get_resolve_virtual_call_stub(),
4494 method, vtable_index);
4495 } else { // neither virtual nor static: opt_virtual
4496 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4497 slow_call = new CallStaticJavaNode(C, tf,
4498 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4499 slow_call->set_optimized_virtual(true);
4500 }
4501 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4502 // To be able to issue a direct call (optimized virtual or virtual)
4503 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4504 // about the method being invoked should be attached to the call site to
4505 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4506 slow_call->set_override_symbolic_info(true);
4507 }
4508 set_arguments_for_java_call(slow_call);
4509 set_edges_for_java_call(slow_call);
4510 return slow_call;
4511 }
4512
4513
4514 /**
4515 * Build special case code for calls to hashCode on an object. This call may
4516 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4517 * slightly different code.
4518 */
4519 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4520 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4521 assert(!(is_virtual && is_static), "either virtual, special, or static");
4522
4523 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4524
4525 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4526 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4527 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4528 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4529 Node* obj = nullptr;
4530 if (!is_static) {
4531 // Check for hashing null object
4532 obj = null_check_receiver();
4533 if (stopped()) return true; // unconditionally null
4534 result_reg->init_req(_null_path, top());
4535 result_val->init_req(_null_path, top());
4536 } else {
4537 // Do a null check, and return zero if null.
4538 // System.identityHashCode(null) == 0
4539 obj = argument(0);
4540 Node* null_ctl = top();
4541 obj = null_check_oop(obj, &null_ctl);
4542 result_reg->init_req(_null_path, null_ctl);
4543 result_val->init_req(_null_path, _gvn.intcon(0));
4544 }
4545
4546 // Unconditionally null? Then return right away.
4547 if (stopped()) {
4548 set_control( result_reg->in(_null_path));
4549 if (!stopped())
4550 set_result(result_val->in(_null_path));
4551 return true;
4552 }
4553
4554 // We only go to the fast case code if we pass a number of guards. The
4555 // paths which do not pass are accumulated in the slow_region.
4556 RegionNode* slow_region = new RegionNode(1);
4557 record_for_igvn(slow_region);
4558
4559 // If this is a virtual call, we generate a funny guard. We pull out
4560 // the vtable entry corresponding to hashCode() from the target object.
4561 // If the target method which we are calling happens to be the native
4562 // Object hashCode() method, we pass the guard. We do not need this
4563 // guard for non-virtual calls -- the caller is known to be the native
4564 // Object hashCode().
4565 if (is_virtual) {
4566 // After null check, get the object's klass.
4567 Node* obj_klass = load_object_klass(obj);
4568 generate_virtual_guard(obj_klass, slow_region);
4569 }
4570
4571 // Get the header out of the object, use LoadMarkNode when available
4572 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4573 // The control of the load must be null. Otherwise, the load can move before
4574 // the null check after castPP removal.
4575 Node* no_ctrl = nullptr;
4576 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4577
4578 // Test the header to see if it is safe to read w.r.t. locking.
4579 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4580 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4581 if (LockingMode == LM_LIGHTWEIGHT) {
4582 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4583 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4584 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4585
4586 generate_slow_guard(test_monitor, slow_region);
4587 } else {
4588 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4589 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4590 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4591
4592 generate_slow_guard(test_not_unlocked, slow_region);
4593 }
4594
4595 // Get the hash value and check to see that it has been properly assigned.
4596 // We depend on hash_mask being at most 32 bits and avoid the use of
4597 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4598 // vm: see markWord.hpp.
4599 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4600 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4601 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4602 // This hack lets the hash bits live anywhere in the mark object now, as long
4603 // as the shift drops the relevant bits into the low 32 bits. Note that
4604 // Java spec says that HashCode is an int so there's no point in capturing
4605 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4606 hshifted_header = ConvX2I(hshifted_header);
4607 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4608
4609 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4610 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4611 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4612
4613 generate_slow_guard(test_assigned, slow_region);
4614
4615 Node* init_mem = reset_memory();
4616 // fill in the rest of the null path:
4617 result_io ->init_req(_null_path, i_o());
4618 result_mem->init_req(_null_path, init_mem);
4619
4620 result_val->init_req(_fast_path, hash_val);
4621 result_reg->init_req(_fast_path, control());
4622 result_io ->init_req(_fast_path, i_o());
4623 result_mem->init_req(_fast_path, init_mem);
4624
4625 // Generate code for the slow case. We make a call to hashCode().
4626 set_control(_gvn.transform(slow_region));
4627 if (!stopped()) {
4628 // No need for PreserveJVMState, because we're using up the present state.
4629 set_all_memory(init_mem);
4630 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4631 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4632 Node* slow_result = set_results_for_java_call(slow_call);
4633 // this->control() comes from set_results_for_java_call
4634 result_reg->init_req(_slow_path, control());
4635 result_val->init_req(_slow_path, slow_result);
4636 result_io ->set_req(_slow_path, i_o());
4637 result_mem ->set_req(_slow_path, reset_memory());
4638 }
4639
4640 // Return the combined state.
4641 set_i_o( _gvn.transform(result_io) );
4642 set_all_memory( _gvn.transform(result_mem));
4643
4644 set_result(result_reg, result_val);
4645 return true;
4646 }
4647
4648 //---------------------------inline_native_getClass----------------------------
4649 // public final native Class<?> java.lang.Object.getClass();
4650 //
4651 // Build special case code for calls to getClass on an object.
4652 bool LibraryCallKit::inline_native_getClass() {
4653 Node* obj = null_check_receiver();
4654 if (stopped()) return true;
4655 set_result(load_mirror_from_klass(load_object_klass(obj)));
4656 return true;
4657 }
4658
4659 //-----------------inline_native_Reflection_getCallerClass---------------------
4660 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4661 //
4662 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4663 //
4664 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4665 // in that it must skip particular security frames and checks for
4666 // caller sensitive methods.
4667 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4668 #ifndef PRODUCT
4669 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4670 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4671 }
4672 #endif
4673
4674 if (!jvms()->has_method()) {
4675 #ifndef PRODUCT
4676 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4677 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4678 }
4679 #endif
4680 return false;
4681 }
4682
4683 // Walk back up the JVM state to find the caller at the required
4684 // depth.
4685 JVMState* caller_jvms = jvms();
4686
4687 // Cf. JVM_GetCallerClass
4688 // NOTE: Start the loop at depth 1 because the current JVM state does
4689 // not include the Reflection.getCallerClass() frame.
4690 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4691 ciMethod* m = caller_jvms->method();
4692 switch (n) {
4693 case 0:
4694 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4695 break;
4696 case 1:
4697 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4698 if (!m->caller_sensitive()) {
4699 #ifndef PRODUCT
4700 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4701 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4702 }
4703 #endif
4704 return false; // bail-out; let JVM_GetCallerClass do the work
4705 }
4706 break;
4707 default:
4708 if (!m->is_ignored_by_security_stack_walk()) {
4709 // We have reached the desired frame; return the holder class.
4710 // Acquire method holder as java.lang.Class and push as constant.
4711 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4712 ciInstance* caller_mirror = caller_klass->java_mirror();
4713 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4714
4715 #ifndef PRODUCT
4716 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4717 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());
4718 tty->print_cr(" JVM state at this point:");
4719 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4720 ciMethod* m = jvms()->of_depth(i)->method();
4721 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4722 }
4723 }
4724 #endif
4725 return true;
4726 }
4727 break;
4728 }
4729 }
4730
4731 #ifndef PRODUCT
4732 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4733 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4734 tty->print_cr(" JVM state at this point:");
4735 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4736 ciMethod* m = jvms()->of_depth(i)->method();
4737 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4738 }
4739 }
4740 #endif
4741
4742 return false; // bail-out; let JVM_GetCallerClass do the work
4743 }
4744
4745 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4746 Node* arg = argument(0);
4747 Node* result = nullptr;
4748
4749 switch (id) {
4750 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4751 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4752 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4753 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4754 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4755 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4756
4757 case vmIntrinsics::_doubleToLongBits: {
4758 // two paths (plus control) merge in a wood
4759 RegionNode *r = new RegionNode(3);
4760 Node *phi = new PhiNode(r, TypeLong::LONG);
4761
4762 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4763 // Build the boolean node
4764 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4765
4766 // Branch either way.
4767 // NaN case is less traveled, which makes all the difference.
4768 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4769 Node *opt_isnan = _gvn.transform(ifisnan);
4770 assert( opt_isnan->is_If(), "Expect an IfNode");
4771 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4772 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4773
4774 set_control(iftrue);
4775
4776 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4777 Node *slow_result = longcon(nan_bits); // return NaN
4778 phi->init_req(1, _gvn.transform( slow_result ));
4779 r->init_req(1, iftrue);
4780
4781 // Else fall through
4782 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4783 set_control(iffalse);
4784
4785 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4786 r->init_req(2, iffalse);
4787
4788 // Post merge
4789 set_control(_gvn.transform(r));
4790 record_for_igvn(r);
4791
4792 C->set_has_split_ifs(true); // Has chance for split-if optimization
4793 result = phi;
4794 assert(result->bottom_type()->isa_long(), "must be");
4795 break;
4796 }
4797
4798 case vmIntrinsics::_floatToIntBits: {
4799 // two paths (plus control) merge in a wood
4800 RegionNode *r = new RegionNode(3);
4801 Node *phi = new PhiNode(r, TypeInt::INT);
4802
4803 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4804 // Build the boolean node
4805 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4806
4807 // Branch either way.
4808 // NaN case is less traveled, which makes all the difference.
4809 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4810 Node *opt_isnan = _gvn.transform(ifisnan);
4811 assert( opt_isnan->is_If(), "Expect an IfNode");
4812 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4813 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4814
4815 set_control(iftrue);
4816
4817 static const jint nan_bits = 0x7fc00000;
4818 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4819 phi->init_req(1, _gvn.transform( slow_result ));
4820 r->init_req(1, iftrue);
4821
4822 // Else fall through
4823 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4824 set_control(iffalse);
4825
4826 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4827 r->init_req(2, iffalse);
4828
4829 // Post merge
4830 set_control(_gvn.transform(r));
4831 record_for_igvn(r);
4832
4833 C->set_has_split_ifs(true); // Has chance for split-if optimization
4834 result = phi;
4835 assert(result->bottom_type()->isa_int(), "must be");
4836 break;
4837 }
4838
4839 default:
4840 fatal_unexpected_iid(id);
4841 break;
4842 }
4843 set_result(_gvn.transform(result));
4844 return true;
4845 }
4846
4847 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
4848 Node* arg = argument(0);
4849 Node* result = nullptr;
4850
4851 switch (id) {
4852 case vmIntrinsics::_floatIsInfinite:
4853 result = new IsInfiniteFNode(arg);
4854 break;
4855 case vmIntrinsics::_floatIsFinite:
4856 result = new IsFiniteFNode(arg);
4857 break;
4858 case vmIntrinsics::_doubleIsInfinite:
4859 result = new IsInfiniteDNode(arg);
4860 break;
4861 case vmIntrinsics::_doubleIsFinite:
4862 result = new IsFiniteDNode(arg);
4863 break;
4864 default:
4865 fatal_unexpected_iid(id);
4866 break;
4867 }
4868 set_result(_gvn.transform(result));
4869 return true;
4870 }
4871
4872 //----------------------inline_unsafe_copyMemory-------------------------
4873 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
4874
4875 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
4876 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
4877 const Type* base_t = gvn.type(base);
4878
4879 bool in_native = (base_t == TypePtr::NULL_PTR);
4880 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
4881 bool is_mixed = !in_heap && !in_native;
4882
4883 if (is_mixed) {
4884 return true; // mixed accesses can touch both on-heap and off-heap memory
4885 }
4886 if (in_heap) {
4887 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
4888 if (!is_prim_array) {
4889 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
4890 // there's not enough type information available to determine proper memory slice for it.
4891 return true;
4892 }
4893 }
4894 return false;
4895 }
4896
4897 bool LibraryCallKit::inline_unsafe_copyMemory() {
4898 if (callee()->is_static()) return false; // caller must have the capability!
4899 null_check_receiver(); // null-check receiver
4900 if (stopped()) return true;
4901
4902 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4903
4904 Node* src_base = argument(1); // type: oop
4905 Node* src_off = ConvL2X(argument(2)); // type: long
4906 Node* dst_base = argument(4); // type: oop
4907 Node* dst_off = ConvL2X(argument(5)); // type: long
4908 Node* size = ConvL2X(argument(7)); // type: long
4909
4910 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4911 "fieldOffset must be byte-scaled");
4912
4913 Node* src_addr = make_unsafe_address(src_base, src_off);
4914 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4915
4916 Node* thread = _gvn.transform(new ThreadLocalNode());
4917 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4918 BasicType doing_unsafe_access_bt = T_BYTE;
4919 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4920
4921 // update volatile field
4922 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4923
4924 int flags = RC_LEAF | RC_NO_FP;
4925
4926 const TypePtr* dst_type = TypePtr::BOTTOM;
4927
4928 // Adjust memory effects of the runtime call based on input values.
4929 if (!has_wide_mem(_gvn, src_addr, src_base) &&
4930 !has_wide_mem(_gvn, dst_addr, dst_base)) {
4931 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4932
4933 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
4934 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
4935 flags |= RC_NARROW_MEM; // narrow in memory
4936 }
4937 }
4938
4939 // Call it. Note that the length argument is not scaled.
4940 make_runtime_call(flags,
4941 OptoRuntime::fast_arraycopy_Type(),
4942 StubRoutines::unsafe_arraycopy(),
4943 "unsafe_arraycopy",
4944 dst_type,
4945 src_addr, dst_addr, size XTOP);
4946
4947 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4948
4949 return true;
4950 }
4951
4952 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
4953 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
4954 bool LibraryCallKit::inline_unsafe_setMemory() {
4955 if (callee()->is_static()) return false; // caller must have the capability!
4956 null_check_receiver(); // null-check receiver
4957 if (stopped()) return true;
4958
4959 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
4960
4961 Node* dst_base = argument(1); // type: oop
4962 Node* dst_off = ConvL2X(argument(2)); // type: long
4963 Node* size = ConvL2X(argument(4)); // type: long
4964 Node* byte = argument(6); // type: byte
4965
4966 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
4967 "fieldOffset must be byte-scaled");
4968
4969 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
4970
4971 Node* thread = _gvn.transform(new ThreadLocalNode());
4972 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
4973 BasicType doing_unsafe_access_bt = T_BYTE;
4974 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
4975
4976 // update volatile field
4977 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4978
4979 int flags = RC_LEAF | RC_NO_FP;
4980
4981 const TypePtr* dst_type = TypePtr::BOTTOM;
4982
4983 // Adjust memory effects of the runtime call based on input values.
4984 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
4985 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
4986
4987 flags |= RC_NARROW_MEM; // narrow in memory
4988 }
4989
4990 // Call it. Note that the length argument is not scaled.
4991 make_runtime_call(flags,
4992 OptoRuntime::make_setmemory_Type(),
4993 StubRoutines::unsafe_setmemory(),
4994 "unsafe_setmemory",
4995 dst_type,
4996 dst_addr, size XTOP, byte);
4997
4998 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
4999
5000 return true;
5001 }
5002
5003 #undef XTOP
5004
5005 //------------------------clone_coping-----------------------------------
5006 // Helper function for inline_native_clone.
5007 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5008 assert(obj_size != nullptr, "");
5009 Node* raw_obj = alloc_obj->in(1);
5010 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5011
5012 AllocateNode* alloc = nullptr;
5013 if (ReduceBulkZeroing) {
5014 // We will be completely responsible for initializing this object -
5015 // mark Initialize node as complete.
5016 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5017 // The object was just allocated - there should be no any stores!
5018 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5019 // Mark as complete_with_arraycopy so that on AllocateNode
5020 // expansion, we know this AllocateNode is initialized by an array
5021 // copy and a StoreStore barrier exists after the array copy.
5022 alloc->initialization()->set_complete_with_arraycopy();
5023 }
5024
5025 Node* size = _gvn.transform(obj_size);
5026 access_clone(obj, alloc_obj, size, is_array);
5027
5028 // Do not let reads from the cloned object float above the arraycopy.
5029 if (alloc != nullptr) {
5030 // Do not let stores that initialize this object be reordered with
5031 // a subsequent store that would make this object accessible by
5032 // other threads.
5033 // Record what AllocateNode this StoreStore protects so that
5034 // escape analysis can go from the MemBarStoreStoreNode to the
5035 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5036 // based on the escape status of the AllocateNode.
5037 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5038 } else {
5039 insert_mem_bar(Op_MemBarCPUOrder);
5040 }
5041 }
5042
5043 //------------------------inline_native_clone----------------------------
5044 // protected native Object java.lang.Object.clone();
5045 //
5046 // Here are the simple edge cases:
5047 // null receiver => normal trap
5048 // virtual and clone was overridden => slow path to out-of-line clone
5049 // not cloneable or finalizer => slow path to out-of-line Object.clone
5050 //
5051 // The general case has two steps, allocation and copying.
5052 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5053 //
5054 // Copying also has two cases, oop arrays and everything else.
5055 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5056 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5057 //
5058 // These steps fold up nicely if and when the cloned object's klass
5059 // can be sharply typed as an object array, a type array, or an instance.
5060 //
5061 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5062 PhiNode* result_val;
5063
5064 // Set the reexecute bit for the interpreter to reexecute
5065 // the bytecode that invokes Object.clone if deoptimization happens.
5066 { PreserveReexecuteState preexecs(this);
5067 jvms()->set_should_reexecute(true);
5068
5069 Node* obj = null_check_receiver();
5070 if (stopped()) return true;
5071
5072 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5073
5074 // If we are going to clone an instance, we need its exact type to
5075 // know the number and types of fields to convert the clone to
5076 // loads/stores. Maybe a speculative type can help us.
5077 if (!obj_type->klass_is_exact() &&
5078 obj_type->speculative_type() != nullptr &&
5079 obj_type->speculative_type()->is_instance_klass()) {
5080 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5081 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5082 !spec_ik->has_injected_fields()) {
5083 if (!obj_type->isa_instptr() ||
5084 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5085 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5086 }
5087 }
5088 }
5089
5090 // Conservatively insert a memory barrier on all memory slices.
5091 // Do not let writes into the original float below the clone.
5092 insert_mem_bar(Op_MemBarCPUOrder);
5093
5094 // paths into result_reg:
5095 enum {
5096 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5097 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5098 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5099 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5100 PATH_LIMIT
5101 };
5102 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5103 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5104 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5105 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5106 record_for_igvn(result_reg);
5107
5108 Node* obj_klass = load_object_klass(obj);
5109 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5110 if (array_ctl != nullptr) {
5111 // It's an array.
5112 PreserveJVMState pjvms(this);
5113 set_control(array_ctl);
5114 Node* obj_length = load_array_length(obj);
5115 Node* array_size = nullptr; // Size of the array without object alignment padding.
5116 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5117
5118 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5119 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5120 // If it is an oop array, it requires very special treatment,
5121 // because gc barriers are required when accessing the array.
5122 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5123 if (is_obja != nullptr) {
5124 PreserveJVMState pjvms2(this);
5125 set_control(is_obja);
5126 // Generate a direct call to the right arraycopy function(s).
5127 // Clones are always tightly coupled.
5128 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5129 ac->set_clone_oop_array();
5130 Node* n = _gvn.transform(ac);
5131 assert(n == ac, "cannot disappear");
5132 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5133
5134 result_reg->init_req(_objArray_path, control());
5135 result_val->init_req(_objArray_path, alloc_obj);
5136 result_i_o ->set_req(_objArray_path, i_o());
5137 result_mem ->set_req(_objArray_path, reset_memory());
5138 }
5139 }
5140 // Otherwise, there are no barriers to worry about.
5141 // (We can dispense with card marks if we know the allocation
5142 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5143 // causes the non-eden paths to take compensating steps to
5144 // simulate a fresh allocation, so that no further
5145 // card marks are required in compiled code to initialize
5146 // the object.)
5147
5148 if (!stopped()) {
5149 copy_to_clone(obj, alloc_obj, array_size, true);
5150
5151 // Present the results of the copy.
5152 result_reg->init_req(_array_path, control());
5153 result_val->init_req(_array_path, alloc_obj);
5154 result_i_o ->set_req(_array_path, i_o());
5155 result_mem ->set_req(_array_path, reset_memory());
5156 }
5157 }
5158
5159 // We only go to the instance fast case code if we pass a number of guards.
5160 // The paths which do not pass are accumulated in the slow_region.
5161 RegionNode* slow_region = new RegionNode(1);
5162 record_for_igvn(slow_region);
5163 if (!stopped()) {
5164 // It's an instance (we did array above). Make the slow-path tests.
5165 // If this is a virtual call, we generate a funny guard. We grab
5166 // the vtable entry corresponding to clone() from the target object.
5167 // If the target method which we are calling happens to be the
5168 // Object clone() method, we pass the guard. We do not need this
5169 // guard for non-virtual calls; the caller is known to be the native
5170 // Object clone().
5171 if (is_virtual) {
5172 generate_virtual_guard(obj_klass, slow_region);
5173 }
5174
5175 // The object must be easily cloneable and must not have a finalizer.
5176 // Both of these conditions may be checked in a single test.
5177 // We could optimize the test further, but we don't care.
5178 generate_access_flags_guard(obj_klass,
5179 // Test both conditions:
5180 JVM_ACC_IS_CLONEABLE_FAST | JVM_ACC_HAS_FINALIZER,
5181 // Must be cloneable but not finalizer:
5182 JVM_ACC_IS_CLONEABLE_FAST,
5183 slow_region);
5184 }
5185
5186 if (!stopped()) {
5187 // It's an instance, and it passed the slow-path tests.
5188 PreserveJVMState pjvms(this);
5189 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5190 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5191 // is reexecuted if deoptimization occurs and there could be problems when merging
5192 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5193 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5194
5195 copy_to_clone(obj, alloc_obj, obj_size, false);
5196
5197 // Present the results of the slow call.
5198 result_reg->init_req(_instance_path, control());
5199 result_val->init_req(_instance_path, alloc_obj);
5200 result_i_o ->set_req(_instance_path, i_o());
5201 result_mem ->set_req(_instance_path, reset_memory());
5202 }
5203
5204 // Generate code for the slow case. We make a call to clone().
5205 set_control(_gvn.transform(slow_region));
5206 if (!stopped()) {
5207 PreserveJVMState pjvms(this);
5208 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5209 // We need to deoptimize on exception (see comment above)
5210 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5211 // this->control() comes from set_results_for_java_call
5212 result_reg->init_req(_slow_path, control());
5213 result_val->init_req(_slow_path, slow_result);
5214 result_i_o ->set_req(_slow_path, i_o());
5215 result_mem ->set_req(_slow_path, reset_memory());
5216 }
5217
5218 // Return the combined state.
5219 set_control( _gvn.transform(result_reg));
5220 set_i_o( _gvn.transform(result_i_o));
5221 set_all_memory( _gvn.transform(result_mem));
5222 } // original reexecute is set back here
5223
5224 set_result(_gvn.transform(result_val));
5225 return true;
5226 }
5227
5228 // If we have a tightly coupled allocation, the arraycopy may take care
5229 // of the array initialization. If one of the guards we insert between
5230 // the allocation and the arraycopy causes a deoptimization, an
5231 // uninitialized array will escape the compiled method. To prevent that
5232 // we set the JVM state for uncommon traps between the allocation and
5233 // the arraycopy to the state before the allocation so, in case of
5234 // deoptimization, we'll reexecute the allocation and the
5235 // initialization.
5236 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5237 if (alloc != nullptr) {
5238 ciMethod* trap_method = alloc->jvms()->method();
5239 int trap_bci = alloc->jvms()->bci();
5240
5241 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5242 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5243 // Make sure there's no store between the allocation and the
5244 // arraycopy otherwise visible side effects could be rexecuted
5245 // in case of deoptimization and cause incorrect execution.
5246 bool no_interfering_store = true;
5247 Node* mem = alloc->in(TypeFunc::Memory);
5248 if (mem->is_MergeMem()) {
5249 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5250 Node* n = mms.memory();
5251 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5252 assert(n->is_Store(), "what else?");
5253 no_interfering_store = false;
5254 break;
5255 }
5256 }
5257 } else {
5258 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5259 Node* n = mms.memory();
5260 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5261 assert(n->is_Store(), "what else?");
5262 no_interfering_store = false;
5263 break;
5264 }
5265 }
5266 }
5267
5268 if (no_interfering_store) {
5269 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5270
5271 JVMState* saved_jvms = jvms();
5272 saved_reexecute_sp = _reexecute_sp;
5273
5274 set_jvms(sfpt->jvms());
5275 _reexecute_sp = jvms()->sp();
5276
5277 return saved_jvms;
5278 }
5279 }
5280 }
5281 return nullptr;
5282 }
5283
5284 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5285 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5286 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5287 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5288 uint size = alloc->req();
5289 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5290 old_jvms->set_map(sfpt);
5291 for (uint i = 0; i < size; i++) {
5292 sfpt->init_req(i, alloc->in(i));
5293 }
5294 // re-push array length for deoptimization
5295 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5296 old_jvms->set_sp(old_jvms->sp()+1);
5297 old_jvms->set_monoff(old_jvms->monoff()+1);
5298 old_jvms->set_scloff(old_jvms->scloff()+1);
5299 old_jvms->set_endoff(old_jvms->endoff()+1);
5300 old_jvms->set_should_reexecute(true);
5301
5302 sfpt->set_i_o(map()->i_o());
5303 sfpt->set_memory(map()->memory());
5304 sfpt->set_control(map()->control());
5305 return sfpt;
5306 }
5307
5308 // In case of a deoptimization, we restart execution at the
5309 // allocation, allocating a new array. We would leave an uninitialized
5310 // array in the heap that GCs wouldn't expect. Move the allocation
5311 // after the traps so we don't allocate the array if we
5312 // deoptimize. This is possible because tightly_coupled_allocation()
5313 // guarantees there's no observer of the allocated array at this point
5314 // and the control flow is simple enough.
5315 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5316 int saved_reexecute_sp, uint new_idx) {
5317 if (saved_jvms_before_guards != nullptr && !stopped()) {
5318 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5319
5320 assert(alloc != nullptr, "only with a tightly coupled allocation");
5321 // restore JVM state to the state at the arraycopy
5322 saved_jvms_before_guards->map()->set_control(map()->control());
5323 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5324 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5325 // If we've improved the types of some nodes (null check) while
5326 // emitting the guards, propagate them to the current state
5327 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5328 set_jvms(saved_jvms_before_guards);
5329 _reexecute_sp = saved_reexecute_sp;
5330
5331 // Remove the allocation from above the guards
5332 CallProjections callprojs;
5333 alloc->extract_projections(&callprojs, true);
5334 InitializeNode* init = alloc->initialization();
5335 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5336 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5337 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5338
5339 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5340 // the allocation (i.e. is only valid if the allocation succeeds):
5341 // 1) replace CastIINode with AllocateArrayNode's length here
5342 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5343 //
5344 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5345 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5346 Node* init_control = init->proj_out(TypeFunc::Control);
5347 Node* alloc_length = alloc->Ideal_length();
5348 #ifdef ASSERT
5349 Node* prev_cast = nullptr;
5350 #endif
5351 for (uint i = 0; i < init_control->outcnt(); i++) {
5352 Node* init_out = init_control->raw_out(i);
5353 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5354 #ifdef ASSERT
5355 if (prev_cast == nullptr) {
5356 prev_cast = init_out;
5357 } else {
5358 if (prev_cast->cmp(*init_out) == false) {
5359 prev_cast->dump();
5360 init_out->dump();
5361 assert(false, "not equal CastIINode");
5362 }
5363 }
5364 #endif
5365 C->gvn_replace_by(init_out, alloc_length);
5366 }
5367 }
5368 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5369
5370 // move the allocation here (after the guards)
5371 _gvn.hash_delete(alloc);
5372 alloc->set_req(TypeFunc::Control, control());
5373 alloc->set_req(TypeFunc::I_O, i_o());
5374 Node *mem = reset_memory();
5375 set_all_memory(mem);
5376 alloc->set_req(TypeFunc::Memory, mem);
5377 set_control(init->proj_out_or_null(TypeFunc::Control));
5378 set_i_o(callprojs.fallthrough_ioproj);
5379
5380 // Update memory as done in GraphKit::set_output_for_allocation()
5381 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5382 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5383 if (ary_type->isa_aryptr() && length_type != nullptr) {
5384 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5385 }
5386 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5387 int elemidx = C->get_alias_index(telemref);
5388 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5389 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5390
5391 Node* allocx = _gvn.transform(alloc);
5392 assert(allocx == alloc, "where has the allocation gone?");
5393 assert(dest->is_CheckCastPP(), "not an allocation result?");
5394
5395 _gvn.hash_delete(dest);
5396 dest->set_req(0, control());
5397 Node* destx = _gvn.transform(dest);
5398 assert(destx == dest, "where has the allocation result gone?");
5399
5400 array_ideal_length(alloc, ary_type, true);
5401 }
5402 }
5403
5404 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5405 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5406 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5407 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5408 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5409 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5410 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5411 JVMState* saved_jvms_before_guards) {
5412 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5413 // There is at least one unrelated uncommon trap which needs to be replaced.
5414 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5415
5416 JVMState* saved_jvms = jvms();
5417 const int saved_reexecute_sp = _reexecute_sp;
5418 set_jvms(sfpt->jvms());
5419 _reexecute_sp = jvms()->sp();
5420
5421 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5422
5423 // Restore state
5424 set_jvms(saved_jvms);
5425 _reexecute_sp = saved_reexecute_sp;
5426 }
5427 }
5428
5429 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5430 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5431 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5432 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5433 while (if_proj->is_IfProj()) {
5434 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5435 if (uncommon_trap != nullptr) {
5436 create_new_uncommon_trap(uncommon_trap);
5437 }
5438 assert(if_proj->in(0)->is_If(), "must be If");
5439 if_proj = if_proj->in(0)->in(0);
5440 }
5441 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5442 "must have reached control projection of init node");
5443 }
5444
5445 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5446 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5447 assert(trap_request != 0, "no valid UCT trap request");
5448 PreserveJVMState pjvms(this);
5449 set_control(uncommon_trap_call->in(0));
5450 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5451 Deoptimization::trap_request_action(trap_request));
5452 assert(stopped(), "Should be stopped");
5453 _gvn.hash_delete(uncommon_trap_call);
5454 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5455 }
5456
5457 //------------------------------inline_array_partition-----------------------
5458 bool LibraryCallKit::inline_array_partition() {
5459
5460 Node* elementType = null_check(argument(0));
5461 Node* obj = argument(1);
5462 Node* offset = argument(2);
5463 Node* fromIndex = argument(4);
5464 Node* toIndex = argument(5);
5465 Node* indexPivot1 = argument(6);
5466 Node* indexPivot2 = argument(7);
5467
5468 Node* pivotIndices = nullptr;
5469
5470 // Set the original stack and the reexecute bit for the interpreter to reexecute
5471 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5472 { PreserveReexecuteState preexecs(this);
5473 jvms()->set_should_reexecute(true);
5474
5475 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5476 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5477 BasicType bt = elem_type->basic_type();
5478 // Disable the intrinsic if the CPU does not support SIMD sort
5479 if (!Matcher::supports_simd_sort(bt)) {
5480 return false;
5481 }
5482 address stubAddr = nullptr;
5483 stubAddr = StubRoutines::select_array_partition_function();
5484 // stub not loaded
5485 if (stubAddr == nullptr) {
5486 return false;
5487 }
5488 // get the address of the array
5489 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5490 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5491 return false; // failed input validation
5492 }
5493 Node* obj_adr = make_unsafe_address(obj, offset);
5494
5495 // create the pivotIndices array of type int and size = 2
5496 Node* size = intcon(2);
5497 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5498 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5499 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5500 guarantee(alloc != nullptr, "created above");
5501 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5502
5503 // pass the basic type enum to the stub
5504 Node* elemType = intcon(bt);
5505
5506 // Call the stub
5507 const char *stubName = "array_partition_stub";
5508 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5509 stubAddr, stubName, TypePtr::BOTTOM,
5510 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5511 indexPivot1, indexPivot2);
5512
5513 } // original reexecute is set back here
5514
5515 if (!stopped()) {
5516 set_result(pivotIndices);
5517 }
5518
5519 return true;
5520 }
5521
5522
5523 //------------------------------inline_array_sort-----------------------
5524 bool LibraryCallKit::inline_array_sort() {
5525
5526 Node* elementType = null_check(argument(0));
5527 Node* obj = argument(1);
5528 Node* offset = argument(2);
5529 Node* fromIndex = argument(4);
5530 Node* toIndex = argument(5);
5531
5532 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5533 ciType* elem_type = elem_klass->const_oop()->as_instance()->java_mirror_type();
5534 BasicType bt = elem_type->basic_type();
5535 // Disable the intrinsic if the CPU does not support SIMD sort
5536 if (!Matcher::supports_simd_sort(bt)) {
5537 return false;
5538 }
5539 address stubAddr = nullptr;
5540 stubAddr = StubRoutines::select_arraysort_function();
5541 //stub not loaded
5542 if (stubAddr == nullptr) {
5543 return false;
5544 }
5545
5546 // get address of the array
5547 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5548 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM ) {
5549 return false; // failed input validation
5550 }
5551 Node* obj_adr = make_unsafe_address(obj, offset);
5552
5553 // pass the basic type enum to the stub
5554 Node* elemType = intcon(bt);
5555
5556 // Call the stub.
5557 const char *stubName = "arraysort_stub";
5558 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5559 stubAddr, stubName, TypePtr::BOTTOM,
5560 obj_adr, elemType, fromIndex, toIndex);
5561
5562 return true;
5563 }
5564
5565
5566 //------------------------------inline_arraycopy-----------------------
5567 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5568 // Object dest, int destPos,
5569 // int length);
5570 bool LibraryCallKit::inline_arraycopy() {
5571 // Get the arguments.
5572 Node* src = argument(0); // type: oop
5573 Node* src_offset = argument(1); // type: int
5574 Node* dest = argument(2); // type: oop
5575 Node* dest_offset = argument(3); // type: int
5576 Node* length = argument(4); // type: int
5577
5578 uint new_idx = C->unique();
5579
5580 // Check for allocation before we add nodes that would confuse
5581 // tightly_coupled_allocation()
5582 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5583
5584 int saved_reexecute_sp = -1;
5585 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5586 // See arraycopy_restore_alloc_state() comment
5587 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5588 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5589 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5590 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5591
5592 // The following tests must be performed
5593 // (1) src and dest are arrays.
5594 // (2) src and dest arrays must have elements of the same BasicType
5595 // (3) src and dest must not be null.
5596 // (4) src_offset must not be negative.
5597 // (5) dest_offset must not be negative.
5598 // (6) length must not be negative.
5599 // (7) src_offset + length must not exceed length of src.
5600 // (8) dest_offset + length must not exceed length of dest.
5601 // (9) each element of an oop array must be assignable
5602
5603 // (3) src and dest must not be null.
5604 // always do this here because we need the JVM state for uncommon traps
5605 Node* null_ctl = top();
5606 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5607 assert(null_ctl->is_top(), "no null control here");
5608 dest = null_check(dest, T_ARRAY);
5609
5610 if (!can_emit_guards) {
5611 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5612 // guards but the arraycopy node could still take advantage of a
5613 // tightly allocated allocation. tightly_coupled_allocation() is
5614 // called again to make sure it takes the null check above into
5615 // account: the null check is mandatory and if it caused an
5616 // uncommon trap to be emitted then the allocation can't be
5617 // considered tightly coupled in this context.
5618 alloc = tightly_coupled_allocation(dest);
5619 }
5620
5621 bool validated = false;
5622
5623 const Type* src_type = _gvn.type(src);
5624 const Type* dest_type = _gvn.type(dest);
5625 const TypeAryPtr* top_src = src_type->isa_aryptr();
5626 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5627
5628 // Do we have the type of src?
5629 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5630 // Do we have the type of dest?
5631 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5632 // Is the type for src from speculation?
5633 bool src_spec = false;
5634 // Is the type for dest from speculation?
5635 bool dest_spec = false;
5636
5637 if ((!has_src || !has_dest) && can_emit_guards) {
5638 // We don't have sufficient type information, let's see if
5639 // speculative types can help. We need to have types for both src
5640 // and dest so that it pays off.
5641
5642 // Do we already have or could we have type information for src
5643 bool could_have_src = has_src;
5644 // Do we already have or could we have type information for dest
5645 bool could_have_dest = has_dest;
5646
5647 ciKlass* src_k = nullptr;
5648 if (!has_src) {
5649 src_k = src_type->speculative_type_not_null();
5650 if (src_k != nullptr && src_k->is_array_klass()) {
5651 could_have_src = true;
5652 }
5653 }
5654
5655 ciKlass* dest_k = nullptr;
5656 if (!has_dest) {
5657 dest_k = dest_type->speculative_type_not_null();
5658 if (dest_k != nullptr && dest_k->is_array_klass()) {
5659 could_have_dest = true;
5660 }
5661 }
5662
5663 if (could_have_src && could_have_dest) {
5664 // This is going to pay off so emit the required guards
5665 if (!has_src) {
5666 src = maybe_cast_profiled_obj(src, src_k, true);
5667 src_type = _gvn.type(src);
5668 top_src = src_type->isa_aryptr();
5669 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5670 src_spec = true;
5671 }
5672 if (!has_dest) {
5673 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5674 dest_type = _gvn.type(dest);
5675 top_dest = dest_type->isa_aryptr();
5676 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5677 dest_spec = true;
5678 }
5679 }
5680 }
5681
5682 if (has_src && has_dest && can_emit_guards) {
5683 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5684 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5685 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5686 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5687
5688 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5689 // If both arrays are object arrays then having the exact types
5690 // for both will remove the need for a subtype check at runtime
5691 // before the call and may make it possible to pick a faster copy
5692 // routine (without a subtype check on every element)
5693 // Do we have the exact type of src?
5694 bool could_have_src = src_spec;
5695 // Do we have the exact type of dest?
5696 bool could_have_dest = dest_spec;
5697 ciKlass* src_k = nullptr;
5698 ciKlass* dest_k = nullptr;
5699 if (!src_spec) {
5700 src_k = src_type->speculative_type_not_null();
5701 if (src_k != nullptr && src_k->is_array_klass()) {
5702 could_have_src = true;
5703 }
5704 }
5705 if (!dest_spec) {
5706 dest_k = dest_type->speculative_type_not_null();
5707 if (dest_k != nullptr && dest_k->is_array_klass()) {
5708 could_have_dest = true;
5709 }
5710 }
5711 if (could_have_src && could_have_dest) {
5712 // If we can have both exact types, emit the missing guards
5713 if (could_have_src && !src_spec) {
5714 src = maybe_cast_profiled_obj(src, src_k, true);
5715 }
5716 if (could_have_dest && !dest_spec) {
5717 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5718 }
5719 }
5720 }
5721 }
5722
5723 ciMethod* trap_method = method();
5724 int trap_bci = bci();
5725 if (saved_jvms_before_guards != nullptr) {
5726 trap_method = alloc->jvms()->method();
5727 trap_bci = alloc->jvms()->bci();
5728 }
5729
5730 bool negative_length_guard_generated = false;
5731
5732 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5733 can_emit_guards &&
5734 !src->is_top() && !dest->is_top()) {
5735 // validate arguments: enables transformation the ArrayCopyNode
5736 validated = true;
5737
5738 RegionNode* slow_region = new RegionNode(1);
5739 record_for_igvn(slow_region);
5740
5741 // (1) src and dest are arrays.
5742 generate_non_array_guard(load_object_klass(src), slow_region);
5743 generate_non_array_guard(load_object_klass(dest), slow_region);
5744
5745 // (2) src and dest arrays must have elements of the same BasicType
5746 // done at macro expansion or at Ideal transformation time
5747
5748 // (4) src_offset must not be negative.
5749 generate_negative_guard(src_offset, slow_region);
5750
5751 // (5) dest_offset must not be negative.
5752 generate_negative_guard(dest_offset, slow_region);
5753
5754 // (7) src_offset + length must not exceed length of src.
5755 generate_limit_guard(src_offset, length,
5756 load_array_length(src),
5757 slow_region);
5758
5759 // (8) dest_offset + length must not exceed length of dest.
5760 generate_limit_guard(dest_offset, length,
5761 load_array_length(dest),
5762 slow_region);
5763
5764 // (6) length must not be negative.
5765 // This is also checked in generate_arraycopy() during macro expansion, but
5766 // we also have to check it here for the case where the ArrayCopyNode will
5767 // be eliminated by Escape Analysis.
5768 if (EliminateAllocations) {
5769 generate_negative_guard(length, slow_region);
5770 negative_length_guard_generated = true;
5771 }
5772
5773 // (9) each element of an oop array must be assignable
5774 Node* dest_klass = load_object_klass(dest);
5775 if (src != dest) {
5776 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5777
5778 if (not_subtype_ctrl != top()) {
5779 PreserveJVMState pjvms(this);
5780 set_control(not_subtype_ctrl);
5781 uncommon_trap(Deoptimization::Reason_intrinsic,
5782 Deoptimization::Action_make_not_entrant);
5783 assert(stopped(), "Should be stopped");
5784 }
5785 }
5786 {
5787 PreserveJVMState pjvms(this);
5788 set_control(_gvn.transform(slow_region));
5789 uncommon_trap(Deoptimization::Reason_intrinsic,
5790 Deoptimization::Action_make_not_entrant);
5791 assert(stopped(), "Should be stopped");
5792 }
5793
5794 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5795 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5796 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5797 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5798 }
5799
5800 if (stopped()) {
5801 return true;
5802 }
5803
5804 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5805 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5806 // so the compiler has a chance to eliminate them: during macro expansion,
5807 // we have to set their control (CastPP nodes are eliminated).
5808 load_object_klass(src), load_object_klass(dest),
5809 load_array_length(src), load_array_length(dest));
5810
5811 ac->set_arraycopy(validated);
5812
5813 Node* n = _gvn.transform(ac);
5814 if (n == ac) {
5815 ac->connect_outputs(this);
5816 } else {
5817 assert(validated, "shouldn't transform if all arguments not validated");
5818 set_all_memory(n);
5819 }
5820 clear_upper_avx();
5821
5822
5823 return true;
5824 }
5825
5826
5827 // Helper function which determines if an arraycopy immediately follows
5828 // an allocation, with no intervening tests or other escapes for the object.
5829 AllocateArrayNode*
5830 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
5831 if (stopped()) return nullptr; // no fast path
5832 if (!C->do_aliasing()) return nullptr; // no MergeMems around
5833
5834 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
5835 if (alloc == nullptr) return nullptr;
5836
5837 Node* rawmem = memory(Compile::AliasIdxRaw);
5838 // Is the allocation's memory state untouched?
5839 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
5840 // Bail out if there have been raw-memory effects since the allocation.
5841 // (Example: There might have been a call or safepoint.)
5842 return nullptr;
5843 }
5844 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
5845 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
5846 return nullptr;
5847 }
5848
5849 // There must be no unexpected observers of this allocation.
5850 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
5851 Node* obs = ptr->fast_out(i);
5852 if (obs != this->map()) {
5853 return nullptr;
5854 }
5855 }
5856
5857 // This arraycopy must unconditionally follow the allocation of the ptr.
5858 Node* alloc_ctl = ptr->in(0);
5859 Node* ctl = control();
5860 while (ctl != alloc_ctl) {
5861 // There may be guards which feed into the slow_region.
5862 // Any other control flow means that we might not get a chance
5863 // to finish initializing the allocated object.
5864 // Various low-level checks bottom out in uncommon traps. These
5865 // are considered safe since we've already checked above that
5866 // there is no unexpected observer of this allocation.
5867 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
5868 assert(ctl->in(0)->is_If(), "must be If");
5869 ctl = ctl->in(0)->in(0);
5870 } else {
5871 return nullptr;
5872 }
5873 }
5874
5875 // If we get this far, we have an allocation which immediately
5876 // precedes the arraycopy, and we can take over zeroing the new object.
5877 // The arraycopy will finish the initialization, and provide
5878 // a new control state to which we will anchor the destination pointer.
5879
5880 return alloc;
5881 }
5882
5883 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
5884 if (node->is_IfProj()) {
5885 Node* other_proj = node->as_IfProj()->other_if_proj();
5886 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
5887 Node* obs = other_proj->fast_out(j);
5888 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
5889 (obs->as_CallStaticJava()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point())) {
5890 return obs->as_CallStaticJava();
5891 }
5892 }
5893 }
5894 return nullptr;
5895 }
5896
5897 //-------------inline_encodeISOArray-----------------------------------
5898 // encode char[] to byte[] in ISO_8859_1 or ASCII
5899 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
5900 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
5901 // no receiver since it is static method
5902 Node *src = argument(0);
5903 Node *src_offset = argument(1);
5904 Node *dst = argument(2);
5905 Node *dst_offset = argument(3);
5906 Node *length = argument(4);
5907
5908 src = must_be_not_null(src, true);
5909 dst = must_be_not_null(dst, true);
5910
5911 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
5912 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
5913 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
5914 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
5915 // failed array check
5916 return false;
5917 }
5918
5919 // Figure out the size and type of the elements we will be copying.
5920 BasicType src_elem = src_type->elem()->array_element_basic_type();
5921 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
5922 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
5923 return false;
5924 }
5925
5926 Node* src_start = array_element_address(src, src_offset, T_CHAR);
5927 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
5928 // 'src_start' points to src array + scaled offset
5929 // 'dst_start' points to dst array + scaled offset
5930
5931 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
5932 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
5933 enc = _gvn.transform(enc);
5934 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
5935 set_memory(res_mem, mtype);
5936 set_result(enc);
5937 clear_upper_avx();
5938
5939 return true;
5940 }
5941
5942 //-------------inline_multiplyToLen-----------------------------------
5943 bool LibraryCallKit::inline_multiplyToLen() {
5944 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
5945
5946 address stubAddr = StubRoutines::multiplyToLen();
5947 if (stubAddr == nullptr) {
5948 return false; // Intrinsic's stub is not implemented on this platform
5949 }
5950 const char* stubName = "multiplyToLen";
5951
5952 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
5953
5954 // no receiver because it is a static method
5955 Node* x = argument(0);
5956 Node* xlen = argument(1);
5957 Node* y = argument(2);
5958 Node* ylen = argument(3);
5959 Node* z = argument(4);
5960
5961 x = must_be_not_null(x, true);
5962 y = must_be_not_null(y, true);
5963
5964 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
5965 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
5966 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
5967 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
5968 // failed array check
5969 return false;
5970 }
5971
5972 BasicType x_elem = x_type->elem()->array_element_basic_type();
5973 BasicType y_elem = y_type->elem()->array_element_basic_type();
5974 if (x_elem != T_INT || y_elem != T_INT) {
5975 return false;
5976 }
5977
5978 // Set the original stack and the reexecute bit for the interpreter to reexecute
5979 // the bytecode that invokes BigInteger.multiplyToLen() if deoptimization happens
5980 // on the return from z array allocation in runtime.
5981 { PreserveReexecuteState preexecs(this);
5982 jvms()->set_should_reexecute(true);
5983
5984 Node* x_start = array_element_address(x, intcon(0), x_elem);
5985 Node* y_start = array_element_address(y, intcon(0), y_elem);
5986 // 'x_start' points to x array + scaled xlen
5987 // 'y_start' points to y array + scaled ylen
5988
5989 // Allocate the result array
5990 Node* zlen = _gvn.transform(new AddINode(xlen, ylen));
5991 ciKlass* klass = ciTypeArrayKlass::make(T_INT);
5992 Node* klass_node = makecon(TypeKlassPtr::make(klass));
5993
5994 IdealKit ideal(this);
5995
5996 #define __ ideal.
5997 Node* one = __ ConI(1);
5998 Node* zero = __ ConI(0);
5999 IdealVariable need_alloc(ideal), z_alloc(ideal); __ declarations_done();
6000 __ set(need_alloc, zero);
6001 __ set(z_alloc, z);
6002 __ if_then(z, BoolTest::eq, null()); {
6003 __ increment (need_alloc, one);
6004 } __ else_(); {
6005 // Update graphKit memory and control from IdealKit.
6006 sync_kit(ideal);
6007 Node* cast = new CastPPNode(control(), z, TypePtr::NOTNULL);
6008 _gvn.set_type(cast, cast->bottom_type());
6009 C->record_for_igvn(cast);
6010
6011 Node* zlen_arg = load_array_length(cast);
6012 // Update IdealKit memory and control from graphKit.
6013 __ sync_kit(this);
6014 __ if_then(zlen_arg, BoolTest::lt, zlen); {
6015 __ increment (need_alloc, one);
6016 } __ end_if();
6017 } __ end_if();
6018
6019 __ if_then(__ value(need_alloc), BoolTest::ne, zero); {
6020 // Update graphKit memory and control from IdealKit.
6021 sync_kit(ideal);
6022 Node * narr = new_array(klass_node, zlen, 1);
6023 // Update IdealKit memory and control from graphKit.
6024 __ sync_kit(this);
6025 __ set(z_alloc, narr);
6026 } __ end_if();
6027
6028 sync_kit(ideal);
6029 z = __ value(z_alloc);
6030 // Can't use TypeAryPtr::INTS which uses Bottom offset.
6031 _gvn.set_type(z, TypeOopPtr::make_from_klass(klass));
6032 // Final sync IdealKit and GraphKit.
6033 final_sync(ideal);
6034 #undef __
6035
6036 Node* z_start = array_element_address(z, intcon(0), T_INT);
6037
6038 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6039 OptoRuntime::multiplyToLen_Type(),
6040 stubAddr, stubName, TypePtr::BOTTOM,
6041 x_start, xlen, y_start, ylen, z_start, zlen);
6042 } // original reexecute is set back here
6043
6044 C->set_has_split_ifs(true); // Has chance for split-if optimization
6045 set_result(z);
6046 return true;
6047 }
6048
6049 //-------------inline_squareToLen------------------------------------
6050 bool LibraryCallKit::inline_squareToLen() {
6051 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6052
6053 address stubAddr = StubRoutines::squareToLen();
6054 if (stubAddr == nullptr) {
6055 return false; // Intrinsic's stub is not implemented on this platform
6056 }
6057 const char* stubName = "squareToLen";
6058
6059 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6060
6061 Node* x = argument(0);
6062 Node* len = argument(1);
6063 Node* z = argument(2);
6064 Node* zlen = argument(3);
6065
6066 x = must_be_not_null(x, true);
6067 z = must_be_not_null(z, true);
6068
6069 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6070 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6071 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6072 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6073 // failed array check
6074 return false;
6075 }
6076
6077 BasicType x_elem = x_type->elem()->array_element_basic_type();
6078 BasicType z_elem = z_type->elem()->array_element_basic_type();
6079 if (x_elem != T_INT || z_elem != T_INT) {
6080 return false;
6081 }
6082
6083
6084 Node* x_start = array_element_address(x, intcon(0), x_elem);
6085 Node* z_start = array_element_address(z, intcon(0), z_elem);
6086
6087 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6088 OptoRuntime::squareToLen_Type(),
6089 stubAddr, stubName, TypePtr::BOTTOM,
6090 x_start, len, z_start, zlen);
6091
6092 set_result(z);
6093 return true;
6094 }
6095
6096 //-------------inline_mulAdd------------------------------------------
6097 bool LibraryCallKit::inline_mulAdd() {
6098 assert(UseMulAddIntrinsic, "not implemented on this platform");
6099
6100 address stubAddr = StubRoutines::mulAdd();
6101 if (stubAddr == nullptr) {
6102 return false; // Intrinsic's stub is not implemented on this platform
6103 }
6104 const char* stubName = "mulAdd";
6105
6106 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6107
6108 Node* out = argument(0);
6109 Node* in = argument(1);
6110 Node* offset = argument(2);
6111 Node* len = argument(3);
6112 Node* k = argument(4);
6113
6114 in = must_be_not_null(in, true);
6115 out = must_be_not_null(out, true);
6116
6117 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6118 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6119 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6120 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6121 // failed array check
6122 return false;
6123 }
6124
6125 BasicType out_elem = out_type->elem()->array_element_basic_type();
6126 BasicType in_elem = in_type->elem()->array_element_basic_type();
6127 if (out_elem != T_INT || in_elem != T_INT) {
6128 return false;
6129 }
6130
6131 Node* outlen = load_array_length(out);
6132 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6133 Node* out_start = array_element_address(out, intcon(0), out_elem);
6134 Node* in_start = array_element_address(in, intcon(0), in_elem);
6135
6136 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6137 OptoRuntime::mulAdd_Type(),
6138 stubAddr, stubName, TypePtr::BOTTOM,
6139 out_start,in_start, new_offset, len, k);
6140 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6141 set_result(result);
6142 return true;
6143 }
6144
6145 //-------------inline_montgomeryMultiply-----------------------------------
6146 bool LibraryCallKit::inline_montgomeryMultiply() {
6147 address stubAddr = StubRoutines::montgomeryMultiply();
6148 if (stubAddr == nullptr) {
6149 return false; // Intrinsic's stub is not implemented on this platform
6150 }
6151
6152 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6153 const char* stubName = "montgomery_multiply";
6154
6155 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6156
6157 Node* a = argument(0);
6158 Node* b = argument(1);
6159 Node* n = argument(2);
6160 Node* len = argument(3);
6161 Node* inv = argument(4);
6162 Node* m = argument(6);
6163
6164 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6165 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6166 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6167 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6168 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6169 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6170 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6171 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6172 // failed array check
6173 return false;
6174 }
6175
6176 BasicType a_elem = a_type->elem()->array_element_basic_type();
6177 BasicType b_elem = b_type->elem()->array_element_basic_type();
6178 BasicType n_elem = n_type->elem()->array_element_basic_type();
6179 BasicType m_elem = m_type->elem()->array_element_basic_type();
6180 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6181 return false;
6182 }
6183
6184 // Make the call
6185 {
6186 Node* a_start = array_element_address(a, intcon(0), a_elem);
6187 Node* b_start = array_element_address(b, intcon(0), b_elem);
6188 Node* n_start = array_element_address(n, intcon(0), n_elem);
6189 Node* m_start = array_element_address(m, intcon(0), m_elem);
6190
6191 Node* call = make_runtime_call(RC_LEAF,
6192 OptoRuntime::montgomeryMultiply_Type(),
6193 stubAddr, stubName, TypePtr::BOTTOM,
6194 a_start, b_start, n_start, len, inv, top(),
6195 m_start);
6196 set_result(m);
6197 }
6198
6199 return true;
6200 }
6201
6202 bool LibraryCallKit::inline_montgomerySquare() {
6203 address stubAddr = StubRoutines::montgomerySquare();
6204 if (stubAddr == nullptr) {
6205 return false; // Intrinsic's stub is not implemented on this platform
6206 }
6207
6208 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6209 const char* stubName = "montgomery_square";
6210
6211 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6212
6213 Node* a = argument(0);
6214 Node* n = argument(1);
6215 Node* len = argument(2);
6216 Node* inv = argument(3);
6217 Node* m = argument(5);
6218
6219 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6220 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6221 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6222 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6223 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6224 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6225 // failed array check
6226 return false;
6227 }
6228
6229 BasicType a_elem = a_type->elem()->array_element_basic_type();
6230 BasicType n_elem = n_type->elem()->array_element_basic_type();
6231 BasicType m_elem = m_type->elem()->array_element_basic_type();
6232 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6233 return false;
6234 }
6235
6236 // Make the call
6237 {
6238 Node* a_start = array_element_address(a, intcon(0), a_elem);
6239 Node* n_start = array_element_address(n, intcon(0), n_elem);
6240 Node* m_start = array_element_address(m, intcon(0), m_elem);
6241
6242 Node* call = make_runtime_call(RC_LEAF,
6243 OptoRuntime::montgomerySquare_Type(),
6244 stubAddr, stubName, TypePtr::BOTTOM,
6245 a_start, n_start, len, inv, top(),
6246 m_start);
6247 set_result(m);
6248 }
6249
6250 return true;
6251 }
6252
6253 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6254 address stubAddr = nullptr;
6255 const char* stubName = nullptr;
6256
6257 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6258 if (stubAddr == nullptr) {
6259 return false; // Intrinsic's stub is not implemented on this platform
6260 }
6261
6262 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6263
6264 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6265
6266 Node* newArr = argument(0);
6267 Node* oldArr = argument(1);
6268 Node* newIdx = argument(2);
6269 Node* shiftCount = argument(3);
6270 Node* numIter = argument(4);
6271
6272 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6273 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6274 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6275 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6276 return false;
6277 }
6278
6279 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6280 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6281 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6282 return false;
6283 }
6284
6285 // Make the call
6286 {
6287 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6288 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6289
6290 Node* call = make_runtime_call(RC_LEAF,
6291 OptoRuntime::bigIntegerShift_Type(),
6292 stubAddr,
6293 stubName,
6294 TypePtr::BOTTOM,
6295 newArr_start,
6296 oldArr_start,
6297 newIdx,
6298 shiftCount,
6299 numIter);
6300 }
6301
6302 return true;
6303 }
6304
6305 //-------------inline_vectorizedMismatch------------------------------
6306 bool LibraryCallKit::inline_vectorizedMismatch() {
6307 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6308
6309 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6310 Node* obja = argument(0); // Object
6311 Node* aoffset = argument(1); // long
6312 Node* objb = argument(3); // Object
6313 Node* boffset = argument(4); // long
6314 Node* length = argument(6); // int
6315 Node* scale = argument(7); // int
6316
6317 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6318 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6319 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6320 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6321 scale == top()) {
6322 return false; // failed input validation
6323 }
6324
6325 Node* obja_adr = make_unsafe_address(obja, aoffset);
6326 Node* objb_adr = make_unsafe_address(objb, boffset);
6327
6328 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6329 //
6330 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6331 // if (length <= inline_limit) {
6332 // inline_path:
6333 // vmask = VectorMaskGen length
6334 // vload1 = LoadVectorMasked obja, vmask
6335 // vload2 = LoadVectorMasked objb, vmask
6336 // result1 = VectorCmpMasked vload1, vload2, vmask
6337 // } else {
6338 // call_stub_path:
6339 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6340 // }
6341 // exit_block:
6342 // return Phi(result1, result2);
6343 //
6344 enum { inline_path = 1, // input is small enough to process it all at once
6345 stub_path = 2, // input is too large; call into the VM
6346 PATH_LIMIT = 3
6347 };
6348
6349 Node* exit_block = new RegionNode(PATH_LIMIT);
6350 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6351 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6352
6353 Node* call_stub_path = control();
6354
6355 BasicType elem_bt = T_ILLEGAL;
6356
6357 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6358 if (scale_t->is_con()) {
6359 switch (scale_t->get_con()) {
6360 case 0: elem_bt = T_BYTE; break;
6361 case 1: elem_bt = T_SHORT; break;
6362 case 2: elem_bt = T_INT; break;
6363 case 3: elem_bt = T_LONG; break;
6364
6365 default: elem_bt = T_ILLEGAL; break; // not supported
6366 }
6367 }
6368
6369 int inline_limit = 0;
6370 bool do_partial_inline = false;
6371
6372 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6373 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6374 do_partial_inline = inline_limit >= 16;
6375 }
6376
6377 if (do_partial_inline) {
6378 assert(elem_bt != T_ILLEGAL, "sanity");
6379
6380 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6381 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6382 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6383
6384 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6385 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6386 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6387
6388 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6389
6390 if (!stopped()) {
6391 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6392
6393 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6394 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6395 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6396 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6397
6398 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6399 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6400 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6401 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6402
6403 exit_block->init_req(inline_path, control());
6404 memory_phi->init_req(inline_path, map()->memory());
6405 result_phi->init_req(inline_path, result);
6406
6407 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6408 clear_upper_avx();
6409 }
6410 }
6411 }
6412
6413 if (call_stub_path != nullptr) {
6414 set_control(call_stub_path);
6415
6416 Node* call = make_runtime_call(RC_LEAF,
6417 OptoRuntime::vectorizedMismatch_Type(),
6418 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6419 obja_adr, objb_adr, length, scale);
6420
6421 exit_block->init_req(stub_path, control());
6422 memory_phi->init_req(stub_path, map()->memory());
6423 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6424 }
6425
6426 exit_block = _gvn.transform(exit_block);
6427 memory_phi = _gvn.transform(memory_phi);
6428 result_phi = _gvn.transform(result_phi);
6429
6430 set_control(exit_block);
6431 set_all_memory(memory_phi);
6432 set_result(result_phi);
6433
6434 return true;
6435 }
6436
6437 //------------------------------inline_vectorizedHashcode----------------------------
6438 bool LibraryCallKit::inline_vectorizedHashCode() {
6439 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6440
6441 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6442 Node* array = argument(0);
6443 Node* offset = argument(1);
6444 Node* length = argument(2);
6445 Node* initialValue = argument(3);
6446 Node* basic_type = argument(4);
6447
6448 if (basic_type == top()) {
6449 return false; // failed input validation
6450 }
6451
6452 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6453 if (!basic_type_t->is_con()) {
6454 return false; // Only intrinsify if mode argument is constant
6455 }
6456
6457 array = must_be_not_null(array, true);
6458
6459 BasicType bt = (BasicType)basic_type_t->get_con();
6460
6461 // Resolve address of first element
6462 Node* array_start = array_element_address(array, offset, bt);
6463
6464 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6465 array_start, length, initialValue, basic_type)));
6466 clear_upper_avx();
6467
6468 return true;
6469 }
6470
6471 /**
6472 * Calculate CRC32 for byte.
6473 * int java.util.zip.CRC32.update(int crc, int b)
6474 */
6475 bool LibraryCallKit::inline_updateCRC32() {
6476 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6477 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6478 // no receiver since it is static method
6479 Node* crc = argument(0); // type: int
6480 Node* b = argument(1); // type: int
6481
6482 /*
6483 * int c = ~ crc;
6484 * b = timesXtoThe32[(b ^ c) & 0xFF];
6485 * b = b ^ (c >>> 8);
6486 * crc = ~b;
6487 */
6488
6489 Node* M1 = intcon(-1);
6490 crc = _gvn.transform(new XorINode(crc, M1));
6491 Node* result = _gvn.transform(new XorINode(crc, b));
6492 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6493
6494 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6495 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6496 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6497 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6498
6499 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6500 result = _gvn.transform(new XorINode(crc, result));
6501 result = _gvn.transform(new XorINode(result, M1));
6502 set_result(result);
6503 return true;
6504 }
6505
6506 /**
6507 * Calculate CRC32 for byte[] array.
6508 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6509 */
6510 bool LibraryCallKit::inline_updateBytesCRC32() {
6511 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6512 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6513 // no receiver since it is static method
6514 Node* crc = argument(0); // type: int
6515 Node* src = argument(1); // type: oop
6516 Node* offset = argument(2); // type: int
6517 Node* length = argument(3); // type: int
6518
6519 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6520 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6521 // failed array check
6522 return false;
6523 }
6524
6525 // Figure out the size and type of the elements we will be copying.
6526 BasicType src_elem = src_type->elem()->array_element_basic_type();
6527 if (src_elem != T_BYTE) {
6528 return false;
6529 }
6530
6531 // 'src_start' points to src array + scaled offset
6532 src = must_be_not_null(src, true);
6533 Node* src_start = array_element_address(src, offset, src_elem);
6534
6535 // We assume that range check is done by caller.
6536 // TODO: generate range check (offset+length < src.length) in debug VM.
6537
6538 // Call the stub.
6539 address stubAddr = StubRoutines::updateBytesCRC32();
6540 const char *stubName = "updateBytesCRC32";
6541
6542 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6543 stubAddr, stubName, TypePtr::BOTTOM,
6544 crc, src_start, length);
6545 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6546 set_result(result);
6547 return true;
6548 }
6549
6550 /**
6551 * Calculate CRC32 for ByteBuffer.
6552 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6553 */
6554 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6555 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6556 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6557 // no receiver since it is static method
6558 Node* crc = argument(0); // type: int
6559 Node* src = argument(1); // type: long
6560 Node* offset = argument(3); // type: int
6561 Node* length = argument(4); // type: int
6562
6563 src = ConvL2X(src); // adjust Java long to machine word
6564 Node* base = _gvn.transform(new CastX2PNode(src));
6565 offset = ConvI2X(offset);
6566
6567 // 'src_start' points to src array + scaled offset
6568 Node* src_start = basic_plus_adr(top(), base, offset);
6569
6570 // Call the stub.
6571 address stubAddr = StubRoutines::updateBytesCRC32();
6572 const char *stubName = "updateBytesCRC32";
6573
6574 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6575 stubAddr, stubName, TypePtr::BOTTOM,
6576 crc, src_start, length);
6577 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6578 set_result(result);
6579 return true;
6580 }
6581
6582 //------------------------------get_table_from_crc32c_class-----------------------
6583 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6584 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6585 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6586
6587 return table;
6588 }
6589
6590 //------------------------------inline_updateBytesCRC32C-----------------------
6591 //
6592 // Calculate CRC32C for byte[] array.
6593 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6594 //
6595 bool LibraryCallKit::inline_updateBytesCRC32C() {
6596 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6597 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6598 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6599 // no receiver since it is a static method
6600 Node* crc = argument(0); // type: int
6601 Node* src = argument(1); // type: oop
6602 Node* offset = argument(2); // type: int
6603 Node* end = argument(3); // type: int
6604
6605 Node* length = _gvn.transform(new SubINode(end, offset));
6606
6607 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6608 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6609 // failed array check
6610 return false;
6611 }
6612
6613 // Figure out the size and type of the elements we will be copying.
6614 BasicType src_elem = src_type->elem()->array_element_basic_type();
6615 if (src_elem != T_BYTE) {
6616 return false;
6617 }
6618
6619 // 'src_start' points to src array + scaled offset
6620 src = must_be_not_null(src, true);
6621 Node* src_start = array_element_address(src, offset, src_elem);
6622
6623 // static final int[] byteTable in class CRC32C
6624 Node* table = get_table_from_crc32c_class(callee()->holder());
6625 table = must_be_not_null(table, true);
6626 Node* table_start = array_element_address(table, intcon(0), T_INT);
6627
6628 // We assume that range check is done by caller.
6629 // TODO: generate range check (offset+length < src.length) in debug VM.
6630
6631 // Call the stub.
6632 address stubAddr = StubRoutines::updateBytesCRC32C();
6633 const char *stubName = "updateBytesCRC32C";
6634
6635 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6636 stubAddr, stubName, TypePtr::BOTTOM,
6637 crc, src_start, length, table_start);
6638 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6639 set_result(result);
6640 return true;
6641 }
6642
6643 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6644 //
6645 // Calculate CRC32C for DirectByteBuffer.
6646 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6647 //
6648 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6649 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6650 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6651 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6652 // no receiver since it is a static method
6653 Node* crc = argument(0); // type: int
6654 Node* src = argument(1); // type: long
6655 Node* offset = argument(3); // type: int
6656 Node* end = argument(4); // type: int
6657
6658 Node* length = _gvn.transform(new SubINode(end, offset));
6659
6660 src = ConvL2X(src); // adjust Java long to machine word
6661 Node* base = _gvn.transform(new CastX2PNode(src));
6662 offset = ConvI2X(offset);
6663
6664 // 'src_start' points to src array + scaled offset
6665 Node* src_start = basic_plus_adr(top(), base, offset);
6666
6667 // static final int[] byteTable in class CRC32C
6668 Node* table = get_table_from_crc32c_class(callee()->holder());
6669 table = must_be_not_null(table, true);
6670 Node* table_start = array_element_address(table, intcon(0), T_INT);
6671
6672 // Call the stub.
6673 address stubAddr = StubRoutines::updateBytesCRC32C();
6674 const char *stubName = "updateBytesCRC32C";
6675
6676 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6677 stubAddr, stubName, TypePtr::BOTTOM,
6678 crc, src_start, length, table_start);
6679 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6680 set_result(result);
6681 return true;
6682 }
6683
6684 //------------------------------inline_updateBytesAdler32----------------------
6685 //
6686 // Calculate Adler32 checksum for byte[] array.
6687 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6688 //
6689 bool LibraryCallKit::inline_updateBytesAdler32() {
6690 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6691 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6692 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6693 // no receiver since it is static method
6694 Node* crc = argument(0); // type: int
6695 Node* src = argument(1); // type: oop
6696 Node* offset = argument(2); // type: int
6697 Node* length = argument(3); // type: int
6698
6699 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6700 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6701 // failed array check
6702 return false;
6703 }
6704
6705 // Figure out the size and type of the elements we will be copying.
6706 BasicType src_elem = src_type->elem()->array_element_basic_type();
6707 if (src_elem != T_BYTE) {
6708 return false;
6709 }
6710
6711 // 'src_start' points to src array + scaled offset
6712 Node* src_start = array_element_address(src, offset, src_elem);
6713
6714 // We assume that range check is done by caller.
6715 // TODO: generate range check (offset+length < src.length) in debug VM.
6716
6717 // Call the stub.
6718 address stubAddr = StubRoutines::updateBytesAdler32();
6719 const char *stubName = "updateBytesAdler32";
6720
6721 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6722 stubAddr, stubName, TypePtr::BOTTOM,
6723 crc, src_start, length);
6724 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6725 set_result(result);
6726 return true;
6727 }
6728
6729 //------------------------------inline_updateByteBufferAdler32---------------
6730 //
6731 // Calculate Adler32 checksum for DirectByteBuffer.
6732 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6733 //
6734 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6735 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6736 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6737 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6738 // no receiver since it is static method
6739 Node* crc = argument(0); // type: int
6740 Node* src = argument(1); // type: long
6741 Node* offset = argument(3); // type: int
6742 Node* length = argument(4); // type: int
6743
6744 src = ConvL2X(src); // adjust Java long to machine word
6745 Node* base = _gvn.transform(new CastX2PNode(src));
6746 offset = ConvI2X(offset);
6747
6748 // 'src_start' points to src array + scaled offset
6749 Node* src_start = basic_plus_adr(top(), base, offset);
6750
6751 // Call the stub.
6752 address stubAddr = StubRoutines::updateBytesAdler32();
6753 const char *stubName = "updateBytesAdler32";
6754
6755 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6756 stubAddr, stubName, TypePtr::BOTTOM,
6757 crc, src_start, length);
6758
6759 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6760 set_result(result);
6761 return true;
6762 }
6763
6764 //----------------------------inline_reference_get----------------------------
6765 // public T java.lang.ref.Reference.get();
6766 bool LibraryCallKit::inline_reference_get() {
6767 const int referent_offset = java_lang_ref_Reference::referent_offset();
6768
6769 // Get the argument:
6770 Node* reference_obj = null_check_receiver();
6771 if (stopped()) return true;
6772
6773 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6774 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6775 decorators, /*is_static*/ false, nullptr);
6776 if (result == nullptr) return false;
6777
6778 // Add memory barrier to prevent commoning reads from this field
6779 // across safepoint since GC can change its value.
6780 insert_mem_bar(Op_MemBarCPUOrder);
6781
6782 set_result(result);
6783 return true;
6784 }
6785
6786 //----------------------------inline_reference_refersTo0----------------------------
6787 // bool java.lang.ref.Reference.refersTo0();
6788 // bool java.lang.ref.PhantomReference.refersTo0();
6789 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6790 // Get arguments:
6791 Node* reference_obj = null_check_receiver();
6792 Node* other_obj = argument(1);
6793 if (stopped()) return true;
6794
6795 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6796 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6797 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6798 decorators, /*is_static*/ false, nullptr);
6799 if (referent == nullptr) return false;
6800
6801 // Add memory barrier to prevent commoning reads from this field
6802 // across safepoint since GC can change its value.
6803 insert_mem_bar(Op_MemBarCPUOrder);
6804
6805 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6806 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6807 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6808
6809 RegionNode* region = new RegionNode(3);
6810 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6811
6812 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6813 region->init_req(1, if_true);
6814 phi->init_req(1, intcon(1));
6815
6816 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6817 region->init_req(2, if_false);
6818 phi->init_req(2, intcon(0));
6819
6820 set_control(_gvn.transform(region));
6821 record_for_igvn(region);
6822 set_result(_gvn.transform(phi));
6823 return true;
6824 }
6825
6826
6827 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6828 DecoratorSet decorators, bool is_static,
6829 ciInstanceKlass* fromKls) {
6830 if (fromKls == nullptr) {
6831 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6832 assert(tinst != nullptr, "obj is null");
6833 assert(tinst->is_loaded(), "obj is not loaded");
6834 fromKls = tinst->instance_klass();
6835 } else {
6836 assert(is_static, "only for static field access");
6837 }
6838 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6839 ciSymbol::make(fieldTypeString),
6840 is_static);
6841
6842 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6843 if (field == nullptr) return (Node *) nullptr;
6844
6845 if (is_static) {
6846 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6847 fromObj = makecon(tip);
6848 }
6849
6850 // Next code copied from Parse::do_get_xxx():
6851
6852 // Compute address and memory type.
6853 int offset = field->offset_in_bytes();
6854 bool is_vol = field->is_volatile();
6855 ciType* field_klass = field->type();
6856 assert(field_klass->is_loaded(), "should be loaded");
6857 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6858 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6859 BasicType bt = field->layout_type();
6860
6861 // Build the resultant type of the load
6862 const Type *type;
6863 if (bt == T_OBJECT) {
6864 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
6865 } else {
6866 type = Type::get_const_basic_type(bt);
6867 }
6868
6869 if (is_vol) {
6870 decorators |= MO_SEQ_CST;
6871 }
6872
6873 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
6874 }
6875
6876 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
6877 bool is_exact /* true */, bool is_static /* false */,
6878 ciInstanceKlass * fromKls /* nullptr */) {
6879 if (fromKls == nullptr) {
6880 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6881 assert(tinst != nullptr, "obj is null");
6882 assert(tinst->is_loaded(), "obj is not loaded");
6883 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
6884 fromKls = tinst->instance_klass();
6885 }
6886 else {
6887 assert(is_static, "only for static field access");
6888 }
6889 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6890 ciSymbol::make(fieldTypeString),
6891 is_static);
6892
6893 assert(field != nullptr, "undefined field");
6894 assert(!field->is_volatile(), "not defined for volatile fields");
6895
6896 if (is_static) {
6897 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6898 fromObj = makecon(tip);
6899 }
6900
6901 // Next code copied from Parse::do_get_xxx():
6902
6903 // Compute address and memory type.
6904 int offset = field->offset_in_bytes();
6905 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6906
6907 return adr;
6908 }
6909
6910 //------------------------------inline_aescrypt_Block-----------------------
6911 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
6912 address stubAddr = nullptr;
6913 const char *stubName;
6914 assert(UseAES, "need AES instruction support");
6915
6916 switch(id) {
6917 case vmIntrinsics::_aescrypt_encryptBlock:
6918 stubAddr = StubRoutines::aescrypt_encryptBlock();
6919 stubName = "aescrypt_encryptBlock";
6920 break;
6921 case vmIntrinsics::_aescrypt_decryptBlock:
6922 stubAddr = StubRoutines::aescrypt_decryptBlock();
6923 stubName = "aescrypt_decryptBlock";
6924 break;
6925 default:
6926 break;
6927 }
6928 if (stubAddr == nullptr) return false;
6929
6930 Node* aescrypt_object = argument(0);
6931 Node* src = argument(1);
6932 Node* src_offset = argument(2);
6933 Node* dest = argument(3);
6934 Node* dest_offset = argument(4);
6935
6936 src = must_be_not_null(src, true);
6937 dest = must_be_not_null(dest, true);
6938
6939 // (1) src and dest are arrays.
6940 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6941 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
6942 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
6943 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
6944
6945 // for the quick and dirty code we will skip all the checks.
6946 // we are just trying to get the call to be generated.
6947 Node* src_start = src;
6948 Node* dest_start = dest;
6949 if (src_offset != nullptr || dest_offset != nullptr) {
6950 assert(src_offset != nullptr && dest_offset != nullptr, "");
6951 src_start = array_element_address(src, src_offset, T_BYTE);
6952 dest_start = array_element_address(dest, dest_offset, T_BYTE);
6953 }
6954
6955 // now need to get the start of its expanded key array
6956 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
6957 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
6958 if (k_start == nullptr) return false;
6959
6960 // Call the stub.
6961 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
6962 stubAddr, stubName, TypePtr::BOTTOM,
6963 src_start, dest_start, k_start);
6964
6965 return true;
6966 }
6967
6968 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
6969 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
6970 address stubAddr = nullptr;
6971 const char *stubName = nullptr;
6972
6973 assert(UseAES, "need AES instruction support");
6974
6975 switch(id) {
6976 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
6977 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
6978 stubName = "cipherBlockChaining_encryptAESCrypt";
6979 break;
6980 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
6981 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
6982 stubName = "cipherBlockChaining_decryptAESCrypt";
6983 break;
6984 default:
6985 break;
6986 }
6987 if (stubAddr == nullptr) return false;
6988
6989 Node* cipherBlockChaining_object = argument(0);
6990 Node* src = argument(1);
6991 Node* src_offset = argument(2);
6992 Node* len = argument(3);
6993 Node* dest = argument(4);
6994 Node* dest_offset = argument(5);
6995
6996 src = must_be_not_null(src, false);
6997 dest = must_be_not_null(dest, false);
6998
6999 // (1) src and dest are arrays.
7000 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7001 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7002 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7003 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7004
7005 // checks are the responsibility of the caller
7006 Node* src_start = src;
7007 Node* dest_start = dest;
7008 if (src_offset != nullptr || dest_offset != nullptr) {
7009 assert(src_offset != nullptr && dest_offset != nullptr, "");
7010 src_start = array_element_address(src, src_offset, T_BYTE);
7011 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7012 }
7013
7014 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7015 // (because of the predicated logic executed earlier).
7016 // so we cast it here safely.
7017 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7018
7019 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7020 if (embeddedCipherObj == nullptr) return false;
7021
7022 // cast it to what we know it will be at runtime
7023 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7024 assert(tinst != nullptr, "CBC obj is null");
7025 assert(tinst->is_loaded(), "CBC obj is not loaded");
7026 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7027 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7028
7029 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7030 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7031 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7032 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7033 aescrypt_object = _gvn.transform(aescrypt_object);
7034
7035 // we need to get the start of the aescrypt_object's expanded key array
7036 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7037 if (k_start == nullptr) return false;
7038
7039 // similarly, get the start address of the r vector
7040 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7041 if (objRvec == nullptr) return false;
7042 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7043
7044 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7045 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7046 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7047 stubAddr, stubName, TypePtr::BOTTOM,
7048 src_start, dest_start, k_start, r_start, len);
7049
7050 // return cipher length (int)
7051 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7052 set_result(retvalue);
7053 return true;
7054 }
7055
7056 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7057 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7058 address stubAddr = nullptr;
7059 const char *stubName = nullptr;
7060
7061 assert(UseAES, "need AES instruction support");
7062
7063 switch (id) {
7064 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7065 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7066 stubName = "electronicCodeBook_encryptAESCrypt";
7067 break;
7068 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7069 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7070 stubName = "electronicCodeBook_decryptAESCrypt";
7071 break;
7072 default:
7073 break;
7074 }
7075
7076 if (stubAddr == nullptr) return false;
7077
7078 Node* electronicCodeBook_object = argument(0);
7079 Node* src = argument(1);
7080 Node* src_offset = argument(2);
7081 Node* len = argument(3);
7082 Node* dest = argument(4);
7083 Node* dest_offset = argument(5);
7084
7085 // (1) src and dest are arrays.
7086 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7087 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7088 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7089 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7090
7091 // checks are the responsibility of the caller
7092 Node* src_start = src;
7093 Node* dest_start = dest;
7094 if (src_offset != nullptr || dest_offset != nullptr) {
7095 assert(src_offset != nullptr && dest_offset != nullptr, "");
7096 src_start = array_element_address(src, src_offset, T_BYTE);
7097 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7098 }
7099
7100 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7101 // (because of the predicated logic executed earlier).
7102 // so we cast it here safely.
7103 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7104
7105 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7106 if (embeddedCipherObj == nullptr) return false;
7107
7108 // cast it to what we know it will be at runtime
7109 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7110 assert(tinst != nullptr, "ECB obj is null");
7111 assert(tinst->is_loaded(), "ECB obj is not loaded");
7112 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7113 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7114
7115 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7116 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7117 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7118 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7119 aescrypt_object = _gvn.transform(aescrypt_object);
7120
7121 // we need to get the start of the aescrypt_object's expanded key array
7122 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7123 if (k_start == nullptr) return false;
7124
7125 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7126 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7127 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7128 stubAddr, stubName, TypePtr::BOTTOM,
7129 src_start, dest_start, k_start, len);
7130
7131 // return cipher length (int)
7132 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7133 set_result(retvalue);
7134 return true;
7135 }
7136
7137 //------------------------------inline_counterMode_AESCrypt-----------------------
7138 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7139 assert(UseAES, "need AES instruction support");
7140 if (!UseAESCTRIntrinsics) return false;
7141
7142 address stubAddr = nullptr;
7143 const char *stubName = nullptr;
7144 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7145 stubAddr = StubRoutines::counterMode_AESCrypt();
7146 stubName = "counterMode_AESCrypt";
7147 }
7148 if (stubAddr == nullptr) return false;
7149
7150 Node* counterMode_object = argument(0);
7151 Node* src = argument(1);
7152 Node* src_offset = argument(2);
7153 Node* len = argument(3);
7154 Node* dest = argument(4);
7155 Node* dest_offset = argument(5);
7156
7157 // (1) src and dest are arrays.
7158 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7159 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7160 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7161 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7162
7163 // checks are the responsibility of the caller
7164 Node* src_start = src;
7165 Node* dest_start = dest;
7166 if (src_offset != nullptr || dest_offset != nullptr) {
7167 assert(src_offset != nullptr && dest_offset != nullptr, "");
7168 src_start = array_element_address(src, src_offset, T_BYTE);
7169 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7170 }
7171
7172 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7173 // (because of the predicated logic executed earlier).
7174 // so we cast it here safely.
7175 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7176 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7177 if (embeddedCipherObj == nullptr) return false;
7178 // cast it to what we know it will be at runtime
7179 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7180 assert(tinst != nullptr, "CTR obj is null");
7181 assert(tinst->is_loaded(), "CTR obj is not loaded");
7182 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7183 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7184 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7185 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7186 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7187 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7188 aescrypt_object = _gvn.transform(aescrypt_object);
7189 // we need to get the start of the aescrypt_object's expanded key array
7190 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7191 if (k_start == nullptr) return false;
7192 // similarly, get the start address of the r vector
7193 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7194 if (obj_counter == nullptr) return false;
7195 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7196
7197 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7198 if (saved_encCounter == nullptr) return false;
7199 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7200 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7201
7202 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7203 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7204 OptoRuntime::counterMode_aescrypt_Type(),
7205 stubAddr, stubName, TypePtr::BOTTOM,
7206 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7207
7208 // return cipher length (int)
7209 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7210 set_result(retvalue);
7211 return true;
7212 }
7213
7214 //------------------------------get_key_start_from_aescrypt_object-----------------------
7215 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7216 #if defined(PPC64) || defined(S390)
7217 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7218 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7219 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7220 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7221 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7222 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7223 if (objSessionK == nullptr) {
7224 return (Node *) nullptr;
7225 }
7226 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7227 #else
7228 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7229 #endif // PPC64
7230 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7231 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7232
7233 // now have the array, need to get the start address of the K array
7234 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7235 return k_start;
7236 }
7237
7238 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7239 // Return node representing slow path of predicate check.
7240 // the pseudo code we want to emulate with this predicate is:
7241 // for encryption:
7242 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7243 // for decryption:
7244 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7245 // note cipher==plain is more conservative than the original java code but that's OK
7246 //
7247 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7248 // The receiver was checked for null already.
7249 Node* objCBC = argument(0);
7250
7251 Node* src = argument(1);
7252 Node* dest = argument(4);
7253
7254 // Load embeddedCipher field of CipherBlockChaining object.
7255 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7256
7257 // get AESCrypt klass for instanceOf check
7258 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7259 // will have same classloader as CipherBlockChaining object
7260 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7261 assert(tinst != nullptr, "CBCobj is null");
7262 assert(tinst->is_loaded(), "CBCobj is not loaded");
7263
7264 // we want to do an instanceof comparison against the AESCrypt class
7265 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7266 if (!klass_AESCrypt->is_loaded()) {
7267 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7268 Node* ctrl = control();
7269 set_control(top()); // no regular fast path
7270 return ctrl;
7271 }
7272
7273 src = must_be_not_null(src, true);
7274 dest = must_be_not_null(dest, true);
7275
7276 // Resolve oops to stable for CmpP below.
7277 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7278
7279 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7280 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7281 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7282
7283 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7284
7285 // for encryption, we are done
7286 if (!decrypting)
7287 return instof_false; // even if it is null
7288
7289 // for decryption, we need to add a further check to avoid
7290 // taking the intrinsic path when cipher and plain are the same
7291 // see the original java code for why.
7292 RegionNode* region = new RegionNode(3);
7293 region->init_req(1, instof_false);
7294
7295 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7296 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7297 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7298 region->init_req(2, src_dest_conjoint);
7299
7300 record_for_igvn(region);
7301 return _gvn.transform(region);
7302 }
7303
7304 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7305 // Return node representing slow path of predicate check.
7306 // the pseudo code we want to emulate with this predicate is:
7307 // for encryption:
7308 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7309 // for decryption:
7310 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7311 // note cipher==plain is more conservative than the original java code but that's OK
7312 //
7313 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7314 // The receiver was checked for null already.
7315 Node* objECB = argument(0);
7316
7317 // Load embeddedCipher field of ElectronicCodeBook object.
7318 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7319
7320 // get AESCrypt klass for instanceOf check
7321 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7322 // will have same classloader as ElectronicCodeBook object
7323 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7324 assert(tinst != nullptr, "ECBobj is null");
7325 assert(tinst->is_loaded(), "ECBobj is not loaded");
7326
7327 // we want to do an instanceof comparison against the AESCrypt class
7328 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7329 if (!klass_AESCrypt->is_loaded()) {
7330 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7331 Node* ctrl = control();
7332 set_control(top()); // no regular fast path
7333 return ctrl;
7334 }
7335 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7336
7337 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7338 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7339 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7340
7341 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7342
7343 // for encryption, we are done
7344 if (!decrypting)
7345 return instof_false; // even if it is null
7346
7347 // for decryption, we need to add a further check to avoid
7348 // taking the intrinsic path when cipher and plain are the same
7349 // see the original java code for why.
7350 RegionNode* region = new RegionNode(3);
7351 region->init_req(1, instof_false);
7352 Node* src = argument(1);
7353 Node* dest = argument(4);
7354 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7355 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7356 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7357 region->init_req(2, src_dest_conjoint);
7358
7359 record_for_igvn(region);
7360 return _gvn.transform(region);
7361 }
7362
7363 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7364 // Return node representing slow path of predicate check.
7365 // the pseudo code we want to emulate with this predicate is:
7366 // for encryption:
7367 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7368 // for decryption:
7369 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7370 // note cipher==plain is more conservative than the original java code but that's OK
7371 //
7372
7373 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7374 // The receiver was checked for null already.
7375 Node* objCTR = argument(0);
7376
7377 // Load embeddedCipher field of CipherBlockChaining object.
7378 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7379
7380 // get AESCrypt klass for instanceOf check
7381 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7382 // will have same classloader as CipherBlockChaining object
7383 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7384 assert(tinst != nullptr, "CTRobj is null");
7385 assert(tinst->is_loaded(), "CTRobj is not loaded");
7386
7387 // we want to do an instanceof comparison against the AESCrypt class
7388 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7389 if (!klass_AESCrypt->is_loaded()) {
7390 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7391 Node* ctrl = control();
7392 set_control(top()); // no regular fast path
7393 return ctrl;
7394 }
7395
7396 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7397 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7398 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7399 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7400 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7401
7402 return instof_false; // even if it is null
7403 }
7404
7405 //------------------------------inline_ghash_processBlocks
7406 bool LibraryCallKit::inline_ghash_processBlocks() {
7407 address stubAddr;
7408 const char *stubName;
7409 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7410
7411 stubAddr = StubRoutines::ghash_processBlocks();
7412 stubName = "ghash_processBlocks";
7413
7414 Node* data = argument(0);
7415 Node* offset = argument(1);
7416 Node* len = argument(2);
7417 Node* state = argument(3);
7418 Node* subkeyH = argument(4);
7419
7420 state = must_be_not_null(state, true);
7421 subkeyH = must_be_not_null(subkeyH, true);
7422 data = must_be_not_null(data, true);
7423
7424 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7425 assert(state_start, "state is null");
7426 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7427 assert(subkeyH_start, "subkeyH is null");
7428 Node* data_start = array_element_address(data, offset, T_BYTE);
7429 assert(data_start, "data is null");
7430
7431 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7432 OptoRuntime::ghash_processBlocks_Type(),
7433 stubAddr, stubName, TypePtr::BOTTOM,
7434 state_start, subkeyH_start, data_start, len);
7435 return true;
7436 }
7437
7438 //------------------------------inline_chacha20Block
7439 bool LibraryCallKit::inline_chacha20Block() {
7440 address stubAddr;
7441 const char *stubName;
7442 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7443
7444 stubAddr = StubRoutines::chacha20Block();
7445 stubName = "chacha20Block";
7446
7447 Node* state = argument(0);
7448 Node* result = argument(1);
7449
7450 state = must_be_not_null(state, true);
7451 result = must_be_not_null(result, true);
7452
7453 Node* state_start = array_element_address(state, intcon(0), T_INT);
7454 assert(state_start, "state is null");
7455 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7456 assert(result_start, "result is null");
7457
7458 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7459 OptoRuntime::chacha20Block_Type(),
7460 stubAddr, stubName, TypePtr::BOTTOM,
7461 state_start, result_start);
7462 // return key stream length (int)
7463 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7464 set_result(retvalue);
7465 return true;
7466 }
7467
7468 bool LibraryCallKit::inline_base64_encodeBlock() {
7469 address stubAddr;
7470 const char *stubName;
7471 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7472 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7473 stubAddr = StubRoutines::base64_encodeBlock();
7474 stubName = "encodeBlock";
7475
7476 if (!stubAddr) return false;
7477 Node* base64obj = argument(0);
7478 Node* src = argument(1);
7479 Node* offset = argument(2);
7480 Node* len = argument(3);
7481 Node* dest = argument(4);
7482 Node* dp = argument(5);
7483 Node* isURL = argument(6);
7484
7485 src = must_be_not_null(src, true);
7486 dest = must_be_not_null(dest, true);
7487
7488 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7489 assert(src_start, "source array is null");
7490 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7491 assert(dest_start, "destination array is null");
7492
7493 Node* base64 = make_runtime_call(RC_LEAF,
7494 OptoRuntime::base64_encodeBlock_Type(),
7495 stubAddr, stubName, TypePtr::BOTTOM,
7496 src_start, offset, len, dest_start, dp, isURL);
7497 return true;
7498 }
7499
7500 bool LibraryCallKit::inline_base64_decodeBlock() {
7501 address stubAddr;
7502 const char *stubName;
7503 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7504 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7505 stubAddr = StubRoutines::base64_decodeBlock();
7506 stubName = "decodeBlock";
7507
7508 if (!stubAddr) return false;
7509 Node* base64obj = argument(0);
7510 Node* src = argument(1);
7511 Node* src_offset = argument(2);
7512 Node* len = argument(3);
7513 Node* dest = argument(4);
7514 Node* dest_offset = argument(5);
7515 Node* isURL = argument(6);
7516 Node* isMIME = argument(7);
7517
7518 src = must_be_not_null(src, true);
7519 dest = must_be_not_null(dest, true);
7520
7521 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7522 assert(src_start, "source array is null");
7523 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7524 assert(dest_start, "destination array is null");
7525
7526 Node* call = make_runtime_call(RC_LEAF,
7527 OptoRuntime::base64_decodeBlock_Type(),
7528 stubAddr, stubName, TypePtr::BOTTOM,
7529 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7530 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7531 set_result(result);
7532 return true;
7533 }
7534
7535 bool LibraryCallKit::inline_poly1305_processBlocks() {
7536 address stubAddr;
7537 const char *stubName;
7538 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7539 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7540 stubAddr = StubRoutines::poly1305_processBlocks();
7541 stubName = "poly1305_processBlocks";
7542
7543 if (!stubAddr) return false;
7544 null_check_receiver(); // null-check receiver
7545 if (stopped()) return true;
7546
7547 Node* input = argument(1);
7548 Node* input_offset = argument(2);
7549 Node* len = argument(3);
7550 Node* alimbs = argument(4);
7551 Node* rlimbs = argument(5);
7552
7553 input = must_be_not_null(input, true);
7554 alimbs = must_be_not_null(alimbs, true);
7555 rlimbs = must_be_not_null(rlimbs, true);
7556
7557 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7558 assert(input_start, "input array is null");
7559 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7560 assert(acc_start, "acc array is null");
7561 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7562 assert(r_start, "r array is null");
7563
7564 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7565 OptoRuntime::poly1305_processBlocks_Type(),
7566 stubAddr, stubName, TypePtr::BOTTOM,
7567 input_start, len, acc_start, r_start);
7568 return true;
7569 }
7570
7571 //------------------------------inline_digestBase_implCompress-----------------------
7572 //
7573 // Calculate MD5 for single-block byte[] array.
7574 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7575 //
7576 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7577 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7578 //
7579 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7580 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7581 //
7582 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7583 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7584 //
7585 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7586 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7587 //
7588 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7589 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7590
7591 Node* digestBase_obj = argument(0);
7592 Node* src = argument(1); // type oop
7593 Node* ofs = argument(2); // type int
7594
7595 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7596 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7597 // failed array check
7598 return false;
7599 }
7600 // Figure out the size and type of the elements we will be copying.
7601 BasicType src_elem = src_type->elem()->array_element_basic_type();
7602 if (src_elem != T_BYTE) {
7603 return false;
7604 }
7605 // 'src_start' points to src array + offset
7606 src = must_be_not_null(src, true);
7607 Node* src_start = array_element_address(src, ofs, src_elem);
7608 Node* state = nullptr;
7609 Node* block_size = nullptr;
7610 address stubAddr;
7611 const char *stubName;
7612
7613 switch(id) {
7614 case vmIntrinsics::_md5_implCompress:
7615 assert(UseMD5Intrinsics, "need MD5 instruction support");
7616 state = get_state_from_digest_object(digestBase_obj, T_INT);
7617 stubAddr = StubRoutines::md5_implCompress();
7618 stubName = "md5_implCompress";
7619 break;
7620 case vmIntrinsics::_sha_implCompress:
7621 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7622 state = get_state_from_digest_object(digestBase_obj, T_INT);
7623 stubAddr = StubRoutines::sha1_implCompress();
7624 stubName = "sha1_implCompress";
7625 break;
7626 case vmIntrinsics::_sha2_implCompress:
7627 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7628 state = get_state_from_digest_object(digestBase_obj, T_INT);
7629 stubAddr = StubRoutines::sha256_implCompress();
7630 stubName = "sha256_implCompress";
7631 break;
7632 case vmIntrinsics::_sha5_implCompress:
7633 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7634 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7635 stubAddr = StubRoutines::sha512_implCompress();
7636 stubName = "sha512_implCompress";
7637 break;
7638 case vmIntrinsics::_sha3_implCompress:
7639 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7640 state = get_state_from_digest_object(digestBase_obj, T_BYTE);
7641 stubAddr = StubRoutines::sha3_implCompress();
7642 stubName = "sha3_implCompress";
7643 block_size = get_block_size_from_digest_object(digestBase_obj);
7644 if (block_size == nullptr) return false;
7645 break;
7646 default:
7647 fatal_unexpected_iid(id);
7648 return false;
7649 }
7650 if (state == nullptr) return false;
7651
7652 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7653 if (stubAddr == nullptr) return false;
7654
7655 // Call the stub.
7656 Node* call;
7657 if (block_size == nullptr) {
7658 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7659 stubAddr, stubName, TypePtr::BOTTOM,
7660 src_start, state);
7661 } else {
7662 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7663 stubAddr, stubName, TypePtr::BOTTOM,
7664 src_start, state, block_size);
7665 }
7666
7667 return true;
7668 }
7669
7670 //------------------------------inline_digestBase_implCompressMB-----------------------
7671 //
7672 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7673 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7674 //
7675 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7676 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7677 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7678 assert((uint)predicate < 5, "sanity");
7679 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7680
7681 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7682 Node* src = argument(1); // byte[] array
7683 Node* ofs = argument(2); // type int
7684 Node* limit = argument(3); // type int
7685
7686 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7687 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7688 // failed array check
7689 return false;
7690 }
7691 // Figure out the size and type of the elements we will be copying.
7692 BasicType src_elem = src_type->elem()->array_element_basic_type();
7693 if (src_elem != T_BYTE) {
7694 return false;
7695 }
7696 // 'src_start' points to src array + offset
7697 src = must_be_not_null(src, false);
7698 Node* src_start = array_element_address(src, ofs, src_elem);
7699
7700 const char* klass_digestBase_name = nullptr;
7701 const char* stub_name = nullptr;
7702 address stub_addr = nullptr;
7703 BasicType elem_type = T_INT;
7704
7705 switch (predicate) {
7706 case 0:
7707 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7708 klass_digestBase_name = "sun/security/provider/MD5";
7709 stub_name = "md5_implCompressMB";
7710 stub_addr = StubRoutines::md5_implCompressMB();
7711 }
7712 break;
7713 case 1:
7714 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7715 klass_digestBase_name = "sun/security/provider/SHA";
7716 stub_name = "sha1_implCompressMB";
7717 stub_addr = StubRoutines::sha1_implCompressMB();
7718 }
7719 break;
7720 case 2:
7721 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7722 klass_digestBase_name = "sun/security/provider/SHA2";
7723 stub_name = "sha256_implCompressMB";
7724 stub_addr = StubRoutines::sha256_implCompressMB();
7725 }
7726 break;
7727 case 3:
7728 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7729 klass_digestBase_name = "sun/security/provider/SHA5";
7730 stub_name = "sha512_implCompressMB";
7731 stub_addr = StubRoutines::sha512_implCompressMB();
7732 elem_type = T_LONG;
7733 }
7734 break;
7735 case 4:
7736 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7737 klass_digestBase_name = "sun/security/provider/SHA3";
7738 stub_name = "sha3_implCompressMB";
7739 stub_addr = StubRoutines::sha3_implCompressMB();
7740 elem_type = T_BYTE;
7741 }
7742 break;
7743 default:
7744 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7745 }
7746 if (klass_digestBase_name != nullptr) {
7747 assert(stub_addr != nullptr, "Stub is generated");
7748 if (stub_addr == nullptr) return false;
7749
7750 // get DigestBase klass to lookup for SHA klass
7751 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7752 assert(tinst != nullptr, "digestBase_obj is not instance???");
7753 assert(tinst->is_loaded(), "DigestBase is not loaded");
7754
7755 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7756 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7757 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7758 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7759 }
7760 return false;
7761 }
7762
7763 //------------------------------inline_digestBase_implCompressMB-----------------------
7764 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7765 BasicType elem_type, address stubAddr, const char *stubName,
7766 Node* src_start, Node* ofs, Node* limit) {
7767 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7768 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7769 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7770 digest_obj = _gvn.transform(digest_obj);
7771
7772 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7773 if (state == nullptr) return false;
7774
7775 Node* block_size = nullptr;
7776 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7777 block_size = get_block_size_from_digest_object(digest_obj);
7778 if (block_size == nullptr) return false;
7779 }
7780
7781 // Call the stub.
7782 Node* call;
7783 if (block_size == nullptr) {
7784 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7785 OptoRuntime::digestBase_implCompressMB_Type(false),
7786 stubAddr, stubName, TypePtr::BOTTOM,
7787 src_start, state, ofs, limit);
7788 } else {
7789 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7790 OptoRuntime::digestBase_implCompressMB_Type(true),
7791 stubAddr, stubName, TypePtr::BOTTOM,
7792 src_start, state, block_size, ofs, limit);
7793 }
7794
7795 // return ofs (int)
7796 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7797 set_result(result);
7798
7799 return true;
7800 }
7801
7802 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
7803 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
7804 assert(UseAES, "need AES instruction support");
7805 address stubAddr = nullptr;
7806 const char *stubName = nullptr;
7807 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
7808 stubName = "galoisCounterMode_AESCrypt";
7809
7810 if (stubAddr == nullptr) return false;
7811
7812 Node* in = argument(0);
7813 Node* inOfs = argument(1);
7814 Node* len = argument(2);
7815 Node* ct = argument(3);
7816 Node* ctOfs = argument(4);
7817 Node* out = argument(5);
7818 Node* outOfs = argument(6);
7819 Node* gctr_object = argument(7);
7820 Node* ghash_object = argument(8);
7821
7822 // (1) in, ct and out are arrays.
7823 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7824 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
7825 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7826 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
7827 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
7828 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
7829
7830 // checks are the responsibility of the caller
7831 Node* in_start = in;
7832 Node* ct_start = ct;
7833 Node* out_start = out;
7834 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
7835 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
7836 in_start = array_element_address(in, inOfs, T_BYTE);
7837 ct_start = array_element_address(ct, ctOfs, T_BYTE);
7838 out_start = array_element_address(out, outOfs, T_BYTE);
7839 }
7840
7841 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7842 // (because of the predicated logic executed earlier).
7843 // so we cast it here safely.
7844 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7845 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7846 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
7847 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
7848 Node* state = load_field_from_object(ghash_object, "state", "[J");
7849
7850 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
7851 return false;
7852 }
7853 // cast it to what we know it will be at runtime
7854 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
7855 assert(tinst != nullptr, "GCTR obj is null");
7856 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7857 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7858 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7859 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7860 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7861 const TypeOopPtr* xtype = aklass->as_instance_type();
7862 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7863 aescrypt_object = _gvn.transform(aescrypt_object);
7864 // we need to get the start of the aescrypt_object's expanded key array
7865 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7866 if (k_start == nullptr) return false;
7867 // similarly, get the start address of the r vector
7868 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
7869 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7870 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
7871
7872
7873 // Call the stub, passing params
7874 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7875 OptoRuntime::galoisCounterMode_aescrypt_Type(),
7876 stubAddr, stubName, TypePtr::BOTTOM,
7877 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
7878
7879 // return cipher length (int)
7880 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
7881 set_result(retvalue);
7882
7883 return true;
7884 }
7885
7886 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
7887 // Return node representing slow path of predicate check.
7888 // the pseudo code we want to emulate with this predicate is:
7889 // for encryption:
7890 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7891 // for decryption:
7892 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7893 // note cipher==plain is more conservative than the original java code but that's OK
7894 //
7895
7896 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
7897 // The receiver was checked for null already.
7898 Node* objGCTR = argument(7);
7899 // Load embeddedCipher field of GCTR object.
7900 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7901 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
7902
7903 // get AESCrypt klass for instanceOf check
7904 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7905 // will have same classloader as CipherBlockChaining object
7906 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
7907 assert(tinst != nullptr, "GCTR obj is null");
7908 assert(tinst->is_loaded(), "GCTR obj is not loaded");
7909
7910 // we want to do an instanceof comparison against the AESCrypt class
7911 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7912 if (!klass_AESCrypt->is_loaded()) {
7913 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7914 Node* ctrl = control();
7915 set_control(top()); // no regular fast path
7916 return ctrl;
7917 }
7918
7919 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7920 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7921 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7922 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7923 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7924
7925 return instof_false; // even if it is null
7926 }
7927
7928 //------------------------------get_state_from_digest_object-----------------------
7929 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
7930 const char* state_type;
7931 switch (elem_type) {
7932 case T_BYTE: state_type = "[B"; break;
7933 case T_INT: state_type = "[I"; break;
7934 case T_LONG: state_type = "[J"; break;
7935 default: ShouldNotReachHere();
7936 }
7937 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
7938 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
7939 if (digest_state == nullptr) return (Node *) nullptr;
7940
7941 // now have the array, need to get the start address of the state array
7942 Node* state = array_element_address(digest_state, intcon(0), elem_type);
7943 return state;
7944 }
7945
7946 //------------------------------get_block_size_from_sha3_object----------------------------------
7947 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
7948 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
7949 assert (block_size != nullptr, "sanity");
7950 return block_size;
7951 }
7952
7953 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
7954 // Return node representing slow path of predicate check.
7955 // the pseudo code we want to emulate with this predicate is:
7956 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
7957 //
7958 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
7959 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7960 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7961 assert((uint)predicate < 5, "sanity");
7962
7963 // The receiver was checked for null already.
7964 Node* digestBaseObj = argument(0);
7965
7966 // get DigestBase klass for instanceOf check
7967 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
7968 assert(tinst != nullptr, "digestBaseObj is null");
7969 assert(tinst->is_loaded(), "DigestBase is not loaded");
7970
7971 const char* klass_name = nullptr;
7972 switch (predicate) {
7973 case 0:
7974 if (UseMD5Intrinsics) {
7975 // we want to do an instanceof comparison against the MD5 class
7976 klass_name = "sun/security/provider/MD5";
7977 }
7978 break;
7979 case 1:
7980 if (UseSHA1Intrinsics) {
7981 // we want to do an instanceof comparison against the SHA class
7982 klass_name = "sun/security/provider/SHA";
7983 }
7984 break;
7985 case 2:
7986 if (UseSHA256Intrinsics) {
7987 // we want to do an instanceof comparison against the SHA2 class
7988 klass_name = "sun/security/provider/SHA2";
7989 }
7990 break;
7991 case 3:
7992 if (UseSHA512Intrinsics) {
7993 // we want to do an instanceof comparison against the SHA5 class
7994 klass_name = "sun/security/provider/SHA5";
7995 }
7996 break;
7997 case 4:
7998 if (UseSHA3Intrinsics) {
7999 // we want to do an instanceof comparison against the SHA3 class
8000 klass_name = "sun/security/provider/SHA3";
8001 }
8002 break;
8003 default:
8004 fatal("unknown SHA intrinsic predicate: %d", predicate);
8005 }
8006
8007 ciKlass* klass = nullptr;
8008 if (klass_name != nullptr) {
8009 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8010 }
8011 if ((klass == nullptr) || !klass->is_loaded()) {
8012 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8013 Node* ctrl = control();
8014 set_control(top()); // no intrinsic path
8015 return ctrl;
8016 }
8017 ciInstanceKlass* instklass = klass->as_instance_klass();
8018
8019 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8020 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8021 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8022 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8023
8024 return instof_false; // even if it is null
8025 }
8026
8027 //-------------inline_fma-----------------------------------
8028 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8029 Node *a = nullptr;
8030 Node *b = nullptr;
8031 Node *c = nullptr;
8032 Node* result = nullptr;
8033 switch (id) {
8034 case vmIntrinsics::_fmaD:
8035 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8036 // no receiver since it is static method
8037 a = round_double_node(argument(0));
8038 b = round_double_node(argument(2));
8039 c = round_double_node(argument(4));
8040 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8041 break;
8042 case vmIntrinsics::_fmaF:
8043 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8044 a = argument(0);
8045 b = argument(1);
8046 c = argument(2);
8047 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8048 break;
8049 default:
8050 fatal_unexpected_iid(id); break;
8051 }
8052 set_result(result);
8053 return true;
8054 }
8055
8056 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8057 // argument(0) is receiver
8058 Node* codePoint = argument(1);
8059 Node* n = nullptr;
8060
8061 switch (id) {
8062 case vmIntrinsics::_isDigit :
8063 n = new DigitNode(control(), codePoint);
8064 break;
8065 case vmIntrinsics::_isLowerCase :
8066 n = new LowerCaseNode(control(), codePoint);
8067 break;
8068 case vmIntrinsics::_isUpperCase :
8069 n = new UpperCaseNode(control(), codePoint);
8070 break;
8071 case vmIntrinsics::_isWhitespace :
8072 n = new WhitespaceNode(control(), codePoint);
8073 break;
8074 default:
8075 fatal_unexpected_iid(id);
8076 }
8077
8078 set_result(_gvn.transform(n));
8079 return true;
8080 }
8081
8082 //------------------------------inline_fp_min_max------------------------------
8083 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8084 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8085
8086 // The intrinsic should be used only when the API branches aren't predictable,
8087 // the last one performing the most important comparison. The following heuristic
8088 // uses the branch statistics to eventually bail out if necessary.
8089
8090 ciMethodData *md = callee()->method_data();
8091
8092 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8093 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8094
8095 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8096 // Bail out if the call-site didn't contribute enough to the statistics.
8097 return false;
8098 }
8099
8100 uint taken = 0, not_taken = 0;
8101
8102 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8103 if (p->is_BranchData()) {
8104 taken = ((ciBranchData*)p)->taken();
8105 not_taken = ((ciBranchData*)p)->not_taken();
8106 }
8107 }
8108
8109 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8110 balance = balance < 0 ? -balance : balance;
8111 if ( balance > 0.2 ) {
8112 // Bail out if the most important branch is predictable enough.
8113 return false;
8114 }
8115 }
8116 */
8117
8118 Node *a = nullptr;
8119 Node *b = nullptr;
8120 Node *n = nullptr;
8121 switch (id) {
8122 case vmIntrinsics::_maxF:
8123 case vmIntrinsics::_minF:
8124 case vmIntrinsics::_maxF_strict:
8125 case vmIntrinsics::_minF_strict:
8126 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8127 a = argument(0);
8128 b = argument(1);
8129 break;
8130 case vmIntrinsics::_maxD:
8131 case vmIntrinsics::_minD:
8132 case vmIntrinsics::_maxD_strict:
8133 case vmIntrinsics::_minD_strict:
8134 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8135 a = round_double_node(argument(0));
8136 b = round_double_node(argument(2));
8137 break;
8138 default:
8139 fatal_unexpected_iid(id);
8140 break;
8141 }
8142 switch (id) {
8143 case vmIntrinsics::_maxF:
8144 case vmIntrinsics::_maxF_strict:
8145 n = new MaxFNode(a, b);
8146 break;
8147 case vmIntrinsics::_minF:
8148 case vmIntrinsics::_minF_strict:
8149 n = new MinFNode(a, b);
8150 break;
8151 case vmIntrinsics::_maxD:
8152 case vmIntrinsics::_maxD_strict:
8153 n = new MaxDNode(a, b);
8154 break;
8155 case vmIntrinsics::_minD:
8156 case vmIntrinsics::_minD_strict:
8157 n = new MinDNode(a, b);
8158 break;
8159 default:
8160 fatal_unexpected_iid(id);
8161 break;
8162 }
8163 set_result(_gvn.transform(n));
8164 return true;
8165 }
8166
8167 bool LibraryCallKit::inline_profileBoolean() {
8168 Node* counts = argument(1);
8169 const TypeAryPtr* ary = nullptr;
8170 ciArray* aobj = nullptr;
8171 if (counts->is_Con()
8172 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8173 && (aobj = ary->const_oop()->as_array()) != nullptr
8174 && (aobj->length() == 2)) {
8175 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8176 jint false_cnt = aobj->element_value(0).as_int();
8177 jint true_cnt = aobj->element_value(1).as_int();
8178
8179 if (C->log() != nullptr) {
8180 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8181 false_cnt, true_cnt);
8182 }
8183
8184 if (false_cnt + true_cnt == 0) {
8185 // According to profile, never executed.
8186 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8187 Deoptimization::Action_reinterpret);
8188 return true;
8189 }
8190
8191 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8192 // is a number of each value occurrences.
8193 Node* result = argument(0);
8194 if (false_cnt == 0 || true_cnt == 0) {
8195 // According to profile, one value has been never seen.
8196 int expected_val = (false_cnt == 0) ? 1 : 0;
8197
8198 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8199 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8200
8201 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8202 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8203 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8204
8205 { // Slow path: uncommon trap for never seen value and then reexecute
8206 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8207 // the value has been seen at least once.
8208 PreserveJVMState pjvms(this);
8209 PreserveReexecuteState preexecs(this);
8210 jvms()->set_should_reexecute(true);
8211
8212 set_control(slow_path);
8213 set_i_o(i_o());
8214
8215 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8216 Deoptimization::Action_reinterpret);
8217 }
8218 // The guard for never seen value enables sharpening of the result and
8219 // returning a constant. It allows to eliminate branches on the same value
8220 // later on.
8221 set_control(fast_path);
8222 result = intcon(expected_val);
8223 }
8224 // Stop profiling.
8225 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8226 // By replacing method body with profile data (represented as ProfileBooleanNode
8227 // on IR level) we effectively disable profiling.
8228 // It enables full speed execution once optimized code is generated.
8229 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8230 C->record_for_igvn(profile);
8231 set_result(profile);
8232 return true;
8233 } else {
8234 // Continue profiling.
8235 // Profile data isn't available at the moment. So, execute method's bytecode version.
8236 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8237 // is compiled and counters aren't available since corresponding MethodHandle
8238 // isn't a compile-time constant.
8239 return false;
8240 }
8241 }
8242
8243 bool LibraryCallKit::inline_isCompileConstant() {
8244 Node* n = argument(0);
8245 set_result(n->is_Con() ? intcon(1) : intcon(0));
8246 return true;
8247 }
8248
8249 //------------------------------- inline_getObjectSize --------------------------------------
8250 //
8251 // Calculate the runtime size of the object/array.
8252 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8253 //
8254 bool LibraryCallKit::inline_getObjectSize() {
8255 Node* obj = argument(3);
8256 Node* klass_node = load_object_klass(obj);
8257
8258 jint layout_con = Klass::_lh_neutral_value;
8259 Node* layout_val = get_layout_helper(klass_node, layout_con);
8260 int layout_is_con = (layout_val == nullptr);
8261
8262 if (layout_is_con) {
8263 // Layout helper is constant, can figure out things at compile time.
8264
8265 if (Klass::layout_helper_is_instance(layout_con)) {
8266 // Instance case: layout_con contains the size itself.
8267 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8268 set_result(size);
8269 } else {
8270 // Array case: size is round(header + element_size*arraylength).
8271 // Since arraylength is different for every array instance, we have to
8272 // compute the whole thing at runtime.
8273
8274 Node* arr_length = load_array_length(obj);
8275
8276 int round_mask = MinObjAlignmentInBytes - 1;
8277 int hsize = Klass::layout_helper_header_size(layout_con);
8278 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8279
8280 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8281 round_mask = 0; // strength-reduce it if it goes away completely
8282 }
8283 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8284 Node* header_size = intcon(hsize + round_mask);
8285
8286 Node* lengthx = ConvI2X(arr_length);
8287 Node* headerx = ConvI2X(header_size);
8288
8289 Node* abody = lengthx;
8290 if (eshift != 0) {
8291 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8292 }
8293 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8294 if (round_mask != 0) {
8295 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8296 }
8297 size = ConvX2L(size);
8298 set_result(size);
8299 }
8300 } else {
8301 // Layout helper is not constant, need to test for array-ness at runtime.
8302
8303 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8304 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8305 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8306 record_for_igvn(result_reg);
8307
8308 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8309 if (array_ctl != nullptr) {
8310 // Array case: size is round(header + element_size*arraylength).
8311 // Since arraylength is different for every array instance, we have to
8312 // compute the whole thing at runtime.
8313
8314 PreserveJVMState pjvms(this);
8315 set_control(array_ctl);
8316 Node* arr_length = load_array_length(obj);
8317
8318 int round_mask = MinObjAlignmentInBytes - 1;
8319 Node* mask = intcon(round_mask);
8320
8321 Node* hss = intcon(Klass::_lh_header_size_shift);
8322 Node* hsm = intcon(Klass::_lh_header_size_mask);
8323 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8324 header_size = _gvn.transform(new AndINode(header_size, hsm));
8325 header_size = _gvn.transform(new AddINode(header_size, mask));
8326
8327 // There is no need to mask or shift this value.
8328 // The semantics of LShiftINode include an implicit mask to 0x1F.
8329 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8330 Node* elem_shift = layout_val;
8331
8332 Node* lengthx = ConvI2X(arr_length);
8333 Node* headerx = ConvI2X(header_size);
8334
8335 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8336 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8337 if (round_mask != 0) {
8338 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8339 }
8340 size = ConvX2L(size);
8341
8342 result_reg->init_req(_array_path, control());
8343 result_val->init_req(_array_path, size);
8344 }
8345
8346 if (!stopped()) {
8347 // Instance case: the layout helper gives us instance size almost directly,
8348 // but we need to mask out the _lh_instance_slow_path_bit.
8349 Node* size = ConvI2X(layout_val);
8350 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8351 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8352 size = _gvn.transform(new AndXNode(size, mask));
8353 size = ConvX2L(size);
8354
8355 result_reg->init_req(_instance_path, control());
8356 result_val->init_req(_instance_path, size);
8357 }
8358
8359 set_result(result_reg, result_val);
8360 }
8361
8362 return true;
8363 }
8364
8365 //------------------------------- inline_blackhole --------------------------------------
8366 //
8367 // Make sure all arguments to this node are alive.
8368 // This matches methods that were requested to be blackholed through compile commands.
8369 //
8370 bool LibraryCallKit::inline_blackhole() {
8371 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8372 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8373 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8374
8375 // Blackhole node pinches only the control, not memory. This allows
8376 // the blackhole to be pinned in the loop that computes blackholed
8377 // values, but have no other side effects, like breaking the optimizations
8378 // across the blackhole.
8379
8380 Node* bh = _gvn.transform(new BlackholeNode(control()));
8381 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8382
8383 // Bind call arguments as blackhole arguments to keep them alive
8384 uint nargs = callee()->arg_size();
8385 for (uint i = 0; i < nargs; i++) {
8386 bh->add_req(argument(i));
8387 }
8388
8389 return true;
8390 }