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::_dtanh:
258 case vmIntrinsics::_dabs:
259 case vmIntrinsics::_fabs:
260 case vmIntrinsics::_iabs:
261 case vmIntrinsics::_labs:
262 case vmIntrinsics::_datan2:
263 case vmIntrinsics::_dsqrt:
264 case vmIntrinsics::_dsqrt_strict:
265 case vmIntrinsics::_dexp:
266 case vmIntrinsics::_dlog:
267 case vmIntrinsics::_dlog10:
268 case vmIntrinsics::_dpow:
269 case vmIntrinsics::_dcopySign:
270 case vmIntrinsics::_fcopySign:
271 case vmIntrinsics::_dsignum:
272 case vmIntrinsics::_roundF:
273 case vmIntrinsics::_roundD:
274 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
275
276 case vmIntrinsics::_notify:
277 case vmIntrinsics::_notifyAll:
278 return inline_notify(intrinsic_id());
279
280 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
281 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
282 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
283 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
284 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
285 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
286 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
287 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
288 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
289 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
290 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
291 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
292 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
293 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
294
295 case vmIntrinsics::_arraycopy: return inline_arraycopy();
296
297 case vmIntrinsics::_arraySort: return inline_array_sort();
298 case vmIntrinsics::_arrayPartition: return inline_array_partition();
299
300 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
301 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
302 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
303 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
304
305 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
306 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
307 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
308 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
309 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
310 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
311 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
312 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
313
314 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
315
316 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
317
318 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
319 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
320 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
321 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
322
323 case vmIntrinsics::_compressStringC:
324 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
325 case vmIntrinsics::_inflateStringC:
326 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
327
328 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
339 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
340 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
341 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
342 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
343 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
344 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
345 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
346 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
347
348 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
359 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
360 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
361 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
362 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
363 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
364 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
365 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
366 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
367
368 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
369 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
370 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
371 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
372
373 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
374 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
375 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
376 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
377
378 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
379 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
380 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
381 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
382 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
383 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
384 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
385 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
386 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
387
388 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
389 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
390 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
391 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
392 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
393 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
394 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
395 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
396 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
397
398 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
409 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
410 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
411 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
412 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
413 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
414 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
415 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
416 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
417
418 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
419 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
420 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
421 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
422 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
423
424 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
425 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
426 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
427 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
428 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
429 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
430 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
431 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
432 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
433 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
434 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
435 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
436 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
437 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
438 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
439 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
440 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
441 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
442 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
443 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
444
445 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
454 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
455 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
456 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
457 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
458 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
459 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
460
461 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
462 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
463 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
464 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
465
466 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
467 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
468 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
469 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
470 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
471
472 case vmIntrinsics::_loadFence:
473 case vmIntrinsics::_storeFence:
474 case vmIntrinsics::_storeStoreFence:
475 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
476
477 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
478
479 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
480 case vmIntrinsics::_currentThread: return inline_native_currentThread();
481 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
482
483 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
484 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
485
486 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
487 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
488
489 #if INCLUDE_JVMTI
490 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
491 "notifyJvmtiStart", true, false);
492 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
493 "notifyJvmtiEnd", false, true);
494 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
495 "notifyJvmtiMount", false, false);
496 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
497 "notifyJvmtiUnmount", false, false);
498 case vmIntrinsics::_notifyJvmtiVThreadHideFrames: return inline_native_notify_jvmti_hide();
499 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
500 #endif
501
502 #ifdef JFR_HAVE_INTRINSICS
503 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
504 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
505 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
506 #endif
507 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
508 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
509 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
510 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
511 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
512 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
513 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
514 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
515 case vmIntrinsics::_getLength: return inline_native_getLength();
516 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
517 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
518 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
519 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
520 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
521 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
522 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
523
524 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
525 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
526
527 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
528
529 case vmIntrinsics::_isInstance:
530 case vmIntrinsics::_getModifiers:
531 case vmIntrinsics::_isInterface:
532 case vmIntrinsics::_isArray:
533 case vmIntrinsics::_isPrimitive:
534 case vmIntrinsics::_isHidden:
535 case vmIntrinsics::_getSuperclass:
536 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
537
538 case vmIntrinsics::_floatToRawIntBits:
539 case vmIntrinsics::_floatToIntBits:
540 case vmIntrinsics::_intBitsToFloat:
541 case vmIntrinsics::_doubleToRawLongBits:
542 case vmIntrinsics::_doubleToLongBits:
543 case vmIntrinsics::_longBitsToDouble:
544 case vmIntrinsics::_floatToFloat16:
545 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
546
547 case vmIntrinsics::_floatIsFinite:
548 case vmIntrinsics::_floatIsInfinite:
549 case vmIntrinsics::_doubleIsFinite:
550 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
551
552 case vmIntrinsics::_numberOfLeadingZeros_i:
553 case vmIntrinsics::_numberOfLeadingZeros_l:
554 case vmIntrinsics::_numberOfTrailingZeros_i:
555 case vmIntrinsics::_numberOfTrailingZeros_l:
556 case vmIntrinsics::_bitCount_i:
557 case vmIntrinsics::_bitCount_l:
558 case vmIntrinsics::_reverse_i:
559 case vmIntrinsics::_reverse_l:
560 case vmIntrinsics::_reverseBytes_i:
561 case vmIntrinsics::_reverseBytes_l:
562 case vmIntrinsics::_reverseBytes_s:
563 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
564
565 case vmIntrinsics::_compress_i:
566 case vmIntrinsics::_compress_l:
567 case vmIntrinsics::_expand_i:
568 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
569
570 case vmIntrinsics::_compareUnsigned_i:
571 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
572
573 case vmIntrinsics::_divideUnsigned_i:
574 case vmIntrinsics::_divideUnsigned_l:
575 case vmIntrinsics::_remainderUnsigned_i:
576 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
577
578 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
579
580 case vmIntrinsics::_Reference_get: return inline_reference_get();
581 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
582 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
583
584 case vmIntrinsics::_Class_cast: return inline_Class_cast();
585
586 case vmIntrinsics::_aescrypt_encryptBlock:
587 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
588
589 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
590 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
591 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
592
593 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
594 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
595 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
596
597 case vmIntrinsics::_counterMode_AESCrypt:
598 return inline_counterMode_AESCrypt(intrinsic_id());
599
600 case vmIntrinsics::_galoisCounterMode_AESCrypt:
601 return inline_galoisCounterMode_AESCrypt();
602
603 case vmIntrinsics::_md5_implCompress:
604 case vmIntrinsics::_sha_implCompress:
605 case vmIntrinsics::_sha2_implCompress:
606 case vmIntrinsics::_sha5_implCompress:
607 case vmIntrinsics::_sha3_implCompress:
608 return inline_digestBase_implCompress(intrinsic_id());
609
610 case vmIntrinsics::_digestBase_implCompressMB:
611 return inline_digestBase_implCompressMB(predicate);
612
613 case vmIntrinsics::_multiplyToLen:
614 return inline_multiplyToLen();
615
616 case vmIntrinsics::_squareToLen:
617 return inline_squareToLen();
618
619 case vmIntrinsics::_mulAdd:
620 return inline_mulAdd();
621
622 case vmIntrinsics::_montgomeryMultiply:
623 return inline_montgomeryMultiply();
624 case vmIntrinsics::_montgomerySquare:
625 return inline_montgomerySquare();
626
627 case vmIntrinsics::_bigIntegerRightShiftWorker:
628 return inline_bigIntegerShift(true);
629 case vmIntrinsics::_bigIntegerLeftShiftWorker:
630 return inline_bigIntegerShift(false);
631
632 case vmIntrinsics::_vectorizedMismatch:
633 return inline_vectorizedMismatch();
634
635 case vmIntrinsics::_ghash_processBlocks:
636 return inline_ghash_processBlocks();
637 case vmIntrinsics::_chacha20Block:
638 return inline_chacha20Block();
639 case vmIntrinsics::_base64_encodeBlock:
640 return inline_base64_encodeBlock();
641 case vmIntrinsics::_base64_decodeBlock:
642 return inline_base64_decodeBlock();
643 case vmIntrinsics::_poly1305_processBlocks:
644 return inline_poly1305_processBlocks();
645 case vmIntrinsics::_intpoly_montgomeryMult_P256:
646 return inline_intpoly_montgomeryMult_P256();
647 case vmIntrinsics::_intpoly_assign:
648 return inline_intpoly_assign();
649 case vmIntrinsics::_encodeISOArray:
650 case vmIntrinsics::_encodeByteISOArray:
651 return inline_encodeISOArray(false);
652 case vmIntrinsics::_encodeAsciiArray:
653 return inline_encodeISOArray(true);
654
655 case vmIntrinsics::_updateCRC32:
656 return inline_updateCRC32();
657 case vmIntrinsics::_updateBytesCRC32:
658 return inline_updateBytesCRC32();
659 case vmIntrinsics::_updateByteBufferCRC32:
660 return inline_updateByteBufferCRC32();
661
662 case vmIntrinsics::_updateBytesCRC32C:
663 return inline_updateBytesCRC32C();
664 case vmIntrinsics::_updateDirectByteBufferCRC32C:
665 return inline_updateDirectByteBufferCRC32C();
666
667 case vmIntrinsics::_updateBytesAdler32:
668 return inline_updateBytesAdler32();
669 case vmIntrinsics::_updateByteBufferAdler32:
670 return inline_updateByteBufferAdler32();
671
672 case vmIntrinsics::_profileBoolean:
673 return inline_profileBoolean();
674 case vmIntrinsics::_isCompileConstant:
675 return inline_isCompileConstant();
676
677 case vmIntrinsics::_countPositives:
678 return inline_countPositives();
679
680 case vmIntrinsics::_fmaD:
681 case vmIntrinsics::_fmaF:
682 return inline_fma(intrinsic_id());
683
684 case vmIntrinsics::_isDigit:
685 case vmIntrinsics::_isLowerCase:
686 case vmIntrinsics::_isUpperCase:
687 case vmIntrinsics::_isWhitespace:
688 return inline_character_compare(intrinsic_id());
689
690 case vmIntrinsics::_min:
691 case vmIntrinsics::_max:
692 case vmIntrinsics::_min_strict:
693 case vmIntrinsics::_max_strict:
694 return inline_min_max(intrinsic_id());
695
696 case vmIntrinsics::_maxF:
697 case vmIntrinsics::_minF:
698 case vmIntrinsics::_maxD:
699 case vmIntrinsics::_minD:
700 case vmIntrinsics::_maxF_strict:
701 case vmIntrinsics::_minF_strict:
702 case vmIntrinsics::_maxD_strict:
703 case vmIntrinsics::_minD_strict:
704 return inline_fp_min_max(intrinsic_id());
705
706 case vmIntrinsics::_VectorUnaryOp:
707 return inline_vector_nary_operation(1);
708 case vmIntrinsics::_VectorBinaryOp:
709 return inline_vector_nary_operation(2);
710 case vmIntrinsics::_VectorTernaryOp:
711 return inline_vector_nary_operation(3);
712 case vmIntrinsics::_VectorFromBitsCoerced:
713 return inline_vector_frombits_coerced();
714 case vmIntrinsics::_VectorShuffleIota:
715 return inline_vector_shuffle_iota();
716 case vmIntrinsics::_VectorMaskOp:
717 return inline_vector_mask_operation();
718 case vmIntrinsics::_VectorShuffleToVector:
719 return inline_vector_shuffle_to_vector();
720 case vmIntrinsics::_VectorWrapShuffleIndexes:
721 return inline_vector_wrap_shuffle_indexes();
722 case vmIntrinsics::_VectorLoadOp:
723 return inline_vector_mem_operation(/*is_store=*/false);
724 case vmIntrinsics::_VectorLoadMaskedOp:
725 return inline_vector_mem_masked_operation(/*is_store*/false);
726 case vmIntrinsics::_VectorStoreOp:
727 return inline_vector_mem_operation(/*is_store=*/true);
728 case vmIntrinsics::_VectorStoreMaskedOp:
729 return inline_vector_mem_masked_operation(/*is_store=*/true);
730 case vmIntrinsics::_VectorGatherOp:
731 return inline_vector_gather_scatter(/*is_scatter*/ false);
732 case vmIntrinsics::_VectorScatterOp:
733 return inline_vector_gather_scatter(/*is_scatter*/ true);
734 case vmIntrinsics::_VectorReductionCoerced:
735 return inline_vector_reduction();
736 case vmIntrinsics::_VectorTest:
737 return inline_vector_test();
738 case vmIntrinsics::_VectorBlend:
739 return inline_vector_blend();
740 case vmIntrinsics::_VectorRearrange:
741 return inline_vector_rearrange();
742 case vmIntrinsics::_VectorSelectFrom:
743 return inline_vector_select_from();
744 case vmIntrinsics::_VectorCompare:
745 return inline_vector_compare();
746 case vmIntrinsics::_VectorBroadcastInt:
747 return inline_vector_broadcast_int();
748 case vmIntrinsics::_VectorConvert:
749 return inline_vector_convert();
750 case vmIntrinsics::_VectorInsert:
751 return inline_vector_insert();
752 case vmIntrinsics::_VectorExtract:
753 return inline_vector_extract();
754 case vmIntrinsics::_VectorCompressExpand:
755 return inline_vector_compress_expand();
756 case vmIntrinsics::_IndexVector:
757 return inline_index_vector();
758 case vmIntrinsics::_IndexPartiallyInUpperRange:
759 return inline_index_partially_in_upper_range();
760
761 case vmIntrinsics::_getObjectSize:
762 return inline_getObjectSize();
763
764 case vmIntrinsics::_blackhole:
765 return inline_blackhole();
766
767 default:
768 // If you get here, it may be that someone has added a new intrinsic
769 // to the list in vmIntrinsics.hpp without implementing it here.
770 #ifndef PRODUCT
771 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
772 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
773 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
774 }
775 #endif
776 return false;
777 }
778 }
779
780 Node* LibraryCallKit::try_to_predicate(int predicate) {
781 if (!jvms()->has_method()) {
782 // Root JVMState has a null method.
783 assert(map()->memory()->Opcode() == Op_Parm, "");
784 // Insert the memory aliasing node
785 set_all_memory(reset_memory());
786 }
787 assert(merged_memory(), "");
788
789 switch (intrinsic_id()) {
790 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
791 return inline_cipherBlockChaining_AESCrypt_predicate(false);
792 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
793 return inline_cipherBlockChaining_AESCrypt_predicate(true);
794 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
795 return inline_electronicCodeBook_AESCrypt_predicate(false);
796 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
797 return inline_electronicCodeBook_AESCrypt_predicate(true);
798 case vmIntrinsics::_counterMode_AESCrypt:
799 return inline_counterMode_AESCrypt_predicate();
800 case vmIntrinsics::_digestBase_implCompressMB:
801 return inline_digestBase_implCompressMB_predicate(predicate);
802 case vmIntrinsics::_galoisCounterMode_AESCrypt:
803 return inline_galoisCounterMode_AESCrypt_predicate();
804
805 default:
806 // If you get here, it may be that someone has added a new intrinsic
807 // to the list in vmIntrinsics.hpp without implementing it here.
808 #ifndef PRODUCT
809 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
810 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
811 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
812 }
813 #endif
814 Node* slow_ctl = control();
815 set_control(top()); // No fast path intrinsic
816 return slow_ctl;
817 }
818 }
819
820 //------------------------------set_result-------------------------------
821 // Helper function for finishing intrinsics.
822 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
823 record_for_igvn(region);
824 set_control(_gvn.transform(region));
825 set_result( _gvn.transform(value));
826 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
827 }
828
829 //------------------------------generate_guard---------------------------
830 // Helper function for generating guarded fast-slow graph structures.
831 // The given 'test', if true, guards a slow path. If the test fails
832 // then a fast path can be taken. (We generally hope it fails.)
833 // In all cases, GraphKit::control() is updated to the fast path.
834 // The returned value represents the control for the slow path.
835 // The return value is never 'top'; it is either a valid control
836 // or null if it is obvious that the slow path can never be taken.
837 // Also, if region and the slow control are not null, the slow edge
838 // is appended to the region.
839 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
840 if (stopped()) {
841 // Already short circuited.
842 return nullptr;
843 }
844
845 // Build an if node and its projections.
846 // If test is true we take the slow path, which we assume is uncommon.
847 if (_gvn.type(test) == TypeInt::ZERO) {
848 // The slow branch is never taken. No need to build this guard.
849 return nullptr;
850 }
851
852 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
853
854 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
855 if (if_slow == top()) {
856 // The slow branch is never taken. No need to build this guard.
857 return nullptr;
858 }
859
860 if (region != nullptr)
861 region->add_req(if_slow);
862
863 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
864 set_control(if_fast);
865
866 return if_slow;
867 }
868
869 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
870 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
871 }
872 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
873 return generate_guard(test, region, PROB_FAIR);
874 }
875
876 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
877 Node* *pos_index) {
878 if (stopped())
879 return nullptr; // already stopped
880 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
881 return nullptr; // index is already adequately typed
882 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
883 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
884 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
885 if (is_neg != nullptr && pos_index != nullptr) {
886 // Emulate effect of Parse::adjust_map_after_if.
887 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
888 (*pos_index) = _gvn.transform(ccast);
889 }
890 return is_neg;
891 }
892
893 // Make sure that 'position' is a valid limit index, in [0..length].
894 // There are two equivalent plans for checking this:
895 // A. (offset + copyLength) unsigned<= arrayLength
896 // B. offset <= (arrayLength - copyLength)
897 // We require that all of the values above, except for the sum and
898 // difference, are already known to be non-negative.
899 // Plan A is robust in the face of overflow, if offset and copyLength
900 // are both hugely positive.
901 //
902 // Plan B is less direct and intuitive, but it does not overflow at
903 // all, since the difference of two non-negatives is always
904 // representable. Whenever Java methods must perform the equivalent
905 // check they generally use Plan B instead of Plan A.
906 // For the moment we use Plan A.
907 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
908 Node* subseq_length,
909 Node* array_length,
910 RegionNode* region) {
911 if (stopped())
912 return nullptr; // already stopped
913 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
914 if (zero_offset && subseq_length->eqv_uncast(array_length))
915 return nullptr; // common case of whole-array copy
916 Node* last = subseq_length;
917 if (!zero_offset) // last += offset
918 last = _gvn.transform(new AddINode(last, offset));
919 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
920 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
921 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
922 return is_over;
923 }
924
925 // Emit range checks for the given String.value byte array
926 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
927 if (stopped()) {
928 return; // already stopped
929 }
930 RegionNode* bailout = new RegionNode(1);
931 record_for_igvn(bailout);
932 if (char_count) {
933 // Convert char count to byte count
934 count = _gvn.transform(new LShiftINode(count, intcon(1)));
935 }
936
937 // Offset and count must not be negative
938 generate_negative_guard(offset, bailout);
939 generate_negative_guard(count, bailout);
940 // Offset + count must not exceed length of array
941 generate_limit_guard(offset, count, load_array_length(array), bailout);
942
943 if (bailout->req() > 1) {
944 PreserveJVMState pjvms(this);
945 set_control(_gvn.transform(bailout));
946 uncommon_trap(Deoptimization::Reason_intrinsic,
947 Deoptimization::Action_maybe_recompile);
948 }
949 }
950
951 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
952 bool is_immutable) {
953 ciKlass* thread_klass = env()->Thread_klass();
954 const Type* thread_type
955 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
956
957 Node* thread = _gvn.transform(new ThreadLocalNode());
958 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
959 tls_output = thread;
960
961 Node* thread_obj_handle
962 = (is_immutable
963 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
964 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
965 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
966 thread_obj_handle = _gvn.transform(thread_obj_handle);
967
968 DecoratorSet decorators = IN_NATIVE;
969 if (is_immutable) {
970 decorators |= C2_IMMUTABLE_MEMORY;
971 }
972 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
973 }
974
975 //--------------------------generate_current_thread--------------------
976 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
977 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
978 /*is_immutable*/false);
979 }
980
981 //--------------------------generate_virtual_thread--------------------
982 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
983 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
984 !C->method()->changes_current_thread());
985 }
986
987 //------------------------------make_string_method_node------------------------
988 // Helper method for String intrinsic functions. This version is called with
989 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
990 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
991 // containing the lengths of str1 and str2.
992 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
993 Node* result = nullptr;
994 switch (opcode) {
995 case Op_StrIndexOf:
996 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
997 str1_start, cnt1, str2_start, cnt2, ae);
998 break;
999 case Op_StrComp:
1000 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1001 str1_start, cnt1, str2_start, cnt2, ae);
1002 break;
1003 case Op_StrEquals:
1004 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1005 // Use the constant length if there is one because optimized match rule may exist.
1006 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1007 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1008 break;
1009 default:
1010 ShouldNotReachHere();
1011 return nullptr;
1012 }
1013
1014 // All these intrinsics have checks.
1015 C->set_has_split_ifs(true); // Has chance for split-if optimization
1016 clear_upper_avx();
1017
1018 return _gvn.transform(result);
1019 }
1020
1021 //------------------------------inline_string_compareTo------------------------
1022 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1023 Node* arg1 = argument(0);
1024 Node* arg2 = argument(1);
1025
1026 arg1 = must_be_not_null(arg1, true);
1027 arg2 = must_be_not_null(arg2, true);
1028
1029 // Get start addr and length of first argument
1030 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1031 Node* arg1_cnt = load_array_length(arg1);
1032
1033 // Get start addr and length of second argument
1034 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1035 Node* arg2_cnt = load_array_length(arg2);
1036
1037 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1038 set_result(result);
1039 return true;
1040 }
1041
1042 //------------------------------inline_string_equals------------------------
1043 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1044 Node* arg1 = argument(0);
1045 Node* arg2 = argument(1);
1046
1047 // paths (plus control) merge
1048 RegionNode* region = new RegionNode(3);
1049 Node* phi = new PhiNode(region, TypeInt::BOOL);
1050
1051 if (!stopped()) {
1052
1053 arg1 = must_be_not_null(arg1, true);
1054 arg2 = must_be_not_null(arg2, true);
1055
1056 // Get start addr and length of first argument
1057 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1058 Node* arg1_cnt = load_array_length(arg1);
1059
1060 // Get start addr and length of second argument
1061 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1062 Node* arg2_cnt = load_array_length(arg2);
1063
1064 // Check for arg1_cnt != arg2_cnt
1065 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1066 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1067 Node* if_ne = generate_slow_guard(bol, nullptr);
1068 if (if_ne != nullptr) {
1069 phi->init_req(2, intcon(0));
1070 region->init_req(2, if_ne);
1071 }
1072
1073 // Check for count == 0 is done by assembler code for StrEquals.
1074
1075 if (!stopped()) {
1076 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1077 phi->init_req(1, equals);
1078 region->init_req(1, control());
1079 }
1080 }
1081
1082 // post merge
1083 set_control(_gvn.transform(region));
1084 record_for_igvn(region);
1085
1086 set_result(_gvn.transform(phi));
1087 return true;
1088 }
1089
1090 //------------------------------inline_array_equals----------------------------
1091 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1092 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1093 Node* arg1 = argument(0);
1094 Node* arg2 = argument(1);
1095
1096 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1097 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1098 clear_upper_avx();
1099
1100 return true;
1101 }
1102
1103
1104 //------------------------------inline_countPositives------------------------------
1105 bool LibraryCallKit::inline_countPositives() {
1106 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1107 return false;
1108 }
1109
1110 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1111 // no receiver since it is static method
1112 Node* ba = argument(0);
1113 Node* offset = argument(1);
1114 Node* len = argument(2);
1115
1116 ba = must_be_not_null(ba, true);
1117
1118 // Range checks
1119 generate_string_range_check(ba, offset, len, false);
1120 if (stopped()) {
1121 return true;
1122 }
1123 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1124 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1125 set_result(_gvn.transform(result));
1126 clear_upper_avx();
1127 return true;
1128 }
1129
1130 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1131 Node* index = argument(0);
1132 Node* length = bt == T_INT ? argument(1) : argument(2);
1133 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1134 return false;
1135 }
1136
1137 // check that length is positive
1138 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1139 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1140
1141 {
1142 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1143 uncommon_trap(Deoptimization::Reason_intrinsic,
1144 Deoptimization::Action_make_not_entrant);
1145 }
1146
1147 if (stopped()) {
1148 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1149 return true;
1150 }
1151
1152 // length is now known positive, add a cast node to make this explicit
1153 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1154 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1155 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1156 ConstraintCastNode::RegularDependency, bt);
1157 casted_length = _gvn.transform(casted_length);
1158 replace_in_map(length, casted_length);
1159 length = casted_length;
1160
1161 // Use an unsigned comparison for the range check itself
1162 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1163 BoolTest::mask btest = BoolTest::lt;
1164 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1165 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1166 _gvn.set_type(rc, rc->Value(&_gvn));
1167 if (!rc_bool->is_Con()) {
1168 record_for_igvn(rc);
1169 }
1170 set_control(_gvn.transform(new IfTrueNode(rc)));
1171 {
1172 PreserveJVMState pjvms(this);
1173 set_control(_gvn.transform(new IfFalseNode(rc)));
1174 uncommon_trap(Deoptimization::Reason_range_check,
1175 Deoptimization::Action_make_not_entrant);
1176 }
1177
1178 if (stopped()) {
1179 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1180 return true;
1181 }
1182
1183 // index is now known to be >= 0 and < length, cast it
1184 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1185 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1186 ConstraintCastNode::RegularDependency, bt);
1187 result = _gvn.transform(result);
1188 set_result(result);
1189 replace_in_map(index, result);
1190 return true;
1191 }
1192
1193 //------------------------------inline_string_indexOf------------------------
1194 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1195 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1196 return false;
1197 }
1198 Node* src = argument(0);
1199 Node* tgt = argument(1);
1200
1201 // Make the merge point
1202 RegionNode* result_rgn = new RegionNode(4);
1203 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1204
1205 src = must_be_not_null(src, true);
1206 tgt = must_be_not_null(tgt, true);
1207
1208 // Get start addr and length of source string
1209 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1210 Node* src_count = load_array_length(src);
1211
1212 // Get start addr and length of substring
1213 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1214 Node* tgt_count = load_array_length(tgt);
1215
1216 Node* result = nullptr;
1217 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1218
1219 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1220 // Divide src size by 2 if String is UTF16 encoded
1221 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1222 }
1223 if (ae == StrIntrinsicNode::UU) {
1224 // Divide substring size by 2 if String is UTF16 encoded
1225 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1226 }
1227
1228 if (call_opt_stub) {
1229 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1230 StubRoutines::_string_indexof_array[ae],
1231 "stringIndexOf", TypePtr::BOTTOM, src_start,
1232 src_count, tgt_start, tgt_count);
1233 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1234 } else {
1235 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1236 result_rgn, result_phi, ae);
1237 }
1238 if (result != nullptr) {
1239 result_phi->init_req(3, result);
1240 result_rgn->init_req(3, control());
1241 }
1242 set_control(_gvn.transform(result_rgn));
1243 record_for_igvn(result_rgn);
1244 set_result(_gvn.transform(result_phi));
1245
1246 return true;
1247 }
1248
1249 //-----------------------------inline_string_indexOfI-----------------------
1250 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1251 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1252 return false;
1253 }
1254 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1255 return false;
1256 }
1257
1258 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1259 Node* src = argument(0); // byte[]
1260 Node* src_count = argument(1); // char count
1261 Node* tgt = argument(2); // byte[]
1262 Node* tgt_count = argument(3); // char count
1263 Node* from_index = argument(4); // char index
1264
1265 src = must_be_not_null(src, true);
1266 tgt = must_be_not_null(tgt, true);
1267
1268 // Multiply byte array index by 2 if String is UTF16 encoded
1269 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1270 src_count = _gvn.transform(new SubINode(src_count, from_index));
1271 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1272 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1273
1274 // Range checks
1275 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1276 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1277 if (stopped()) {
1278 return true;
1279 }
1280
1281 RegionNode* region = new RegionNode(5);
1282 Node* phi = new PhiNode(region, TypeInt::INT);
1283 Node* result = nullptr;
1284
1285 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1286
1287 if (call_opt_stub) {
1288 assert(arrayOopDesc::base_offset_in_bytes(T_BYTE) >= 16, "Needed for indexOf");
1289 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1290 StubRoutines::_string_indexof_array[ae],
1291 "stringIndexOf", TypePtr::BOTTOM, src_start,
1292 src_count, tgt_start, tgt_count);
1293 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1294 } else {
1295 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1296 region, phi, ae);
1297 }
1298 if (result != nullptr) {
1299 // The result is index relative to from_index if substring was found, -1 otherwise.
1300 // Generate code which will fold into cmove.
1301 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1302 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1303
1304 Node* if_lt = generate_slow_guard(bol, nullptr);
1305 if (if_lt != nullptr) {
1306 // result == -1
1307 phi->init_req(3, result);
1308 region->init_req(3, if_lt);
1309 }
1310 if (!stopped()) {
1311 result = _gvn.transform(new AddINode(result, from_index));
1312 phi->init_req(4, result);
1313 region->init_req(4, control());
1314 }
1315 }
1316
1317 set_control(_gvn.transform(region));
1318 record_for_igvn(region);
1319 set_result(_gvn.transform(phi));
1320 clear_upper_avx();
1321
1322 return true;
1323 }
1324
1325 // Create StrIndexOfNode with fast path checks
1326 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1327 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1328 // Check for substr count > string count
1329 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1330 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1331 Node* if_gt = generate_slow_guard(bol, nullptr);
1332 if (if_gt != nullptr) {
1333 phi->init_req(1, intcon(-1));
1334 region->init_req(1, if_gt);
1335 }
1336 if (!stopped()) {
1337 // Check for substr count == 0
1338 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1339 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1340 Node* if_zero = generate_slow_guard(bol, nullptr);
1341 if (if_zero != nullptr) {
1342 phi->init_req(2, intcon(0));
1343 region->init_req(2, if_zero);
1344 }
1345 }
1346 if (!stopped()) {
1347 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1348 }
1349 return nullptr;
1350 }
1351
1352 //-----------------------------inline_string_indexOfChar-----------------------
1353 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1354 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1355 return false;
1356 }
1357 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1358 return false;
1359 }
1360 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1361 Node* src = argument(0); // byte[]
1362 Node* int_ch = argument(1);
1363 Node* from_index = argument(2);
1364 Node* max = argument(3);
1365
1366 src = must_be_not_null(src, true);
1367
1368 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1369 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1370 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1371
1372 // Range checks
1373 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1374
1375 // Check for int_ch >= 0
1376 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1377 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1378 {
1379 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1380 uncommon_trap(Deoptimization::Reason_intrinsic,
1381 Deoptimization::Action_maybe_recompile);
1382 }
1383 if (stopped()) {
1384 return true;
1385 }
1386
1387 RegionNode* region = new RegionNode(3);
1388 Node* phi = new PhiNode(region, TypeInt::INT);
1389
1390 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1391 C->set_has_split_ifs(true); // Has chance for split-if optimization
1392 _gvn.transform(result);
1393
1394 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1395 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1396
1397 Node* if_lt = generate_slow_guard(bol, nullptr);
1398 if (if_lt != nullptr) {
1399 // result == -1
1400 phi->init_req(2, result);
1401 region->init_req(2, if_lt);
1402 }
1403 if (!stopped()) {
1404 result = _gvn.transform(new AddINode(result, from_index));
1405 phi->init_req(1, result);
1406 region->init_req(1, control());
1407 }
1408 set_control(_gvn.transform(region));
1409 record_for_igvn(region);
1410 set_result(_gvn.transform(phi));
1411 clear_upper_avx();
1412
1413 return true;
1414 }
1415 //---------------------------inline_string_copy---------------------
1416 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1417 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1418 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1419 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1420 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1421 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1422 bool LibraryCallKit::inline_string_copy(bool compress) {
1423 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1424 return false;
1425 }
1426 int nargs = 5; // 2 oops, 3 ints
1427 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1428
1429 Node* src = argument(0);
1430 Node* src_offset = argument(1);
1431 Node* dst = argument(2);
1432 Node* dst_offset = argument(3);
1433 Node* length = argument(4);
1434
1435 // Check for allocation before we add nodes that would confuse
1436 // tightly_coupled_allocation()
1437 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1438
1439 // Figure out the size and type of the elements we will be copying.
1440 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1441 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1442 if (src_type == nullptr || dst_type == nullptr) {
1443 return false;
1444 }
1445 BasicType src_elem = src_type->elem()->array_element_basic_type();
1446 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1447 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1448 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1449 "Unsupported array types for inline_string_copy");
1450
1451 src = must_be_not_null(src, true);
1452 dst = must_be_not_null(dst, true);
1453
1454 // Convert char[] offsets to byte[] offsets
1455 bool convert_src = (compress && src_elem == T_BYTE);
1456 bool convert_dst = (!compress && dst_elem == T_BYTE);
1457 if (convert_src) {
1458 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1459 } else if (convert_dst) {
1460 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1461 }
1462
1463 // Range checks
1464 generate_string_range_check(src, src_offset, length, convert_src);
1465 generate_string_range_check(dst, dst_offset, length, convert_dst);
1466 if (stopped()) {
1467 return true;
1468 }
1469
1470 Node* src_start = array_element_address(src, src_offset, src_elem);
1471 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1472 // 'src_start' points to src array + scaled offset
1473 // 'dst_start' points to dst array + scaled offset
1474 Node* count = nullptr;
1475 if (compress) {
1476 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1477 } else {
1478 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1479 }
1480
1481 if (alloc != nullptr) {
1482 if (alloc->maybe_set_complete(&_gvn)) {
1483 // "You break it, you buy it."
1484 InitializeNode* init = alloc->initialization();
1485 assert(init->is_complete(), "we just did this");
1486 init->set_complete_with_arraycopy();
1487 assert(dst->is_CheckCastPP(), "sanity");
1488 assert(dst->in(0)->in(0) == init, "dest pinned");
1489 }
1490 // Do not let stores that initialize this object be reordered with
1491 // a subsequent store that would make this object accessible by
1492 // other threads.
1493 // Record what AllocateNode this StoreStore protects so that
1494 // escape analysis can go from the MemBarStoreStoreNode to the
1495 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1496 // based on the escape status of the AllocateNode.
1497 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1498 }
1499 if (compress) {
1500 set_result(_gvn.transform(count));
1501 }
1502 clear_upper_avx();
1503
1504 return true;
1505 }
1506
1507 #ifdef _LP64
1508 #define XTOP ,top() /*additional argument*/
1509 #else //_LP64
1510 #define XTOP /*no additional argument*/
1511 #endif //_LP64
1512
1513 //------------------------inline_string_toBytesU--------------------------
1514 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1515 bool LibraryCallKit::inline_string_toBytesU() {
1516 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1517 return false;
1518 }
1519 // Get the arguments.
1520 Node* value = argument(0);
1521 Node* offset = argument(1);
1522 Node* length = argument(2);
1523
1524 Node* newcopy = nullptr;
1525
1526 // Set the original stack and the reexecute bit for the interpreter to reexecute
1527 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1528 { PreserveReexecuteState preexecs(this);
1529 jvms()->set_should_reexecute(true);
1530
1531 // Check if a null path was taken unconditionally.
1532 value = null_check(value);
1533
1534 RegionNode* bailout = new RegionNode(1);
1535 record_for_igvn(bailout);
1536
1537 // Range checks
1538 generate_negative_guard(offset, bailout);
1539 generate_negative_guard(length, bailout);
1540 generate_limit_guard(offset, length, load_array_length(value), bailout);
1541 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1542 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1543
1544 if (bailout->req() > 1) {
1545 PreserveJVMState pjvms(this);
1546 set_control(_gvn.transform(bailout));
1547 uncommon_trap(Deoptimization::Reason_intrinsic,
1548 Deoptimization::Action_maybe_recompile);
1549 }
1550 if (stopped()) {
1551 return true;
1552 }
1553
1554 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1555 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1556 newcopy = new_array(klass_node, size, 0); // no arguments to push
1557 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1558 guarantee(alloc != nullptr, "created above");
1559
1560 // Calculate starting addresses.
1561 Node* src_start = array_element_address(value, offset, T_CHAR);
1562 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1563
1564 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1565 const TypeInt* toffset = gvn().type(offset)->is_int();
1566 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1567
1568 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1569 const char* copyfunc_name = "arraycopy";
1570 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1571 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1572 OptoRuntime::fast_arraycopy_Type(),
1573 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1574 src_start, dst_start, ConvI2X(length) XTOP);
1575 // Do not let reads from the cloned object float above the arraycopy.
1576 if (alloc->maybe_set_complete(&_gvn)) {
1577 // "You break it, you buy it."
1578 InitializeNode* init = alloc->initialization();
1579 assert(init->is_complete(), "we just did this");
1580 init->set_complete_with_arraycopy();
1581 assert(newcopy->is_CheckCastPP(), "sanity");
1582 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1583 }
1584 // Do not let stores that initialize this object be reordered with
1585 // a subsequent store that would make this object accessible by
1586 // other threads.
1587 // Record what AllocateNode this StoreStore protects so that
1588 // escape analysis can go from the MemBarStoreStoreNode to the
1589 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1590 // based on the escape status of the AllocateNode.
1591 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1592 } // original reexecute is set back here
1593
1594 C->set_has_split_ifs(true); // Has chance for split-if optimization
1595 if (!stopped()) {
1596 set_result(newcopy);
1597 }
1598 clear_upper_avx();
1599
1600 return true;
1601 }
1602
1603 //------------------------inline_string_getCharsU--------------------------
1604 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1605 bool LibraryCallKit::inline_string_getCharsU() {
1606 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1607 return false;
1608 }
1609
1610 // Get the arguments.
1611 Node* src = argument(0);
1612 Node* src_begin = argument(1);
1613 Node* src_end = argument(2); // exclusive offset (i < src_end)
1614 Node* dst = argument(3);
1615 Node* dst_begin = argument(4);
1616
1617 // Check for allocation before we add nodes that would confuse
1618 // tightly_coupled_allocation()
1619 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1620
1621 // Check if a null path was taken unconditionally.
1622 src = null_check(src);
1623 dst = null_check(dst);
1624 if (stopped()) {
1625 return true;
1626 }
1627
1628 // Get length and convert char[] offset to byte[] offset
1629 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1630 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1631
1632 // Range checks
1633 generate_string_range_check(src, src_begin, length, true);
1634 generate_string_range_check(dst, dst_begin, length, false);
1635 if (stopped()) {
1636 return true;
1637 }
1638
1639 if (!stopped()) {
1640 // Calculate starting addresses.
1641 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1642 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1643
1644 // Check if array addresses are aligned to HeapWordSize
1645 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1646 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1647 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1648 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1649
1650 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1651 const char* copyfunc_name = "arraycopy";
1652 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1653 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1654 OptoRuntime::fast_arraycopy_Type(),
1655 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1656 src_start, dst_start, ConvI2X(length) XTOP);
1657 // Do not let reads from the cloned object float above the arraycopy.
1658 if (alloc != nullptr) {
1659 if (alloc->maybe_set_complete(&_gvn)) {
1660 // "You break it, you buy it."
1661 InitializeNode* init = alloc->initialization();
1662 assert(init->is_complete(), "we just did this");
1663 init->set_complete_with_arraycopy();
1664 assert(dst->is_CheckCastPP(), "sanity");
1665 assert(dst->in(0)->in(0) == init, "dest pinned");
1666 }
1667 // Do not let stores that initialize this object be reordered with
1668 // a subsequent store that would make this object accessible by
1669 // other threads.
1670 // Record what AllocateNode this StoreStore protects so that
1671 // escape analysis can go from the MemBarStoreStoreNode to the
1672 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1673 // based on the escape status of the AllocateNode.
1674 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1675 } else {
1676 insert_mem_bar(Op_MemBarCPUOrder);
1677 }
1678 }
1679
1680 C->set_has_split_ifs(true); // Has chance for split-if optimization
1681 return true;
1682 }
1683
1684 //----------------------inline_string_char_access----------------------------
1685 // Store/Load char to/from byte[] array.
1686 // static void StringUTF16.putChar(byte[] val, int index, int c)
1687 // static char StringUTF16.getChar(byte[] val, int index)
1688 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1689 Node* value = argument(0);
1690 Node* index = argument(1);
1691 Node* ch = is_store ? argument(2) : nullptr;
1692
1693 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1694 // correctly requires matched array shapes.
1695 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1696 "sanity: byte[] and char[] bases agree");
1697 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1698 "sanity: byte[] and char[] scales agree");
1699
1700 // Bail when getChar over constants is requested: constant folding would
1701 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1702 // Java method would constant fold nicely instead.
1703 if (!is_store && value->is_Con() && index->is_Con()) {
1704 return false;
1705 }
1706
1707 // Save state and restore on bailout
1708 uint old_sp = sp();
1709 SafePointNode* old_map = clone_map();
1710
1711 value = must_be_not_null(value, true);
1712
1713 Node* adr = array_element_address(value, index, T_CHAR);
1714 if (adr->is_top()) {
1715 set_map(old_map);
1716 set_sp(old_sp);
1717 return false;
1718 }
1719 destruct_map_clone(old_map);
1720 if (is_store) {
1721 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1722 } else {
1723 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);
1724 set_result(ch);
1725 }
1726 return true;
1727 }
1728
1729 //--------------------------round_double_node--------------------------------
1730 // Round a double node if necessary.
1731 Node* LibraryCallKit::round_double_node(Node* n) {
1732 if (Matcher::strict_fp_requires_explicit_rounding) {
1733 #ifdef IA32
1734 if (UseSSE < 2) {
1735 n = _gvn.transform(new RoundDoubleNode(nullptr, n));
1736 }
1737 #else
1738 Unimplemented();
1739 #endif // IA32
1740 }
1741 return n;
1742 }
1743
1744 //------------------------------inline_math-----------------------------------
1745 // public static double Math.abs(double)
1746 // public static double Math.sqrt(double)
1747 // public static double Math.log(double)
1748 // public static double Math.log10(double)
1749 // public static double Math.round(double)
1750 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1751 Node* arg = round_double_node(argument(0));
1752 Node* n = nullptr;
1753 switch (id) {
1754 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1755 case vmIntrinsics::_dsqrt:
1756 case vmIntrinsics::_dsqrt_strict:
1757 n = new SqrtDNode(C, control(), arg); break;
1758 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1759 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1760 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1761 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1762 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, round_double_node(argument(2))); break;
1763 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1764 default: fatal_unexpected_iid(id); break;
1765 }
1766 set_result(_gvn.transform(n));
1767 return true;
1768 }
1769
1770 //------------------------------inline_math-----------------------------------
1771 // public static float Math.abs(float)
1772 // public static int Math.abs(int)
1773 // public static long Math.abs(long)
1774 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1775 Node* arg = argument(0);
1776 Node* n = nullptr;
1777 switch (id) {
1778 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1779 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1780 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1781 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1782 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1783 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1784 default: fatal_unexpected_iid(id); break;
1785 }
1786 set_result(_gvn.transform(n));
1787 return true;
1788 }
1789
1790 //------------------------------runtime_math-----------------------------
1791 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1792 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1793 "must be (DD)D or (D)D type");
1794
1795 // Inputs
1796 Node* a = round_double_node(argument(0));
1797 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? round_double_node(argument(2)) : nullptr;
1798
1799 const TypePtr* no_memory_effects = nullptr;
1800 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1801 no_memory_effects,
1802 a, top(), b, b ? top() : nullptr);
1803 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1804 #ifdef ASSERT
1805 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1806 assert(value_top == top(), "second value must be top");
1807 #endif
1808
1809 set_result(value);
1810 return true;
1811 }
1812
1813 //------------------------------inline_math_pow-----------------------------
1814 bool LibraryCallKit::inline_math_pow() {
1815 Node* exp = round_double_node(argument(2));
1816 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1817 if (d != nullptr) {
1818 if (d->getd() == 2.0) {
1819 // Special case: pow(x, 2.0) => x * x
1820 Node* base = round_double_node(argument(0));
1821 set_result(_gvn.transform(new MulDNode(base, base)));
1822 return true;
1823 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1824 // Special case: pow(x, 0.5) => sqrt(x)
1825 Node* base = round_double_node(argument(0));
1826 Node* zero = _gvn.zerocon(T_DOUBLE);
1827
1828 RegionNode* region = new RegionNode(3);
1829 Node* phi = new PhiNode(region, Type::DOUBLE);
1830
1831 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1832 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1833 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1834 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1835 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1836
1837 Node* if_pow = generate_slow_guard(test, nullptr);
1838 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1839 phi->init_req(1, value_sqrt);
1840 region->init_req(1, control());
1841
1842 if (if_pow != nullptr) {
1843 set_control(if_pow);
1844 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1845 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1846 const TypePtr* no_memory_effects = nullptr;
1847 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1848 no_memory_effects, base, top(), exp, top());
1849 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1850 #ifdef ASSERT
1851 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1852 assert(value_top == top(), "second value must be top");
1853 #endif
1854 phi->init_req(2, value_pow);
1855 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1856 }
1857
1858 C->set_has_split_ifs(true); // Has chance for split-if optimization
1859 set_control(_gvn.transform(region));
1860 record_for_igvn(region);
1861 set_result(_gvn.transform(phi));
1862
1863 return true;
1864 }
1865 }
1866
1867 return StubRoutines::dpow() != nullptr ?
1868 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1869 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1870 }
1871
1872 //------------------------------inline_math_native-----------------------------
1873 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1874 switch (id) {
1875 case vmIntrinsics::_dsin:
1876 return StubRoutines::dsin() != nullptr ?
1877 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1878 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1879 case vmIntrinsics::_dcos:
1880 return StubRoutines::dcos() != nullptr ?
1881 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1882 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1883 case vmIntrinsics::_dtan:
1884 return StubRoutines::dtan() != nullptr ?
1885 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1886 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1887 case vmIntrinsics::_dtanh:
1888 return StubRoutines::dtanh() != nullptr ?
1889 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1890 case vmIntrinsics::_dexp:
1891 return StubRoutines::dexp() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1894 case vmIntrinsics::_dlog:
1895 return StubRoutines::dlog() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1897 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1898 case vmIntrinsics::_dlog10:
1899 return StubRoutines::dlog10() != nullptr ?
1900 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1901 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1902
1903 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1904 case vmIntrinsics::_ceil:
1905 case vmIntrinsics::_floor:
1906 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1907
1908 case vmIntrinsics::_dsqrt:
1909 case vmIntrinsics::_dsqrt_strict:
1910 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1911 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1912 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1913 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1914 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1915
1916 case vmIntrinsics::_dpow: return inline_math_pow();
1917 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1918 case vmIntrinsics::_fcopySign: return inline_math(id);
1919 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1920 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1921 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1922
1923 // These intrinsics are not yet correctly implemented
1924 case vmIntrinsics::_datan2:
1925 return false;
1926
1927 default:
1928 fatal_unexpected_iid(id);
1929 return false;
1930 }
1931 }
1932
1933 //----------------------------inline_notify-----------------------------------*
1934 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1935 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1936 address func;
1937 if (id == vmIntrinsics::_notify) {
1938 func = OptoRuntime::monitor_notify_Java();
1939 } else {
1940 func = OptoRuntime::monitor_notifyAll_Java();
1941 }
1942 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1943 make_slow_call_ex(call, env()->Throwable_klass(), false);
1944 return true;
1945 }
1946
1947
1948 //----------------------------inline_min_max-----------------------------------
1949 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1950 set_result(generate_min_max(id, argument(0), argument(1)));
1951 return true;
1952 }
1953
1954 void LibraryCallKit::inline_math_mathExact(Node* math, Node *test) {
1955 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
1956 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
1957 Node* fast_path = _gvn.transform( new IfFalseNode(check));
1958 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
1959
1960 {
1961 PreserveJVMState pjvms(this);
1962 PreserveReexecuteState preexecs(this);
1963 jvms()->set_should_reexecute(true);
1964
1965 set_control(slow_path);
1966 set_i_o(i_o());
1967
1968 uncommon_trap(Deoptimization::Reason_intrinsic,
1969 Deoptimization::Action_none);
1970 }
1971
1972 set_control(fast_path);
1973 set_result(math);
1974 }
1975
1976 template <typename OverflowOp>
1977 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
1978 typedef typename OverflowOp::MathOp MathOp;
1979
1980 MathOp* mathOp = new MathOp(arg1, arg2);
1981 Node* operation = _gvn.transform( mathOp );
1982 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
1983 inline_math_mathExact(operation, ofcheck);
1984 return true;
1985 }
1986
1987 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
1988 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
1989 }
1990
1991 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
1992 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
1993 }
1994
1995 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
1996 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
1997 }
1998
1999 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2000 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2001 }
2002
2003 bool LibraryCallKit::inline_math_negateExactI() {
2004 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2005 }
2006
2007 bool LibraryCallKit::inline_math_negateExactL() {
2008 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2009 }
2010
2011 bool LibraryCallKit::inline_math_multiplyExactI() {
2012 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2013 }
2014
2015 bool LibraryCallKit::inline_math_multiplyExactL() {
2016 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2017 }
2018
2019 bool LibraryCallKit::inline_math_multiplyHigh() {
2020 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2021 return true;
2022 }
2023
2024 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2025 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2026 return true;
2027 }
2028
2029 Node*
2030 LibraryCallKit::generate_min_max(vmIntrinsics::ID id, Node* x0, Node* y0) {
2031 Node* result_val = nullptr;
2032 switch (id) {
2033 case vmIntrinsics::_min:
2034 case vmIntrinsics::_min_strict:
2035 result_val = _gvn.transform(new MinINode(x0, y0));
2036 break;
2037 case vmIntrinsics::_max:
2038 case vmIntrinsics::_max_strict:
2039 result_val = _gvn.transform(new MaxINode(x0, y0));
2040 break;
2041 default:
2042 fatal_unexpected_iid(id);
2043 break;
2044 }
2045 return result_val;
2046 }
2047
2048 inline int
2049 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2050 const TypePtr* base_type = TypePtr::NULL_PTR;
2051 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2052 if (base_type == nullptr) {
2053 // Unknown type.
2054 return Type::AnyPtr;
2055 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2056 // Since this is a null+long form, we have to switch to a rawptr.
2057 base = _gvn.transform(new CastX2PNode(offset));
2058 offset = MakeConX(0);
2059 return Type::RawPtr;
2060 } else if (base_type->base() == Type::RawPtr) {
2061 return Type::RawPtr;
2062 } else if (base_type->isa_oopptr()) {
2063 // Base is never null => always a heap address.
2064 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2065 return Type::OopPtr;
2066 }
2067 // Offset is small => always a heap address.
2068 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2069 if (offset_type != nullptr &&
2070 base_type->offset() == 0 && // (should always be?)
2071 offset_type->_lo >= 0 &&
2072 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2073 return Type::OopPtr;
2074 } else if (type == T_OBJECT) {
2075 // off heap access to an oop doesn't make any sense. Has to be on
2076 // heap.
2077 return Type::OopPtr;
2078 }
2079 // Otherwise, it might either be oop+off or null+addr.
2080 return Type::AnyPtr;
2081 } else {
2082 // No information:
2083 return Type::AnyPtr;
2084 }
2085 }
2086
2087 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2088 Node* uncasted_base = base;
2089 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2090 if (kind == Type::RawPtr) {
2091 return basic_plus_adr(top(), uncasted_base, offset);
2092 } else if (kind == Type::AnyPtr) {
2093 assert(base == uncasted_base, "unexpected base change");
2094 if (can_cast) {
2095 if (!_gvn.type(base)->speculative_maybe_null() &&
2096 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2097 // According to profiling, this access is always on
2098 // heap. Casting the base to not null and thus avoiding membars
2099 // around the access should allow better optimizations
2100 Node* null_ctl = top();
2101 base = null_check_oop(base, &null_ctl, true, true, true);
2102 assert(null_ctl->is_top(), "no null control here");
2103 return basic_plus_adr(base, offset);
2104 } else if (_gvn.type(base)->speculative_always_null() &&
2105 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2106 // According to profiling, this access is always off
2107 // heap.
2108 base = null_assert(base);
2109 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2110 offset = MakeConX(0);
2111 return basic_plus_adr(top(), raw_base, offset);
2112 }
2113 }
2114 // We don't know if it's an on heap or off heap access. Fall back
2115 // to raw memory access.
2116 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2117 return basic_plus_adr(top(), raw, offset);
2118 } else {
2119 assert(base == uncasted_base, "unexpected base change");
2120 // We know it's an on heap access so base can't be null
2121 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2122 base = must_be_not_null(base, true);
2123 }
2124 return basic_plus_adr(base, offset);
2125 }
2126 }
2127
2128 //--------------------------inline_number_methods-----------------------------
2129 // inline int Integer.numberOfLeadingZeros(int)
2130 // inline int Long.numberOfLeadingZeros(long)
2131 //
2132 // inline int Integer.numberOfTrailingZeros(int)
2133 // inline int Long.numberOfTrailingZeros(long)
2134 //
2135 // inline int Integer.bitCount(int)
2136 // inline int Long.bitCount(long)
2137 //
2138 // inline char Character.reverseBytes(char)
2139 // inline short Short.reverseBytes(short)
2140 // inline int Integer.reverseBytes(int)
2141 // inline long Long.reverseBytes(long)
2142 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2143 Node* arg = argument(0);
2144 Node* n = nullptr;
2145 switch (id) {
2146 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2147 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2148 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2149 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2150 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2151 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2152 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode(nullptr, arg); break;
2153 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( nullptr, arg); break;
2154 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( nullptr, arg); break;
2155 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( nullptr, arg); break;
2156 case vmIntrinsics::_reverse_i: n = new ReverseINode(nullptr, arg); break;
2157 case vmIntrinsics::_reverse_l: n = new ReverseLNode(nullptr, arg); break;
2158 default: fatal_unexpected_iid(id); break;
2159 }
2160 set_result(_gvn.transform(n));
2161 return true;
2162 }
2163
2164 //--------------------------inline_bitshuffle_methods-----------------------------
2165 // inline int Integer.compress(int, int)
2166 // inline int Integer.expand(int, int)
2167 // inline long Long.compress(long, long)
2168 // inline long Long.expand(long, long)
2169 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2170 Node* n = nullptr;
2171 switch (id) {
2172 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2173 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2174 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2175 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2176 default: fatal_unexpected_iid(id); break;
2177 }
2178 set_result(_gvn.transform(n));
2179 return true;
2180 }
2181
2182 //--------------------------inline_number_methods-----------------------------
2183 // inline int Integer.compareUnsigned(int, int)
2184 // inline int Long.compareUnsigned(long, long)
2185 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2186 Node* arg1 = argument(0);
2187 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2188 Node* n = nullptr;
2189 switch (id) {
2190 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2191 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2192 default: fatal_unexpected_iid(id); break;
2193 }
2194 set_result(_gvn.transform(n));
2195 return true;
2196 }
2197
2198 //--------------------------inline_unsigned_divmod_methods-----------------------------
2199 // inline int Integer.divideUnsigned(int, int)
2200 // inline int Integer.remainderUnsigned(int, int)
2201 // inline long Long.divideUnsigned(long, long)
2202 // inline long Long.remainderUnsigned(long, long)
2203 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2204 Node* n = nullptr;
2205 switch (id) {
2206 case vmIntrinsics::_divideUnsigned_i: {
2207 zero_check_int(argument(1));
2208 // Compile-time detect of null-exception
2209 if (stopped()) {
2210 return true; // keep the graph constructed so far
2211 }
2212 n = new UDivINode(control(), argument(0), argument(1));
2213 break;
2214 }
2215 case vmIntrinsics::_divideUnsigned_l: {
2216 zero_check_long(argument(2));
2217 // Compile-time detect of null-exception
2218 if (stopped()) {
2219 return true; // keep the graph constructed so far
2220 }
2221 n = new UDivLNode(control(), argument(0), argument(2));
2222 break;
2223 }
2224 case vmIntrinsics::_remainderUnsigned_i: {
2225 zero_check_int(argument(1));
2226 // Compile-time detect of null-exception
2227 if (stopped()) {
2228 return true; // keep the graph constructed so far
2229 }
2230 n = new UModINode(control(), argument(0), argument(1));
2231 break;
2232 }
2233 case vmIntrinsics::_remainderUnsigned_l: {
2234 zero_check_long(argument(2));
2235 // Compile-time detect of null-exception
2236 if (stopped()) {
2237 return true; // keep the graph constructed so far
2238 }
2239 n = new UModLNode(control(), argument(0), argument(2));
2240 break;
2241 }
2242 default: fatal_unexpected_iid(id); break;
2243 }
2244 set_result(_gvn.transform(n));
2245 return true;
2246 }
2247
2248 //----------------------------inline_unsafe_access----------------------------
2249
2250 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2251 // Attempt to infer a sharper value type from the offset and base type.
2252 ciKlass* sharpened_klass = nullptr;
2253
2254 // See if it is an instance field, with an object type.
2255 if (alias_type->field() != nullptr) {
2256 if (alias_type->field()->type()->is_klass()) {
2257 sharpened_klass = alias_type->field()->type()->as_klass();
2258 }
2259 }
2260
2261 const TypeOopPtr* result = nullptr;
2262 // See if it is a narrow oop array.
2263 if (adr_type->isa_aryptr()) {
2264 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2265 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2266 if (elem_type != nullptr && elem_type->is_loaded()) {
2267 // Sharpen the value type.
2268 result = elem_type;
2269 }
2270 }
2271 }
2272
2273 // The sharpened class might be unloaded if there is no class loader
2274 // contraint in place.
2275 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2276 // Sharpen the value type.
2277 result = TypeOopPtr::make_from_klass(sharpened_klass);
2278 }
2279 if (result != nullptr) {
2280 #ifndef PRODUCT
2281 if (C->print_intrinsics() || C->print_inlining()) {
2282 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2283 tty->print(" sharpened value: "); result->dump(); tty->cr();
2284 }
2285 #endif
2286 }
2287 return result;
2288 }
2289
2290 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2291 switch (kind) {
2292 case Relaxed:
2293 return MO_UNORDERED;
2294 case Opaque:
2295 return MO_RELAXED;
2296 case Acquire:
2297 return MO_ACQUIRE;
2298 case Release:
2299 return MO_RELEASE;
2300 case Volatile:
2301 return MO_SEQ_CST;
2302 default:
2303 ShouldNotReachHere();
2304 return 0;
2305 }
2306 }
2307
2308 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2309 if (callee()->is_static()) return false; // caller must have the capability!
2310 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2311 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2312 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2313 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2314
2315 if (is_reference_type(type)) {
2316 decorators |= ON_UNKNOWN_OOP_REF;
2317 }
2318
2319 if (unaligned) {
2320 decorators |= C2_UNALIGNED;
2321 }
2322
2323 #ifndef PRODUCT
2324 {
2325 ResourceMark rm;
2326 // Check the signatures.
2327 ciSignature* sig = callee()->signature();
2328 #ifdef ASSERT
2329 if (!is_store) {
2330 // Object getReference(Object base, int/long offset), etc.
2331 BasicType rtype = sig->return_type()->basic_type();
2332 assert(rtype == type, "getter must return the expected value");
2333 assert(sig->count() == 2, "oop getter has 2 arguments");
2334 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2335 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2336 } else {
2337 // void putReference(Object base, int/long offset, Object x), etc.
2338 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2339 assert(sig->count() == 3, "oop putter has 3 arguments");
2340 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2341 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2342 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2343 assert(vtype == type, "putter must accept the expected value");
2344 }
2345 #endif // ASSERT
2346 }
2347 #endif //PRODUCT
2348
2349 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2350
2351 Node* receiver = argument(0); // type: oop
2352
2353 // Build address expression.
2354 Node* heap_base_oop = top();
2355
2356 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2357 Node* base = argument(1); // type: oop
2358 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2359 Node* offset = argument(2); // type: long
2360 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2361 // to be plain byte offsets, which are also the same as those accepted
2362 // by oopDesc::field_addr.
2363 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2364 "fieldOffset must be byte-scaled");
2365 // 32-bit machines ignore the high half!
2366 offset = ConvL2X(offset);
2367
2368 // Save state and restore on bailout
2369 uint old_sp = sp();
2370 SafePointNode* old_map = clone_map();
2371
2372 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2373 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2374
2375 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2376 if (type != T_OBJECT) {
2377 decorators |= IN_NATIVE; // off-heap primitive access
2378 } else {
2379 set_map(old_map);
2380 set_sp(old_sp);
2381 return false; // off-heap oop accesses are not supported
2382 }
2383 } else {
2384 heap_base_oop = base; // on-heap or mixed access
2385 }
2386
2387 // Can base be null? Otherwise, always on-heap access.
2388 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2389
2390 if (!can_access_non_heap) {
2391 decorators |= IN_HEAP;
2392 }
2393
2394 Node* val = is_store ? argument(4) : nullptr;
2395
2396 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2397 if (adr_type == TypePtr::NULL_PTR) {
2398 set_map(old_map);
2399 set_sp(old_sp);
2400 return false; // off-heap access with zero address
2401 }
2402
2403 // Try to categorize the address.
2404 Compile::AliasType* alias_type = C->alias_type(adr_type);
2405 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2406
2407 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2408 alias_type->adr_type() == TypeAryPtr::RANGE) {
2409 set_map(old_map);
2410 set_sp(old_sp);
2411 return false; // not supported
2412 }
2413
2414 bool mismatched = false;
2415 BasicType bt = alias_type->basic_type();
2416 if (bt != T_ILLEGAL) {
2417 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2418 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2419 // Alias type doesn't differentiate between byte[] and boolean[]).
2420 // Use address type to get the element type.
2421 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2422 }
2423 if (is_reference_type(bt, true)) {
2424 // accessing an array field with getReference is not a mismatch
2425 bt = T_OBJECT;
2426 }
2427 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2428 // Don't intrinsify mismatched object accesses
2429 set_map(old_map);
2430 set_sp(old_sp);
2431 return false;
2432 }
2433 mismatched = (bt != type);
2434 } else if (alias_type->adr_type()->isa_oopptr()) {
2435 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2436 }
2437
2438 destruct_map_clone(old_map);
2439 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2440
2441 if (mismatched) {
2442 decorators |= C2_MISMATCHED;
2443 }
2444
2445 // First guess at the value type.
2446 const Type *value_type = Type::get_const_basic_type(type);
2447
2448 // Figure out the memory ordering.
2449 decorators |= mo_decorator_for_access_kind(kind);
2450
2451 if (!is_store && type == T_OBJECT) {
2452 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2453 if (tjp != nullptr) {
2454 value_type = tjp;
2455 }
2456 }
2457
2458 receiver = null_check(receiver);
2459 if (stopped()) {
2460 return true;
2461 }
2462 // Heap pointers get a null-check from the interpreter,
2463 // as a courtesy. However, this is not guaranteed by Unsafe,
2464 // and it is not possible to fully distinguish unintended nulls
2465 // from intended ones in this API.
2466
2467 if (!is_store) {
2468 Node* p = nullptr;
2469 // Try to constant fold a load from a constant field
2470 ciField* field = alias_type->field();
2471 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2472 // final or stable field
2473 p = make_constant_from_field(field, heap_base_oop);
2474 }
2475
2476 if (p == nullptr) { // Could not constant fold the load
2477 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2478 // Normalize the value returned by getBoolean in the following cases
2479 if (type == T_BOOLEAN &&
2480 (mismatched ||
2481 heap_base_oop == top() || // - heap_base_oop is null or
2482 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2483 // and the unsafe access is made to large offset
2484 // (i.e., larger than the maximum offset necessary for any
2485 // field access)
2486 ) {
2487 IdealKit ideal = IdealKit(this);
2488 #define __ ideal.
2489 IdealVariable normalized_result(ideal);
2490 __ declarations_done();
2491 __ set(normalized_result, p);
2492 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2493 __ set(normalized_result, ideal.ConI(1));
2494 ideal.end_if();
2495 final_sync(ideal);
2496 p = __ value(normalized_result);
2497 #undef __
2498 }
2499 }
2500 if (type == T_ADDRESS) {
2501 p = gvn().transform(new CastP2XNode(nullptr, p));
2502 p = ConvX2UL(p);
2503 }
2504 // The load node has the control of the preceding MemBarCPUOrder. All
2505 // following nodes will have the control of the MemBarCPUOrder inserted at
2506 // the end of this method. So, pushing the load onto the stack at a later
2507 // point is fine.
2508 set_result(p);
2509 } else {
2510 if (bt == T_ADDRESS) {
2511 // Repackage the long as a pointer.
2512 val = ConvL2X(val);
2513 val = gvn().transform(new CastX2PNode(val));
2514 }
2515 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2516 }
2517
2518 return true;
2519 }
2520
2521 //----------------------------inline_unsafe_load_store----------------------------
2522 // This method serves a couple of different customers (depending on LoadStoreKind):
2523 //
2524 // LS_cmp_swap:
2525 //
2526 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2527 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2528 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2529 //
2530 // LS_cmp_swap_weak:
2531 //
2532 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2533 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2534 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2535 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2536 //
2537 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2538 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2539 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2540 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2541 //
2542 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2543 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2544 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2545 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2546 //
2547 // LS_cmp_exchange:
2548 //
2549 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2550 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2551 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2552 //
2553 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2554 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2555 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2556 //
2557 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2558 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2559 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2560 //
2561 // LS_get_add:
2562 //
2563 // int getAndAddInt( Object o, long offset, int delta)
2564 // long getAndAddLong(Object o, long offset, long delta)
2565 //
2566 // LS_get_set:
2567 //
2568 // int getAndSet(Object o, long offset, int newValue)
2569 // long getAndSet(Object o, long offset, long newValue)
2570 // Object getAndSet(Object o, long offset, Object newValue)
2571 //
2572 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2573 // This basic scheme here is the same as inline_unsafe_access, but
2574 // differs in enough details that combining them would make the code
2575 // overly confusing. (This is a true fact! I originally combined
2576 // them, but even I was confused by it!) As much code/comments as
2577 // possible are retained from inline_unsafe_access though to make
2578 // the correspondences clearer. - dl
2579
2580 if (callee()->is_static()) return false; // caller must have the capability!
2581
2582 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2583 decorators |= mo_decorator_for_access_kind(access_kind);
2584
2585 #ifndef PRODUCT
2586 BasicType rtype;
2587 {
2588 ResourceMark rm;
2589 // Check the signatures.
2590 ciSignature* sig = callee()->signature();
2591 rtype = sig->return_type()->basic_type();
2592 switch(kind) {
2593 case LS_get_add:
2594 case LS_get_set: {
2595 // Check the signatures.
2596 #ifdef ASSERT
2597 assert(rtype == type, "get and set must return the expected type");
2598 assert(sig->count() == 3, "get and set has 3 arguments");
2599 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2600 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2601 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2602 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2603 #endif // ASSERT
2604 break;
2605 }
2606 case LS_cmp_swap:
2607 case LS_cmp_swap_weak: {
2608 // Check the signatures.
2609 #ifdef ASSERT
2610 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2611 assert(sig->count() == 4, "CAS has 4 arguments");
2612 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2613 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2614 #endif // ASSERT
2615 break;
2616 }
2617 case LS_cmp_exchange: {
2618 // Check the signatures.
2619 #ifdef ASSERT
2620 assert(rtype == type, "CAS must return the expected type");
2621 assert(sig->count() == 4, "CAS has 4 arguments");
2622 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2623 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2624 #endif // ASSERT
2625 break;
2626 }
2627 default:
2628 ShouldNotReachHere();
2629 }
2630 }
2631 #endif //PRODUCT
2632
2633 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2634
2635 // Get arguments:
2636 Node* receiver = nullptr;
2637 Node* base = nullptr;
2638 Node* offset = nullptr;
2639 Node* oldval = nullptr;
2640 Node* newval = nullptr;
2641 switch(kind) {
2642 case LS_cmp_swap:
2643 case LS_cmp_swap_weak:
2644 case LS_cmp_exchange: {
2645 const bool two_slot_type = type2size[type] == 2;
2646 receiver = argument(0); // type: oop
2647 base = argument(1); // type: oop
2648 offset = argument(2); // type: long
2649 oldval = argument(4); // type: oop, int, or long
2650 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2651 break;
2652 }
2653 case LS_get_add:
2654 case LS_get_set: {
2655 receiver = argument(0); // type: oop
2656 base = argument(1); // type: oop
2657 offset = argument(2); // type: long
2658 oldval = nullptr;
2659 newval = argument(4); // type: oop, int, or long
2660 break;
2661 }
2662 default:
2663 ShouldNotReachHere();
2664 }
2665
2666 // Build field offset expression.
2667 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2668 // to be plain byte offsets, which are also the same as those accepted
2669 // by oopDesc::field_addr.
2670 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2671 // 32-bit machines ignore the high half of long offsets
2672 offset = ConvL2X(offset);
2673 // Save state and restore on bailout
2674 uint old_sp = sp();
2675 SafePointNode* old_map = clone_map();
2676 Node* adr = make_unsafe_address(base, offset,type, false);
2677 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2678
2679 Compile::AliasType* alias_type = C->alias_type(adr_type);
2680 BasicType bt = alias_type->basic_type();
2681 if (bt != T_ILLEGAL &&
2682 (is_reference_type(bt) != (type == T_OBJECT))) {
2683 // Don't intrinsify mismatched object accesses.
2684 set_map(old_map);
2685 set_sp(old_sp);
2686 return false;
2687 }
2688
2689 destruct_map_clone(old_map);
2690
2691 // For CAS, unlike inline_unsafe_access, there seems no point in
2692 // trying to refine types. Just use the coarse types here.
2693 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2694 const Type *value_type = Type::get_const_basic_type(type);
2695
2696 switch (kind) {
2697 case LS_get_set:
2698 case LS_cmp_exchange: {
2699 if (type == T_OBJECT) {
2700 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2701 if (tjp != nullptr) {
2702 value_type = tjp;
2703 }
2704 }
2705 break;
2706 }
2707 case LS_cmp_swap:
2708 case LS_cmp_swap_weak:
2709 case LS_get_add:
2710 break;
2711 default:
2712 ShouldNotReachHere();
2713 }
2714
2715 // Null check receiver.
2716 receiver = null_check(receiver);
2717 if (stopped()) {
2718 return true;
2719 }
2720
2721 int alias_idx = C->get_alias_index(adr_type);
2722
2723 if (is_reference_type(type)) {
2724 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2725
2726 // Transformation of a value which could be null pointer (CastPP #null)
2727 // could be delayed during Parse (for example, in adjust_map_after_if()).
2728 // Execute transformation here to avoid barrier generation in such case.
2729 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2730 newval = _gvn.makecon(TypePtr::NULL_PTR);
2731
2732 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2733 // Refine the value to a null constant, when it is known to be null
2734 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2735 }
2736 }
2737
2738 Node* result = nullptr;
2739 switch (kind) {
2740 case LS_cmp_exchange: {
2741 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2742 oldval, newval, value_type, type, decorators);
2743 break;
2744 }
2745 case LS_cmp_swap_weak:
2746 decorators |= C2_WEAK_CMPXCHG;
2747 case LS_cmp_swap: {
2748 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2749 oldval, newval, value_type, type, decorators);
2750 break;
2751 }
2752 case LS_get_set: {
2753 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2754 newval, value_type, type, decorators);
2755 break;
2756 }
2757 case LS_get_add: {
2758 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2759 newval, value_type, type, decorators);
2760 break;
2761 }
2762 default:
2763 ShouldNotReachHere();
2764 }
2765
2766 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2767 set_result(result);
2768 return true;
2769 }
2770
2771 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2772 // Regardless of form, don't allow previous ld/st to move down,
2773 // then issue acquire, release, or volatile mem_bar.
2774 insert_mem_bar(Op_MemBarCPUOrder);
2775 switch(id) {
2776 case vmIntrinsics::_loadFence:
2777 insert_mem_bar(Op_LoadFence);
2778 return true;
2779 case vmIntrinsics::_storeFence:
2780 insert_mem_bar(Op_StoreFence);
2781 return true;
2782 case vmIntrinsics::_storeStoreFence:
2783 insert_mem_bar(Op_StoreStoreFence);
2784 return true;
2785 case vmIntrinsics::_fullFence:
2786 insert_mem_bar(Op_MemBarVolatile);
2787 return true;
2788 default:
2789 fatal_unexpected_iid(id);
2790 return false;
2791 }
2792 }
2793
2794 bool LibraryCallKit::inline_onspinwait() {
2795 insert_mem_bar(Op_OnSpinWait);
2796 return true;
2797 }
2798
2799 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2800 if (!kls->is_Con()) {
2801 return true;
2802 }
2803 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2804 if (klsptr == nullptr) {
2805 return true;
2806 }
2807 ciInstanceKlass* ik = klsptr->instance_klass();
2808 // don't need a guard for a klass that is already initialized
2809 return !ik->is_initialized();
2810 }
2811
2812 //----------------------------inline_unsafe_writeback0-------------------------
2813 // public native void Unsafe.writeback0(long address)
2814 bool LibraryCallKit::inline_unsafe_writeback0() {
2815 if (!Matcher::has_match_rule(Op_CacheWB)) {
2816 return false;
2817 }
2818 #ifndef PRODUCT
2819 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2820 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2821 ciSignature* sig = callee()->signature();
2822 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2823 #endif
2824 null_check_receiver(); // null-check, then ignore
2825 Node *addr = argument(1);
2826 addr = new CastX2PNode(addr);
2827 addr = _gvn.transform(addr);
2828 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2829 flush = _gvn.transform(flush);
2830 set_memory(flush, TypeRawPtr::BOTTOM);
2831 return true;
2832 }
2833
2834 //----------------------------inline_unsafe_writeback0-------------------------
2835 // public native void Unsafe.writeback0(long address)
2836 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2837 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2838 return false;
2839 }
2840 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2841 return false;
2842 }
2843 #ifndef PRODUCT
2844 assert(Matcher::has_match_rule(Op_CacheWB),
2845 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2846 : "found match rule for CacheWBPostSync but not CacheWB"));
2847
2848 #endif
2849 null_check_receiver(); // null-check, then ignore
2850 Node *sync;
2851 if (is_pre) {
2852 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2853 } else {
2854 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2855 }
2856 sync = _gvn.transform(sync);
2857 set_memory(sync, TypeRawPtr::BOTTOM);
2858 return true;
2859 }
2860
2861 //----------------------------inline_unsafe_allocate---------------------------
2862 // public native Object Unsafe.allocateInstance(Class<?> cls);
2863 bool LibraryCallKit::inline_unsafe_allocate() {
2864
2865 #if INCLUDE_JVMTI
2866 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2867 return false;
2868 }
2869 #endif //INCLUDE_JVMTI
2870
2871 if (callee()->is_static()) return false; // caller must have the capability!
2872
2873 null_check_receiver(); // null-check, then ignore
2874 Node* cls = null_check(argument(1));
2875 if (stopped()) return true;
2876
2877 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2878 kls = null_check(kls);
2879 if (stopped()) return true; // argument was like int.class
2880
2881 #if INCLUDE_JVMTI
2882 // Don't try to access new allocated obj in the intrinsic.
2883 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2884 // Deoptimize and allocate in interpreter instead.
2885 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2886 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2887 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2888 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2889 {
2890 BuildCutout unless(this, tst, PROB_MAX);
2891 uncommon_trap(Deoptimization::Reason_intrinsic,
2892 Deoptimization::Action_make_not_entrant);
2893 }
2894 if (stopped()) {
2895 return true;
2896 }
2897 #endif //INCLUDE_JVMTI
2898
2899 Node* test = nullptr;
2900 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2901 // Note: The argument might still be an illegal value like
2902 // Serializable.class or Object[].class. The runtime will handle it.
2903 // But we must make an explicit check for initialization.
2904 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2905 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2906 // can generate code to load it as unsigned byte.
2907 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2908 Node* bits = intcon(InstanceKlass::fully_initialized);
2909 test = _gvn.transform(new SubINode(inst, bits));
2910 // The 'test' is non-zero if we need to take a slow path.
2911 }
2912
2913 Node* obj = new_instance(kls, test);
2914 set_result(obj);
2915 return true;
2916 }
2917
2918 //------------------------inline_native_time_funcs--------------
2919 // inline code for System.currentTimeMillis() and System.nanoTime()
2920 // these have the same type and signature
2921 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2922 const TypeFunc* tf = OptoRuntime::void_long_Type();
2923 const TypePtr* no_memory_effects = nullptr;
2924 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2925 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2926 #ifdef ASSERT
2927 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2928 assert(value_top == top(), "second value must be top");
2929 #endif
2930 set_result(value);
2931 return true;
2932 }
2933
2934
2935 #if INCLUDE_JVMTI
2936
2937 // When notifications are disabled then just update the VTMS transition bit and return.
2938 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2939 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2940 if (!DoJVMTIVirtualThreadTransitions) {
2941 return true;
2942 }
2943 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
2944 IdealKit ideal(this);
2945
2946 Node* ONE = ideal.ConI(1);
2947 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
2948 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
2949 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
2950
2951 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
2952 sync_kit(ideal);
2953 // if notifyJvmti enabled then make a call to the given SharedRuntime function
2954 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
2955 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
2956 ideal.sync_kit(this);
2957 } ideal.else_(); {
2958 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
2959 Node* thread = ideal.thread();
2960 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
2961 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
2962 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2963
2964 sync_kit(ideal);
2965 access_store_at(nullptr, jt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2966 access_store_at(nullptr, vt_addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2967
2968 ideal.sync_kit(this);
2969 } ideal.end_if();
2970 final_sync(ideal);
2971
2972 return true;
2973 }
2974
2975 // Always update the temporary VTMS transition bit.
2976 bool LibraryCallKit::inline_native_notify_jvmti_hide() {
2977 if (!DoJVMTIVirtualThreadTransitions) {
2978 return true;
2979 }
2980 IdealKit ideal(this);
2981
2982 {
2983 // unconditionally update the temporary VTMS transition bit in current JavaThread
2984 Node* thread = ideal.thread();
2985 Node* hide = _gvn.transform(argument(0)); // hide argument for temporary VTMS transition notification
2986 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_tmp_VTMS_transition_offset()));
2987 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
2988
2989 sync_kit(ideal);
2990 access_store_at(nullptr, addr, addr_type, hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
2991 ideal.sync_kit(this);
2992 }
2993 final_sync(ideal);
2994
2995 return true;
2996 }
2997
2998 // Always update the is_disable_suspend bit.
2999 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3000 if (!DoJVMTIVirtualThreadTransitions) {
3001 return true;
3002 }
3003 IdealKit ideal(this);
3004
3005 {
3006 // unconditionally update the is_disable_suspend bit in current JavaThread
3007 Node* thread = ideal.thread();
3008 Node* arg = _gvn.transform(argument(0)); // argument for notification
3009 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3010 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3011
3012 sync_kit(ideal);
3013 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3014 ideal.sync_kit(this);
3015 }
3016 final_sync(ideal);
3017
3018 return true;
3019 }
3020
3021 #endif // INCLUDE_JVMTI
3022
3023 #ifdef JFR_HAVE_INTRINSICS
3024
3025 /**
3026 * if oop->klass != null
3027 * // normal class
3028 * epoch = _epoch_state ? 2 : 1
3029 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3030 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3031 * }
3032 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3033 * else
3034 * // primitive class
3035 * if oop->array_klass != null
3036 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3037 * else
3038 * id = LAST_TYPE_ID + 1 // void class path
3039 * if (!signaled)
3040 * signaled = true
3041 */
3042 bool LibraryCallKit::inline_native_classID() {
3043 Node* cls = argument(0);
3044
3045 IdealKit ideal(this);
3046 #define __ ideal.
3047 IdealVariable result(ideal); __ declarations_done();
3048 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3049 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3050 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3051
3052
3053 __ if_then(kls, BoolTest::ne, null()); {
3054 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3055 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3056
3057 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3058 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3059 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3060 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3061 mask = _gvn.transform(new OrLNode(mask, epoch));
3062 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3063
3064 float unlikely = PROB_UNLIKELY(0.999);
3065 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3066 sync_kit(ideal);
3067 make_runtime_call(RC_LEAF,
3068 OptoRuntime::class_id_load_barrier_Type(),
3069 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3070 "class id load barrier",
3071 TypePtr::BOTTOM,
3072 kls);
3073 ideal.sync_kit(this);
3074 } __ end_if();
3075
3076 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3077 } __ else_(); {
3078 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(),
3079 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3080 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3081 __ if_then(array_kls, BoolTest::ne, null()); {
3082 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3083 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3084 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3085 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3086 } __ else_(); {
3087 // void class case
3088 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3089 } __ end_if();
3090
3091 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3092 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3093 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3094 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3095 } __ end_if();
3096 } __ end_if();
3097
3098 final_sync(ideal);
3099 set_result(ideal.value(result));
3100 #undef __
3101 return true;
3102 }
3103
3104 //------------------------inline_native_jvm_commit------------------
3105 bool LibraryCallKit::inline_native_jvm_commit() {
3106 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3107
3108 // Save input memory and i_o state.
3109 Node* input_memory_state = reset_memory();
3110 set_all_memory(input_memory_state);
3111 Node* input_io_state = i_o();
3112
3113 // TLS.
3114 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3115 // Jfr java buffer.
3116 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3117 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3118 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3119
3120 // Load the current value of the notified field in the JfrThreadLocal.
3121 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3122 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3123
3124 // Test for notification.
3125 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3126 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3127 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3128
3129 // True branch, is notified.
3130 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3131 set_control(is_notified);
3132
3133 // Reset notified state.
3134 Node* notified_reset_memory = store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered);
3135
3136 // 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.
3137 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3138 // Convert the machine-word to a long.
3139 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3140
3141 // False branch, not notified.
3142 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3143 set_control(not_notified);
3144 set_all_memory(input_memory_state);
3145
3146 // Arg is the next position as a long.
3147 Node* arg = argument(0);
3148 // Convert long to machine-word.
3149 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3150
3151 // Store the next_position to the underlying jfr java buffer.
3152 Node* commit_memory;
3153 #ifdef _LP64
3154 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_LONG, Compile::AliasIdxRaw, MemNode::release);
3155 #else
3156 commit_memory = store_to_memory(control(), java_buffer_pos_offset, next_pos_X, T_INT, Compile::AliasIdxRaw, MemNode::release);
3157 #endif
3158
3159 // 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.
3160 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3161 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3162 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3163
3164 // And flags with lease constant.
3165 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3166
3167 // Branch on lease to conditionalize returning the leased java buffer.
3168 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3169 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3170 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3171
3172 // False branch, not a lease.
3173 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3174
3175 // True branch, is lease.
3176 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3177 set_control(is_lease);
3178
3179 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3180 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3181 OptoRuntime::void_void_Type(),
3182 SharedRuntime::jfr_return_lease(),
3183 "return_lease", TypePtr::BOTTOM);
3184 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3185
3186 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3187 record_for_igvn(lease_compare_rgn);
3188 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3189 record_for_igvn(lease_compare_mem);
3190 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3191 record_for_igvn(lease_compare_io);
3192 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3193 record_for_igvn(lease_result_value);
3194
3195 // Update control and phi nodes.
3196 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3197 lease_compare_rgn->init_req(_false_path, not_lease);
3198
3199 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3200 lease_compare_mem->init_req(_false_path, commit_memory);
3201
3202 lease_compare_io->init_req(_true_path, i_o());
3203 lease_compare_io->init_req(_false_path, input_io_state);
3204
3205 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3206 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3207
3208 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3209 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3210 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3211 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3212
3213 // Update control and phi nodes.
3214 result_rgn->init_req(_true_path, is_notified);
3215 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3216
3217 result_mem->init_req(_true_path, notified_reset_memory);
3218 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3219
3220 result_io->init_req(_true_path, input_io_state);
3221 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3222
3223 result_value->init_req(_true_path, current_pos);
3224 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3225
3226 // Set output state.
3227 set_control(_gvn.transform(result_rgn));
3228 set_all_memory(_gvn.transform(result_mem));
3229 set_i_o(_gvn.transform(result_io));
3230 set_result(result_rgn, result_value);
3231 return true;
3232 }
3233
3234 /*
3235 * The intrinsic is a model of this pseudo-code:
3236 *
3237 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3238 * jobject h_event_writer = tl->java_event_writer();
3239 * if (h_event_writer == nullptr) {
3240 * return nullptr;
3241 * }
3242 * oop threadObj = Thread::threadObj();
3243 * oop vthread = java_lang_Thread::vthread(threadObj);
3244 * traceid tid;
3245 * bool pinVirtualThread;
3246 * bool excluded;
3247 * if (vthread != threadObj) { // i.e. current thread is virtual
3248 * tid = java_lang_Thread::tid(vthread);
3249 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3250 * pinVirtualThread = VMContinuations;
3251 * excluded = vthread_epoch_raw & excluded_mask;
3252 * if (!excluded) {
3253 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3254 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3255 * if (vthread_epoch != current_epoch) {
3256 * write_checkpoint();
3257 * }
3258 * }
3259 * } else {
3260 * tid = java_lang_Thread::tid(threadObj);
3261 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3262 * pinVirtualThread = false;
3263 * excluded = thread_epoch_raw & excluded_mask;
3264 * }
3265 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3266 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3267 * if (tid_in_event_writer != tid) {
3268 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3269 * setField(event_writer, "excluded", excluded);
3270 * setField(event_writer, "threadID", tid);
3271 * }
3272 * return event_writer
3273 */
3274 bool LibraryCallKit::inline_native_getEventWriter() {
3275 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3276
3277 // Save input memory and i_o state.
3278 Node* input_memory_state = reset_memory();
3279 set_all_memory(input_memory_state);
3280 Node* input_io_state = i_o();
3281
3282 Node* excluded_mask = _gvn.intcon(32768);
3283 Node* epoch_mask = _gvn.intcon(32767);
3284
3285 // TLS
3286 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3287
3288 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3289 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3290
3291 // Load the eventwriter jobject handle.
3292 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3293
3294 // Null check the jobject handle.
3295 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3296 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3297 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3298
3299 // False path, jobj is null.
3300 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3301
3302 // True path, jobj is not null.
3303 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3304
3305 set_control(jobj_is_not_null);
3306
3307 // Load the threadObj for the CarrierThread.
3308 Node* threadObj = generate_current_thread(tls_ptr);
3309
3310 // Load the vthread.
3311 Node* vthread = generate_virtual_thread(tls_ptr);
3312
3313 // If vthread != threadObj, this is a virtual thread.
3314 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3315 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3316 IfNode* iff_vthread_not_equal_threadObj =
3317 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3318
3319 // False branch, fallback to threadObj.
3320 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3321 set_control(vthread_equal_threadObj);
3322
3323 // Load the tid field from the vthread object.
3324 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3325
3326 // Load the raw epoch value from the threadObj.
3327 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3328 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3329 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3330
3331 // Mask off the excluded information from the epoch.
3332 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3333
3334 // True branch, this is a virtual thread.
3335 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3336 set_control(vthread_not_equal_threadObj);
3337
3338 // Load the tid field from the vthread object.
3339 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3340
3341 // Continuation support determines if a virtual thread should be pinned.
3342 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3343 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3344
3345 // Load the raw epoch value from the vthread.
3346 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3347 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3348 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3349
3350 // Mask off the excluded information from the epoch.
3351 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3352
3353 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3354 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3355 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3356 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3357
3358 // False branch, vthread is excluded, no need to write epoch info.
3359 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3360
3361 // True branch, vthread is included, update epoch info.
3362 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3363 set_control(included);
3364
3365 // Get epoch value.
3366 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3367
3368 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3369 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3370 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3371
3372 // Compare the epoch in the vthread to the current epoch generation.
3373 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3374 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3375 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3376
3377 // False path, epoch is equal, checkpoint information is valid.
3378 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3379
3380 // True path, epoch is not equal, write a checkpoint for the vthread.
3381 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3382
3383 set_control(epoch_is_not_equal);
3384
3385 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3386 // The call also updates the native thread local thread id and the vthread with the current epoch.
3387 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3388 OptoRuntime::jfr_write_checkpoint_Type(),
3389 SharedRuntime::jfr_write_checkpoint(),
3390 "write_checkpoint", TypePtr::BOTTOM);
3391 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3392
3393 // vthread epoch != current epoch
3394 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3395 record_for_igvn(epoch_compare_rgn);
3396 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3397 record_for_igvn(epoch_compare_mem);
3398 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3399 record_for_igvn(epoch_compare_io);
3400
3401 // Update control and phi nodes.
3402 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3403 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3404 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3405 epoch_compare_mem->init_req(_false_path, input_memory_state);
3406 epoch_compare_io->init_req(_true_path, i_o());
3407 epoch_compare_io->init_req(_false_path, input_io_state);
3408
3409 // excluded != true
3410 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3411 record_for_igvn(exclude_compare_rgn);
3412 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3413 record_for_igvn(exclude_compare_mem);
3414 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3415 record_for_igvn(exclude_compare_io);
3416
3417 // Update control and phi nodes.
3418 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3419 exclude_compare_rgn->init_req(_false_path, excluded);
3420 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3421 exclude_compare_mem->init_req(_false_path, input_memory_state);
3422 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3423 exclude_compare_io->init_req(_false_path, input_io_state);
3424
3425 // vthread != threadObj
3426 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3427 record_for_igvn(vthread_compare_rgn);
3428 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3429 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3430 record_for_igvn(vthread_compare_io);
3431 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3432 record_for_igvn(tid);
3433 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3434 record_for_igvn(exclusion);
3435 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3436 record_for_igvn(pinVirtualThread);
3437
3438 // Update control and phi nodes.
3439 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3440 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3441 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3442 vthread_compare_mem->init_req(_false_path, input_memory_state);
3443 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3444 vthread_compare_io->init_req(_false_path, input_io_state);
3445 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3446 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3447 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3448 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3449 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3450 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3451
3452 // Update branch state.
3453 set_control(_gvn.transform(vthread_compare_rgn));
3454 set_all_memory(_gvn.transform(vthread_compare_mem));
3455 set_i_o(_gvn.transform(vthread_compare_io));
3456
3457 // Load the event writer oop by dereferencing the jobject handle.
3458 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3459 assert(klass_EventWriter->is_loaded(), "invariant");
3460 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3461 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3462 const TypeOopPtr* const xtype = aklass->as_instance_type();
3463 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3464 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3465
3466 // Load the current thread id from the event writer object.
3467 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3468 // Get the field offset to, conditionally, store an updated tid value later.
3469 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3470 const TypePtr* event_writer_tid_field_type = _gvn.type(event_writer_tid_field)->isa_ptr();
3471 // Get the field offset to, conditionally, store an updated exclusion value later.
3472 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3473 const TypePtr* event_writer_excluded_field_type = _gvn.type(event_writer_excluded_field)->isa_ptr();
3474 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3475 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3476 const TypePtr* event_writer_pin_field_type = _gvn.type(event_writer_pin_field)->isa_ptr();
3477
3478 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3479 record_for_igvn(event_writer_tid_compare_rgn);
3480 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3481 record_for_igvn(event_writer_tid_compare_mem);
3482 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3483 record_for_igvn(event_writer_tid_compare_io);
3484
3485 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3486 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3487 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3488 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3489
3490 // False path, tids are the same.
3491 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3492
3493 // True path, tid is not equal, need to update the tid in the event writer.
3494 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3495 record_for_igvn(tid_is_not_equal);
3496
3497 // Store the pin state to the event writer.
3498 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, event_writer_pin_field_type, MemNode::unordered);
3499
3500 // Store the exclusion state to the event writer.
3501 store_to_memory(tid_is_not_equal, event_writer_excluded_field, _gvn.transform(exclusion), T_BOOLEAN, event_writer_excluded_field_type, MemNode::unordered);
3502
3503 // Store the tid to the event writer.
3504 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, event_writer_tid_field_type, MemNode::unordered);
3505
3506 // Update control and phi nodes.
3507 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3508 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3509 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3510 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3511 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3512 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3513
3514 // Result of top level CFG, Memory, IO and Value.
3515 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3516 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3517 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3518 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3519
3520 // Result control.
3521 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3522 result_rgn->init_req(_false_path, jobj_is_null);
3523
3524 // Result memory.
3525 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3526 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3527
3528 // Result IO.
3529 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3530 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3531
3532 // Result value.
3533 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3534 result_value->init_req(_false_path, null()); // return null
3535
3536 // Set output state.
3537 set_control(_gvn.transform(result_rgn));
3538 set_all_memory(_gvn.transform(result_mem));
3539 set_i_o(_gvn.transform(result_io));
3540 set_result(result_rgn, result_value);
3541 return true;
3542 }
3543
3544 /*
3545 * The intrinsic is a model of this pseudo-code:
3546 *
3547 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3548 * if (carrierThread != thread) { // is virtual thread
3549 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3550 * bool excluded = vthread_epoch_raw & excluded_mask;
3551 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3552 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3553 * if (!excluded) {
3554 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3555 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3556 * }
3557 * Atomic::release_store(&tl->_vthread, true);
3558 * return;
3559 * }
3560 * Atomic::release_store(&tl->_vthread, false);
3561 */
3562 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3563 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3564
3565 Node* input_memory_state = reset_memory();
3566 set_all_memory(input_memory_state);
3567
3568 Node* excluded_mask = _gvn.intcon(32768);
3569 Node* epoch_mask = _gvn.intcon(32767);
3570
3571 Node* const carrierThread = generate_current_thread(jt);
3572 // If thread != carrierThread, this is a virtual thread.
3573 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3574 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3575 IfNode* iff_thread_not_equal_carrierThread =
3576 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3577
3578 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3579
3580 // False branch, is carrierThread.
3581 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3582 // Store release
3583 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3584
3585 set_all_memory(input_memory_state);
3586
3587 // True branch, is virtual thread.
3588 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3589 set_control(thread_not_equal_carrierThread);
3590
3591 // Load the raw epoch value from the vthread.
3592 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3593 Node* epoch_raw = access_load_at(thread, epoch_offset, TypeRawPtr::BOTTOM, TypeInt::CHAR, T_CHAR,
3594 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3595
3596 // Mask off the excluded information from the epoch.
3597 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3598
3599 // Load the tid field from the thread.
3600 Node* tid = load_field_from_object(thread, "tid", "J");
3601
3602 // Store the vthread tid to the jfr thread local.
3603 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3604 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, Compile::AliasIdxRaw, MemNode::unordered, true);
3605
3606 // Branch is_excluded to conditionalize updating the epoch .
3607 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3608 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3609 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3610
3611 // True branch, vthread is excluded, no need to write epoch info.
3612 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3613 set_control(excluded);
3614 Node* vthread_is_excluded = _gvn.intcon(1);
3615
3616 // False branch, vthread is included, update epoch info.
3617 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3618 set_control(included);
3619 Node* vthread_is_included = _gvn.intcon(0);
3620
3621 // Get epoch value.
3622 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3623
3624 // Store the vthread epoch to the jfr thread local.
3625 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3626 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, Compile::AliasIdxRaw, MemNode::unordered, true);
3627
3628 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3629 record_for_igvn(excluded_rgn);
3630 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3631 record_for_igvn(excluded_mem);
3632 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3633 record_for_igvn(exclusion);
3634
3635 // Merge the excluded control and memory.
3636 excluded_rgn->init_req(_true_path, excluded);
3637 excluded_rgn->init_req(_false_path, included);
3638 excluded_mem->init_req(_true_path, tid_memory);
3639 excluded_mem->init_req(_false_path, included_memory);
3640 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3641 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3642
3643 // Set intermediate state.
3644 set_control(_gvn.transform(excluded_rgn));
3645 set_all_memory(excluded_mem);
3646
3647 // Store the vthread exclusion state to the jfr thread local.
3648 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3649 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::unordered, true);
3650
3651 // Store release
3652 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3653
3654 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3655 record_for_igvn(thread_compare_rgn);
3656 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3657 record_for_igvn(thread_compare_mem);
3658 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3659 record_for_igvn(vthread);
3660
3661 // Merge the thread_compare control and memory.
3662 thread_compare_rgn->init_req(_true_path, control());
3663 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3664 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3665 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3666
3667 // Set output state.
3668 set_control(_gvn.transform(thread_compare_rgn));
3669 set_all_memory(_gvn.transform(thread_compare_mem));
3670 }
3671
3672 #endif // JFR_HAVE_INTRINSICS
3673
3674 //------------------------inline_native_currentCarrierThread------------------
3675 bool LibraryCallKit::inline_native_currentCarrierThread() {
3676 Node* junk = nullptr;
3677 set_result(generate_current_thread(junk));
3678 return true;
3679 }
3680
3681 //------------------------inline_native_currentThread------------------
3682 bool LibraryCallKit::inline_native_currentThread() {
3683 Node* junk = nullptr;
3684 set_result(generate_virtual_thread(junk));
3685 return true;
3686 }
3687
3688 //------------------------inline_native_setVthread------------------
3689 bool LibraryCallKit::inline_native_setCurrentThread() {
3690 assert(C->method()->changes_current_thread(),
3691 "method changes current Thread but is not annotated ChangesCurrentThread");
3692 Node* arr = argument(1);
3693 Node* thread = _gvn.transform(new ThreadLocalNode());
3694 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3695 Node* thread_obj_handle
3696 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3697 thread_obj_handle = _gvn.transform(thread_obj_handle);
3698 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3699 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3700 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3701 return true;
3702 }
3703
3704 const Type* LibraryCallKit::scopedValueCache_type() {
3705 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3706 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3707 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3708
3709 // Because we create the scopedValue cache lazily we have to make the
3710 // type of the result BotPTR.
3711 bool xk = etype->klass_is_exact();
3712 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3713 return objects_type;
3714 }
3715
3716 Node* LibraryCallKit::scopedValueCache_helper() {
3717 Node* thread = _gvn.transform(new ThreadLocalNode());
3718 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3719 // We cannot use immutable_memory() because we might flip onto a
3720 // different carrier thread, at which point we'll need to use that
3721 // carrier thread's cache.
3722 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3723 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3724 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3725 }
3726
3727 //------------------------inline_native_scopedValueCache------------------
3728 bool LibraryCallKit::inline_native_scopedValueCache() {
3729 Node* cache_obj_handle = scopedValueCache_helper();
3730 const Type* objects_type = scopedValueCache_type();
3731 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3732
3733 return true;
3734 }
3735
3736 //------------------------inline_native_setScopedValueCache------------------
3737 bool LibraryCallKit::inline_native_setScopedValueCache() {
3738 Node* arr = argument(0);
3739 Node* cache_obj_handle = scopedValueCache_helper();
3740 const Type* objects_type = scopedValueCache_type();
3741
3742 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3743 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3744
3745 return true;
3746 }
3747
3748 //------------------------inline_native_Continuation_pin and unpin-----------
3749
3750 // Shared implementation routine for both pin and unpin.
3751 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3752 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3753
3754 // Save input memory.
3755 Node* input_memory_state = reset_memory();
3756 set_all_memory(input_memory_state);
3757
3758 // TLS
3759 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3760 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3761 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3762
3763 // Null check the last continuation object.
3764 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3765 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3766 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3767
3768 // False path, last continuation is null.
3769 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3770
3771 // True path, last continuation is not null.
3772 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3773
3774 set_control(continuation_is_not_null);
3775
3776 // Load the pin count from the last continuation.
3777 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3778 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3779
3780 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3781 Node* pin_count_rhs;
3782 if (unpin) {
3783 pin_count_rhs = _gvn.intcon(0);
3784 } else {
3785 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3786 }
3787 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3788 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3789 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3790
3791 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3792 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3793 set_control(valid_pin_count);
3794
3795 Node* next_pin_count;
3796 if (unpin) {
3797 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3798 } else {
3799 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3800 }
3801
3802 Node* updated_pin_count_memory = store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, Compile::AliasIdxRaw, MemNode::unordered);
3803
3804 // True branch, pin count over/underflow.
3805 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3806 {
3807 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3808 // which will throw IllegalStateException for pin count over/underflow.
3809 PreserveJVMState pjvms(this);
3810 set_control(pin_count_over_underflow);
3811 set_all_memory(input_memory_state);
3812 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
3813 Deoptimization::Action_none);
3814 assert(stopped(), "invariant");
3815 }
3816
3817 // Result of top level CFG and Memory.
3818 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3819 record_for_igvn(result_rgn);
3820 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3821 record_for_igvn(result_mem);
3822
3823 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3824 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3825 result_mem->init_req(_true_path, _gvn.transform(updated_pin_count_memory));
3826 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3827
3828 // Set output state.
3829 set_control(_gvn.transform(result_rgn));
3830 set_all_memory(_gvn.transform(result_mem));
3831
3832 return true;
3833 }
3834
3835 //---------------------------load_mirror_from_klass----------------------------
3836 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3837 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3838 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3839 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3840 // mirror = ((OopHandle)mirror)->resolve();
3841 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3842 }
3843
3844 //-----------------------load_klass_from_mirror_common-------------------------
3845 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3846 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3847 // and branch to the given path on the region.
3848 // If never_see_null, take an uncommon trap on null, so we can optimistically
3849 // compile for the non-null case.
3850 // If the region is null, force never_see_null = true.
3851 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3852 bool never_see_null,
3853 RegionNode* region,
3854 int null_path,
3855 int offset) {
3856 if (region == nullptr) never_see_null = true;
3857 Node* p = basic_plus_adr(mirror, offset);
3858 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3859 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3860 Node* null_ctl = top();
3861 kls = null_check_oop(kls, &null_ctl, never_see_null);
3862 if (region != nullptr) {
3863 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3864 region->init_req(null_path, null_ctl);
3865 } else {
3866 assert(null_ctl == top(), "no loose ends");
3867 }
3868 return kls;
3869 }
3870
3871 //--------------------(inline_native_Class_query helpers)---------------------
3872 // Use this for JVM_ACC_INTERFACE.
3873 // Fall through if (mods & mask) == bits, take the guard otherwise.
3874 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3875 ByteSize offset, const Type* type, BasicType bt) {
3876 // Branch around if the given klass has the given modifier bit set.
3877 // Like generate_guard, adds a new path onto the region.
3878 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3879 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3880 Node* mask = intcon(modifier_mask);
3881 Node* bits = intcon(modifier_bits);
3882 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3883 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3884 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3885 return generate_fair_guard(bol, region);
3886 }
3887 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3888 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3889 Klass::access_flags_offset(), TypeInt::INT, T_INT);
3890 }
3891
3892 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3893 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3894 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3895 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3896 }
3897
3898 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3899 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3900 }
3901
3902 //-------------------------inline_native_Class_query-------------------
3903 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3904 const Type* return_type = TypeInt::BOOL;
3905 Node* prim_return_value = top(); // what happens if it's a primitive class?
3906 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3907 bool expect_prim = false; // most of these guys expect to work on refs
3908
3909 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3910
3911 Node* mirror = argument(0);
3912 Node* obj = top();
3913
3914 switch (id) {
3915 case vmIntrinsics::_isInstance:
3916 // nothing is an instance of a primitive type
3917 prim_return_value = intcon(0);
3918 obj = argument(1);
3919 break;
3920 case vmIntrinsics::_getModifiers:
3921 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3922 assert(is_power_of_2((int)JVM_ACC_WRITTEN_FLAGS+1), "change next line");
3923 return_type = TypeInt::make(0, JVM_ACC_WRITTEN_FLAGS, Type::WidenMin);
3924 break;
3925 case vmIntrinsics::_isInterface:
3926 prim_return_value = intcon(0);
3927 break;
3928 case vmIntrinsics::_isArray:
3929 prim_return_value = intcon(0);
3930 expect_prim = true; // cf. ObjectStreamClass.getClassSignature
3931 break;
3932 case vmIntrinsics::_isPrimitive:
3933 prim_return_value = intcon(1);
3934 expect_prim = true; // obviously
3935 break;
3936 case vmIntrinsics::_isHidden:
3937 prim_return_value = intcon(0);
3938 break;
3939 case vmIntrinsics::_getSuperclass:
3940 prim_return_value = null();
3941 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3942 break;
3943 case vmIntrinsics::_getClassAccessFlags:
3944 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3945 return_type = TypeInt::INT; // not bool! 6297094
3946 break;
3947 default:
3948 fatal_unexpected_iid(id);
3949 break;
3950 }
3951
3952 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3953 if (mirror_con == nullptr) return false; // cannot happen?
3954
3955 #ifndef PRODUCT
3956 if (C->print_intrinsics() || C->print_inlining()) {
3957 ciType* k = mirror_con->java_mirror_type();
3958 if (k) {
3959 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3960 k->print_name();
3961 tty->cr();
3962 }
3963 }
3964 #endif
3965
3966 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3967 RegionNode* region = new RegionNode(PATH_LIMIT);
3968 record_for_igvn(region);
3969 PhiNode* phi = new PhiNode(region, return_type);
3970
3971 // The mirror will never be null of Reflection.getClassAccessFlags, however
3972 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
3973 // if it is. See bug 4774291.
3974
3975 // For Reflection.getClassAccessFlags(), the null check occurs in
3976 // the wrong place; see inline_unsafe_access(), above, for a similar
3977 // situation.
3978 mirror = null_check(mirror);
3979 // If mirror or obj is dead, only null-path is taken.
3980 if (stopped()) return true;
3981
3982 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
3983
3984 // Now load the mirror's klass metaobject, and null-check it.
3985 // Side-effects region with the control path if the klass is null.
3986 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
3987 // If kls is null, we have a primitive mirror.
3988 phi->init_req(_prim_path, prim_return_value);
3989 if (stopped()) { set_result(region, phi); return true; }
3990 bool safe_for_replace = (region->in(_prim_path) == top());
3991
3992 Node* p; // handy temp
3993 Node* null_ctl;
3994
3995 // Now that we have the non-null klass, we can perform the real query.
3996 // For constant classes, the query will constant-fold in LoadNode::Value.
3997 Node* query_value = top();
3998 switch (id) {
3999 case vmIntrinsics::_isInstance:
4000 // nothing is an instance of a primitive type
4001 query_value = gen_instanceof(obj, kls, safe_for_replace);
4002 break;
4003
4004 case vmIntrinsics::_getModifiers:
4005 p = basic_plus_adr(kls, in_bytes(Klass::modifier_flags_offset()));
4006 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4007 break;
4008
4009 case vmIntrinsics::_isInterface:
4010 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4011 if (generate_interface_guard(kls, region) != nullptr)
4012 // A guard was added. If the guard is taken, it was an interface.
4013 phi->add_req(intcon(1));
4014 // If we fall through, it's a plain class.
4015 query_value = intcon(0);
4016 break;
4017
4018 case vmIntrinsics::_isArray:
4019 // (To verify this code sequence, check the asserts in JVM_IsArrayClass.)
4020 if (generate_array_guard(kls, region) != nullptr)
4021 // A guard was added. If the guard is taken, it was an array.
4022 phi->add_req(intcon(1));
4023 // If we fall through, it's a plain class.
4024 query_value = intcon(0);
4025 break;
4026
4027 case vmIntrinsics::_isPrimitive:
4028 query_value = intcon(0); // "normal" path produces false
4029 break;
4030
4031 case vmIntrinsics::_isHidden:
4032 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4033 if (generate_hidden_class_guard(kls, region) != nullptr)
4034 // A guard was added. If the guard is taken, it was an hidden class.
4035 phi->add_req(intcon(1));
4036 // If we fall through, it's a plain class.
4037 query_value = intcon(0);
4038 break;
4039
4040
4041 case vmIntrinsics::_getSuperclass:
4042 // The rules here are somewhat unfortunate, but we can still do better
4043 // with random logic than with a JNI call.
4044 // Interfaces store null or Object as _super, but must report null.
4045 // Arrays store an intermediate super as _super, but must report Object.
4046 // Other types can report the actual _super.
4047 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4048 if (generate_interface_guard(kls, region) != nullptr)
4049 // A guard was added. If the guard is taken, it was an interface.
4050 phi->add_req(null());
4051 if (generate_array_guard(kls, region) != nullptr)
4052 // A guard was added. If the guard is taken, it was an array.
4053 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4054 // If we fall through, it's a plain class. Get its _super.
4055 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4056 kls = _gvn.transform(LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4057 null_ctl = top();
4058 kls = null_check_oop(kls, &null_ctl);
4059 if (null_ctl != top()) {
4060 // If the guard is taken, Object.superClass is null (both klass and mirror).
4061 region->add_req(null_ctl);
4062 phi ->add_req(null());
4063 }
4064 if (!stopped()) {
4065 query_value = load_mirror_from_klass(kls);
4066 }
4067 break;
4068
4069 case vmIntrinsics::_getClassAccessFlags:
4070 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4071 query_value = make_load(nullptr, p, TypeInt::INT, T_INT, MemNode::unordered);
4072 break;
4073
4074 default:
4075 fatal_unexpected_iid(id);
4076 break;
4077 }
4078
4079 // Fall-through is the normal case of a query to a real class.
4080 phi->init_req(1, query_value);
4081 region->init_req(1, control());
4082
4083 C->set_has_split_ifs(true); // Has chance for split-if optimization
4084 set_result(region, phi);
4085 return true;
4086 }
4087
4088 //-------------------------inline_Class_cast-------------------
4089 bool LibraryCallKit::inline_Class_cast() {
4090 Node* mirror = argument(0); // Class
4091 Node* obj = argument(1);
4092 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4093 if (mirror_con == nullptr) {
4094 return false; // dead path (mirror->is_top()).
4095 }
4096 if (obj == nullptr || obj->is_top()) {
4097 return false; // dead path
4098 }
4099 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4100
4101 // First, see if Class.cast() can be folded statically.
4102 // java_mirror_type() returns non-null for compile-time Class constants.
4103 ciType* tm = mirror_con->java_mirror_type();
4104 if (tm != nullptr && tm->is_klass() &&
4105 tp != nullptr) {
4106 if (!tp->is_loaded()) {
4107 // Don't use intrinsic when class is not loaded.
4108 return false;
4109 } else {
4110 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4111 if (static_res == Compile::SSC_always_true) {
4112 // isInstance() is true - fold the code.
4113 set_result(obj);
4114 return true;
4115 } else if (static_res == Compile::SSC_always_false) {
4116 // Don't use intrinsic, have to throw ClassCastException.
4117 // If the reference is null, the non-intrinsic bytecode will
4118 // be optimized appropriately.
4119 return false;
4120 }
4121 }
4122 }
4123
4124 // Bailout intrinsic and do normal inlining if exception path is frequent.
4125 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4126 return false;
4127 }
4128
4129 // Generate dynamic checks.
4130 // Class.cast() is java implementation of _checkcast bytecode.
4131 // Do checkcast (Parse::do_checkcast()) optimizations here.
4132
4133 mirror = null_check(mirror);
4134 // If mirror is dead, only null-path is taken.
4135 if (stopped()) {
4136 return true;
4137 }
4138
4139 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4140 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4141 RegionNode* region = new RegionNode(PATH_LIMIT);
4142 record_for_igvn(region);
4143
4144 // Now load the mirror's klass metaobject, and null-check it.
4145 // If kls is null, we have a primitive mirror and
4146 // nothing is an instance of a primitive type.
4147 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4148
4149 Node* res = top();
4150 if (!stopped()) {
4151 Node* bad_type_ctrl = top();
4152 // Do checkcast optimizations.
4153 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4154 region->init_req(_bad_type_path, bad_type_ctrl);
4155 }
4156 if (region->in(_prim_path) != top() ||
4157 region->in(_bad_type_path) != top()) {
4158 // Let Interpreter throw ClassCastException.
4159 PreserveJVMState pjvms(this);
4160 set_control(_gvn.transform(region));
4161 uncommon_trap(Deoptimization::Reason_intrinsic,
4162 Deoptimization::Action_maybe_recompile);
4163 }
4164 if (!stopped()) {
4165 set_result(res);
4166 }
4167 return true;
4168 }
4169
4170
4171 //--------------------------inline_native_subtype_check------------------------
4172 // This intrinsic takes the JNI calls out of the heart of
4173 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4174 bool LibraryCallKit::inline_native_subtype_check() {
4175 // Pull both arguments off the stack.
4176 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4177 args[0] = argument(0);
4178 args[1] = argument(1);
4179 Node* klasses[2]; // corresponding Klasses: superk, subk
4180 klasses[0] = klasses[1] = top();
4181
4182 enum {
4183 // A full decision tree on {superc is prim, subc is prim}:
4184 _prim_0_path = 1, // {P,N} => false
4185 // {P,P} & superc!=subc => false
4186 _prim_same_path, // {P,P} & superc==subc => true
4187 _prim_1_path, // {N,P} => false
4188 _ref_subtype_path, // {N,N} & subtype check wins => true
4189 _both_ref_path, // {N,N} & subtype check loses => false
4190 PATH_LIMIT
4191 };
4192
4193 RegionNode* region = new RegionNode(PATH_LIMIT);
4194 Node* phi = new PhiNode(region, TypeInt::BOOL);
4195 record_for_igvn(region);
4196
4197 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4198 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4199 int class_klass_offset = java_lang_Class::klass_offset();
4200
4201 // First null-check both mirrors and load each mirror's klass metaobject.
4202 int which_arg;
4203 for (which_arg = 0; which_arg <= 1; which_arg++) {
4204 Node* arg = args[which_arg];
4205 arg = null_check(arg);
4206 if (stopped()) break;
4207 args[which_arg] = arg;
4208
4209 Node* p = basic_plus_adr(arg, class_klass_offset);
4210 Node* kls = LoadKlassNode::make(_gvn, nullptr, immutable_memory(), p, adr_type, kls_type);
4211 klasses[which_arg] = _gvn.transform(kls);
4212 }
4213
4214 // Having loaded both klasses, test each for null.
4215 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4216 for (which_arg = 0; which_arg <= 1; which_arg++) {
4217 Node* kls = klasses[which_arg];
4218 Node* null_ctl = top();
4219 kls = null_check_oop(kls, &null_ctl, never_see_null);
4220 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4221 region->init_req(prim_path, null_ctl);
4222 if (stopped()) break;
4223 klasses[which_arg] = kls;
4224 }
4225
4226 if (!stopped()) {
4227 // now we have two reference types, in klasses[0..1]
4228 Node* subk = klasses[1]; // the argument to isAssignableFrom
4229 Node* superk = klasses[0]; // the receiver
4230 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4231 // now we have a successful reference subtype check
4232 region->set_req(_ref_subtype_path, control());
4233 }
4234
4235 // If both operands are primitive (both klasses null), then
4236 // we must return true when they are identical primitives.
4237 // It is convenient to test this after the first null klass check.
4238 set_control(region->in(_prim_0_path)); // go back to first null check
4239 if (!stopped()) {
4240 // Since superc is primitive, make a guard for the superc==subc case.
4241 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4242 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4243 generate_guard(bol_eq, region, PROB_FAIR);
4244 if (region->req() == PATH_LIMIT+1) {
4245 // A guard was added. If the added guard is taken, superc==subc.
4246 region->swap_edges(PATH_LIMIT, _prim_same_path);
4247 region->del_req(PATH_LIMIT);
4248 }
4249 region->set_req(_prim_0_path, control()); // Not equal after all.
4250 }
4251
4252 // these are the only paths that produce 'true':
4253 phi->set_req(_prim_same_path, intcon(1));
4254 phi->set_req(_ref_subtype_path, intcon(1));
4255
4256 // pull together the cases:
4257 assert(region->req() == PATH_LIMIT, "sane region");
4258 for (uint i = 1; i < region->req(); i++) {
4259 Node* ctl = region->in(i);
4260 if (ctl == nullptr || ctl == top()) {
4261 region->set_req(i, top());
4262 phi ->set_req(i, top());
4263 } else if (phi->in(i) == nullptr) {
4264 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4265 }
4266 }
4267
4268 set_control(_gvn.transform(region));
4269 set_result(_gvn.transform(phi));
4270 return true;
4271 }
4272
4273 //---------------------generate_array_guard_common------------------------
4274 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4275 bool obj_array, bool not_array) {
4276
4277 if (stopped()) {
4278 return nullptr;
4279 }
4280
4281 // If obj_array/non_array==false/false:
4282 // Branch around if the given klass is in fact an array (either obj or prim).
4283 // If obj_array/non_array==false/true:
4284 // Branch around if the given klass is not an array klass of any kind.
4285 // If obj_array/non_array==true/true:
4286 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4287 // If obj_array/non_array==true/false:
4288 // Branch around if the kls is an oop array (Object[] or subtype)
4289 //
4290 // Like generate_guard, adds a new path onto the region.
4291 jint layout_con = 0;
4292 Node* layout_val = get_layout_helper(kls, layout_con);
4293 if (layout_val == nullptr) {
4294 bool query = (obj_array
4295 ? Klass::layout_helper_is_objArray(layout_con)
4296 : Klass::layout_helper_is_array(layout_con));
4297 if (query == not_array) {
4298 return nullptr; // never a branch
4299 } else { // always a branch
4300 Node* always_branch = control();
4301 if (region != nullptr)
4302 region->add_req(always_branch);
4303 set_control(top());
4304 return always_branch;
4305 }
4306 }
4307 // Now test the correct condition.
4308 jint nval = (obj_array
4309 ? (jint)(Klass::_lh_array_tag_type_value
4310 << Klass::_lh_array_tag_shift)
4311 : Klass::_lh_neutral_value);
4312 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4313 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4314 // invert the test if we are looking for a non-array
4315 if (not_array) btest = BoolTest(btest).negate();
4316 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4317 return generate_fair_guard(bol, region);
4318 }
4319
4320
4321 //-----------------------inline_native_newArray--------------------------
4322 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4323 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4324 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4325 Node* mirror;
4326 Node* count_val;
4327 if (uninitialized) {
4328 null_check_receiver();
4329 mirror = argument(1);
4330 count_val = argument(2);
4331 } else {
4332 mirror = argument(0);
4333 count_val = argument(1);
4334 }
4335
4336 mirror = null_check(mirror);
4337 // If mirror or obj is dead, only null-path is taken.
4338 if (stopped()) return true;
4339
4340 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4341 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4342 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4343 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4344 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4345
4346 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4347 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4348 result_reg, _slow_path);
4349 Node* normal_ctl = control();
4350 Node* no_array_ctl = result_reg->in(_slow_path);
4351
4352 // Generate code for the slow case. We make a call to newArray().
4353 set_control(no_array_ctl);
4354 if (!stopped()) {
4355 // Either the input type is void.class, or else the
4356 // array klass has not yet been cached. Either the
4357 // ensuing call will throw an exception, or else it
4358 // will cache the array klass for next time.
4359 PreserveJVMState pjvms(this);
4360 CallJavaNode* slow_call = nullptr;
4361 if (uninitialized) {
4362 // Generate optimized virtual call (holder class 'Unsafe' is final)
4363 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4364 } else {
4365 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4366 }
4367 Node* slow_result = set_results_for_java_call(slow_call);
4368 // this->control() comes from set_results_for_java_call
4369 result_reg->set_req(_slow_path, control());
4370 result_val->set_req(_slow_path, slow_result);
4371 result_io ->set_req(_slow_path, i_o());
4372 result_mem->set_req(_slow_path, reset_memory());
4373 }
4374
4375 set_control(normal_ctl);
4376 if (!stopped()) {
4377 // Normal case: The array type has been cached in the java.lang.Class.
4378 // The following call works fine even if the array type is polymorphic.
4379 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4380 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4381 result_reg->init_req(_normal_path, control());
4382 result_val->init_req(_normal_path, obj);
4383 result_io ->init_req(_normal_path, i_o());
4384 result_mem->init_req(_normal_path, reset_memory());
4385
4386 if (uninitialized) {
4387 // Mark the allocation so that zeroing is skipped
4388 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4389 alloc->maybe_set_complete(&_gvn);
4390 }
4391 }
4392
4393 // Return the combined state.
4394 set_i_o( _gvn.transform(result_io) );
4395 set_all_memory( _gvn.transform(result_mem));
4396
4397 C->set_has_split_ifs(true); // Has chance for split-if optimization
4398 set_result(result_reg, result_val);
4399 return true;
4400 }
4401
4402 //----------------------inline_native_getLength--------------------------
4403 // public static native int java.lang.reflect.Array.getLength(Object array);
4404 bool LibraryCallKit::inline_native_getLength() {
4405 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4406
4407 Node* array = null_check(argument(0));
4408 // If array is dead, only null-path is taken.
4409 if (stopped()) return true;
4410
4411 // Deoptimize if it is a non-array.
4412 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr);
4413
4414 if (non_array != nullptr) {
4415 PreserveJVMState pjvms(this);
4416 set_control(non_array);
4417 uncommon_trap(Deoptimization::Reason_intrinsic,
4418 Deoptimization::Action_maybe_recompile);
4419 }
4420
4421 // If control is dead, only non-array-path is taken.
4422 if (stopped()) return true;
4423
4424 // The works fine even if the array type is polymorphic.
4425 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4426 Node* result = load_array_length(array);
4427
4428 C->set_has_split_ifs(true); // Has chance for split-if optimization
4429 set_result(result);
4430 return true;
4431 }
4432
4433 //------------------------inline_array_copyOf----------------------------
4434 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4435 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4436 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4437 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4438
4439 // Get the arguments.
4440 Node* original = argument(0);
4441 Node* start = is_copyOfRange? argument(1): intcon(0);
4442 Node* end = is_copyOfRange? argument(2): argument(1);
4443 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4444
4445 Node* newcopy = nullptr;
4446
4447 // Set the original stack and the reexecute bit for the interpreter to reexecute
4448 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4449 { PreserveReexecuteState preexecs(this);
4450 jvms()->set_should_reexecute(true);
4451
4452 array_type_mirror = null_check(array_type_mirror);
4453 original = null_check(original);
4454
4455 // Check if a null path was taken unconditionally.
4456 if (stopped()) return true;
4457
4458 Node* orig_length = load_array_length(original);
4459
4460 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4461 klass_node = null_check(klass_node);
4462
4463 RegionNode* bailout = new RegionNode(1);
4464 record_for_igvn(bailout);
4465
4466 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4467 // Bail out if that is so.
4468 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4469 if (not_objArray != nullptr) {
4470 // Improve the klass node's type from the new optimistic assumption:
4471 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4472 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4473 Node* cast = new CastPPNode(control(), klass_node, akls);
4474 klass_node = _gvn.transform(cast);
4475 }
4476
4477 // Bail out if either start or end is negative.
4478 generate_negative_guard(start, bailout, &start);
4479 generate_negative_guard(end, bailout, &end);
4480
4481 Node* length = end;
4482 if (_gvn.type(start) != TypeInt::ZERO) {
4483 length = _gvn.transform(new SubINode(end, start));
4484 }
4485
4486 // Bail out if length is negative (i.e., if start > end).
4487 // Without this the new_array would throw
4488 // NegativeArraySizeException but IllegalArgumentException is what
4489 // should be thrown
4490 generate_negative_guard(length, bailout, &length);
4491
4492 // Bail out if start is larger than the original length
4493 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4494 generate_negative_guard(orig_tail, bailout, &orig_tail);
4495
4496 if (bailout->req() > 1) {
4497 PreserveJVMState pjvms(this);
4498 set_control(_gvn.transform(bailout));
4499 uncommon_trap(Deoptimization::Reason_intrinsic,
4500 Deoptimization::Action_maybe_recompile);
4501 }
4502
4503 if (!stopped()) {
4504 // How many elements will we copy from the original?
4505 // The answer is MinI(orig_tail, length).
4506 Node* moved = generate_min_max(vmIntrinsics::_min, orig_tail, length);
4507
4508 // Generate a direct call to the right arraycopy function(s).
4509 // We know the copy is disjoint but we might not know if the
4510 // oop stores need checking.
4511 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4512 // This will fail a store-check if x contains any non-nulls.
4513
4514 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4515 // loads/stores but it is legal only if we're sure the
4516 // Arrays.copyOf would succeed. So we need all input arguments
4517 // to the copyOf to be validated, including that the copy to the
4518 // new array won't trigger an ArrayStoreException. That subtype
4519 // check can be optimized if we know something on the type of
4520 // the input array from type speculation.
4521 if (_gvn.type(klass_node)->singleton()) {
4522 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4523 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4524
4525 int test = C->static_subtype_check(superk, subk);
4526 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4527 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4528 if (t_original->speculative_type() != nullptr) {
4529 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4530 }
4531 }
4532 }
4533
4534 bool validated = false;
4535 // Reason_class_check rather than Reason_intrinsic because we
4536 // want to intrinsify even if this traps.
4537 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4538 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4539
4540 if (not_subtype_ctrl != top()) {
4541 PreserveJVMState pjvms(this);
4542 set_control(not_subtype_ctrl);
4543 uncommon_trap(Deoptimization::Reason_class_check,
4544 Deoptimization::Action_make_not_entrant);
4545 assert(stopped(), "Should be stopped");
4546 }
4547 validated = true;
4548 }
4549
4550 if (!stopped()) {
4551 newcopy = new_array(klass_node, length, 0); // no arguments to push
4552
4553 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4554 load_object_klass(original), klass_node);
4555 if (!is_copyOfRange) {
4556 ac->set_copyof(validated);
4557 } else {
4558 ac->set_copyofrange(validated);
4559 }
4560 Node* n = _gvn.transform(ac);
4561 if (n == ac) {
4562 ac->connect_outputs(this);
4563 } else {
4564 assert(validated, "shouldn't transform if all arguments not validated");
4565 set_all_memory(n);
4566 }
4567 }
4568 }
4569 } // original reexecute is set back here
4570
4571 C->set_has_split_ifs(true); // Has chance for split-if optimization
4572 if (!stopped()) {
4573 set_result(newcopy);
4574 }
4575 return true;
4576 }
4577
4578
4579 //----------------------generate_virtual_guard---------------------------
4580 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4581 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4582 RegionNode* slow_region) {
4583 ciMethod* method = callee();
4584 int vtable_index = method->vtable_index();
4585 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4586 "bad index %d", vtable_index);
4587 // Get the Method* out of the appropriate vtable entry.
4588 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4589 vtable_index*vtableEntry::size_in_bytes() +
4590 in_bytes(vtableEntry::method_offset());
4591 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4592 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4593
4594 // Compare the target method with the expected method (e.g., Object.hashCode).
4595 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4596
4597 Node* native_call = makecon(native_call_addr);
4598 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4599 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4600
4601 return generate_slow_guard(test_native, slow_region);
4602 }
4603
4604 //-----------------------generate_method_call----------------------------
4605 // Use generate_method_call to make a slow-call to the real
4606 // method if the fast path fails. An alternative would be to
4607 // use a stub like OptoRuntime::slow_arraycopy_Java.
4608 // This only works for expanding the current library call,
4609 // not another intrinsic. (E.g., don't use this for making an
4610 // arraycopy call inside of the copyOf intrinsic.)
4611 CallJavaNode*
4612 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4613 // When compiling the intrinsic method itself, do not use this technique.
4614 guarantee(callee() != C->method(), "cannot make slow-call to self");
4615
4616 ciMethod* method = callee();
4617 // ensure the JVMS we have will be correct for this call
4618 guarantee(method_id == method->intrinsic_id(), "must match");
4619
4620 const TypeFunc* tf = TypeFunc::make(method);
4621 if (res_not_null) {
4622 assert(tf->return_type() == T_OBJECT, "");
4623 const TypeTuple* range = tf->range();
4624 const Type** fields = TypeTuple::fields(range->cnt());
4625 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4626 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4627 tf = TypeFunc::make(tf->domain(), new_range);
4628 }
4629 CallJavaNode* slow_call;
4630 if (is_static) {
4631 assert(!is_virtual, "");
4632 slow_call = new CallStaticJavaNode(C, tf,
4633 SharedRuntime::get_resolve_static_call_stub(), method);
4634 } else if (is_virtual) {
4635 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4636 int vtable_index = Method::invalid_vtable_index;
4637 if (UseInlineCaches) {
4638 // Suppress the vtable call
4639 } else {
4640 // hashCode and clone are not a miranda methods,
4641 // so the vtable index is fixed.
4642 // No need to use the linkResolver to get it.
4643 vtable_index = method->vtable_index();
4644 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4645 "bad index %d", vtable_index);
4646 }
4647 slow_call = new CallDynamicJavaNode(tf,
4648 SharedRuntime::get_resolve_virtual_call_stub(),
4649 method, vtable_index);
4650 } else { // neither virtual nor static: opt_virtual
4651 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4652 slow_call = new CallStaticJavaNode(C, tf,
4653 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4654 slow_call->set_optimized_virtual(true);
4655 }
4656 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4657 // To be able to issue a direct call (optimized virtual or virtual)
4658 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4659 // about the method being invoked should be attached to the call site to
4660 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4661 slow_call->set_override_symbolic_info(true);
4662 }
4663 set_arguments_for_java_call(slow_call);
4664 set_edges_for_java_call(slow_call);
4665 return slow_call;
4666 }
4667
4668
4669 /**
4670 * Build special case code for calls to hashCode on an object. This call may
4671 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4672 * slightly different code.
4673 */
4674 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4675 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4676 assert(!(is_virtual && is_static), "either virtual, special, or static");
4677
4678 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4679
4680 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4681 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4682 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4683 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4684 Node* obj = nullptr;
4685 if (!is_static) {
4686 // Check for hashing null object
4687 obj = null_check_receiver();
4688 if (stopped()) return true; // unconditionally null
4689 result_reg->init_req(_null_path, top());
4690 result_val->init_req(_null_path, top());
4691 } else {
4692 // Do a null check, and return zero if null.
4693 // System.identityHashCode(null) == 0
4694 obj = argument(0);
4695 Node* null_ctl = top();
4696 obj = null_check_oop(obj, &null_ctl);
4697 result_reg->init_req(_null_path, null_ctl);
4698 result_val->init_req(_null_path, _gvn.intcon(0));
4699 }
4700
4701 // Unconditionally null? Then return right away.
4702 if (stopped()) {
4703 set_control( result_reg->in(_null_path));
4704 if (!stopped())
4705 set_result(result_val->in(_null_path));
4706 return true;
4707 }
4708
4709 // We only go to the fast case code if we pass a number of guards. The
4710 // paths which do not pass are accumulated in the slow_region.
4711 RegionNode* slow_region = new RegionNode(1);
4712 record_for_igvn(slow_region);
4713
4714 // If this is a virtual call, we generate a funny guard. We pull out
4715 // the vtable entry corresponding to hashCode() from the target object.
4716 // If the target method which we are calling happens to be the native
4717 // Object hashCode() method, we pass the guard. We do not need this
4718 // guard for non-virtual calls -- the caller is known to be the native
4719 // Object hashCode().
4720 if (is_virtual) {
4721 // After null check, get the object's klass.
4722 Node* obj_klass = load_object_klass(obj);
4723 generate_virtual_guard(obj_klass, slow_region);
4724 }
4725
4726 // Get the header out of the object, use LoadMarkNode when available
4727 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4728 // The control of the load must be null. Otherwise, the load can move before
4729 // the null check after castPP removal.
4730 Node* no_ctrl = nullptr;
4731 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4732
4733 if (!UseObjectMonitorTable) {
4734 // Test the header to see if it is safe to read w.r.t. locking.
4735 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4736 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4737 if (LockingMode == LM_LIGHTWEIGHT) {
4738 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4739 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4740 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4741
4742 generate_slow_guard(test_monitor, slow_region);
4743 } else {
4744 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4745 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4746 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4747
4748 generate_slow_guard(test_not_unlocked, slow_region);
4749 }
4750 }
4751
4752 // Get the hash value and check to see that it has been properly assigned.
4753 // We depend on hash_mask being at most 32 bits and avoid the use of
4754 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4755 // vm: see markWord.hpp.
4756 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4757 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4758 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4759 // This hack lets the hash bits live anywhere in the mark object now, as long
4760 // as the shift drops the relevant bits into the low 32 bits. Note that
4761 // Java spec says that HashCode is an int so there's no point in capturing
4762 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4763 hshifted_header = ConvX2I(hshifted_header);
4764 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4765
4766 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4767 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4768 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4769
4770 generate_slow_guard(test_assigned, slow_region);
4771
4772 Node* init_mem = reset_memory();
4773 // fill in the rest of the null path:
4774 result_io ->init_req(_null_path, i_o());
4775 result_mem->init_req(_null_path, init_mem);
4776
4777 result_val->init_req(_fast_path, hash_val);
4778 result_reg->init_req(_fast_path, control());
4779 result_io ->init_req(_fast_path, i_o());
4780 result_mem->init_req(_fast_path, init_mem);
4781
4782 // Generate code for the slow case. We make a call to hashCode().
4783 set_control(_gvn.transform(slow_region));
4784 if (!stopped()) {
4785 // No need for PreserveJVMState, because we're using up the present state.
4786 set_all_memory(init_mem);
4787 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4788 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4789 Node* slow_result = set_results_for_java_call(slow_call);
4790 // this->control() comes from set_results_for_java_call
4791 result_reg->init_req(_slow_path, control());
4792 result_val->init_req(_slow_path, slow_result);
4793 result_io ->set_req(_slow_path, i_o());
4794 result_mem ->set_req(_slow_path, reset_memory());
4795 }
4796
4797 // Return the combined state.
4798 set_i_o( _gvn.transform(result_io) );
4799 set_all_memory( _gvn.transform(result_mem));
4800
4801 set_result(result_reg, result_val);
4802 return true;
4803 }
4804
4805 //---------------------------inline_native_getClass----------------------------
4806 // public final native Class<?> java.lang.Object.getClass();
4807 //
4808 // Build special case code for calls to getClass on an object.
4809 bool LibraryCallKit::inline_native_getClass() {
4810 Node* obj = null_check_receiver();
4811 if (stopped()) return true;
4812 set_result(load_mirror_from_klass(load_object_klass(obj)));
4813 return true;
4814 }
4815
4816 //-----------------inline_native_Reflection_getCallerClass---------------------
4817 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4818 //
4819 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4820 //
4821 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4822 // in that it must skip particular security frames and checks for
4823 // caller sensitive methods.
4824 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4825 #ifndef PRODUCT
4826 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4827 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4828 }
4829 #endif
4830
4831 if (!jvms()->has_method()) {
4832 #ifndef PRODUCT
4833 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4834 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4835 }
4836 #endif
4837 return false;
4838 }
4839
4840 // Walk back up the JVM state to find the caller at the required
4841 // depth.
4842 JVMState* caller_jvms = jvms();
4843
4844 // Cf. JVM_GetCallerClass
4845 // NOTE: Start the loop at depth 1 because the current JVM state does
4846 // not include the Reflection.getCallerClass() frame.
4847 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4848 ciMethod* m = caller_jvms->method();
4849 switch (n) {
4850 case 0:
4851 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4852 break;
4853 case 1:
4854 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4855 if (!m->caller_sensitive()) {
4856 #ifndef PRODUCT
4857 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4858 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4859 }
4860 #endif
4861 return false; // bail-out; let JVM_GetCallerClass do the work
4862 }
4863 break;
4864 default:
4865 if (!m->is_ignored_by_security_stack_walk()) {
4866 // We have reached the desired frame; return the holder class.
4867 // Acquire method holder as java.lang.Class and push as constant.
4868 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4869 ciInstance* caller_mirror = caller_klass->java_mirror();
4870 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4871
4872 #ifndef PRODUCT
4873 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4874 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());
4875 tty->print_cr(" JVM state at this point:");
4876 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4877 ciMethod* m = jvms()->of_depth(i)->method();
4878 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4879 }
4880 }
4881 #endif
4882 return true;
4883 }
4884 break;
4885 }
4886 }
4887
4888 #ifndef PRODUCT
4889 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4890 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4891 tty->print_cr(" JVM state at this point:");
4892 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4893 ciMethod* m = jvms()->of_depth(i)->method();
4894 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4895 }
4896 }
4897 #endif
4898
4899 return false; // bail-out; let JVM_GetCallerClass do the work
4900 }
4901
4902 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4903 Node* arg = argument(0);
4904 Node* result = nullptr;
4905
4906 switch (id) {
4907 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4908 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4909 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4910 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4911 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4912 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4913
4914 case vmIntrinsics::_doubleToLongBits: {
4915 // two paths (plus control) merge in a wood
4916 RegionNode *r = new RegionNode(3);
4917 Node *phi = new PhiNode(r, TypeLong::LONG);
4918
4919 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4920 // Build the boolean node
4921 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4922
4923 // Branch either way.
4924 // NaN case is less traveled, which makes all the difference.
4925 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4926 Node *opt_isnan = _gvn.transform(ifisnan);
4927 assert( opt_isnan->is_If(), "Expect an IfNode");
4928 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4929 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4930
4931 set_control(iftrue);
4932
4933 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4934 Node *slow_result = longcon(nan_bits); // return NaN
4935 phi->init_req(1, _gvn.transform( slow_result ));
4936 r->init_req(1, iftrue);
4937
4938 // Else fall through
4939 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4940 set_control(iffalse);
4941
4942 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4943 r->init_req(2, iffalse);
4944
4945 // Post merge
4946 set_control(_gvn.transform(r));
4947 record_for_igvn(r);
4948
4949 C->set_has_split_ifs(true); // Has chance for split-if optimization
4950 result = phi;
4951 assert(result->bottom_type()->isa_long(), "must be");
4952 break;
4953 }
4954
4955 case vmIntrinsics::_floatToIntBits: {
4956 // two paths (plus control) merge in a wood
4957 RegionNode *r = new RegionNode(3);
4958 Node *phi = new PhiNode(r, TypeInt::INT);
4959
4960 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4961 // Build the boolean node
4962 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4963
4964 // Branch either way.
4965 // NaN case is less traveled, which makes all the difference.
4966 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4967 Node *opt_isnan = _gvn.transform(ifisnan);
4968 assert( opt_isnan->is_If(), "Expect an IfNode");
4969 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4970 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4971
4972 set_control(iftrue);
4973
4974 static const jint nan_bits = 0x7fc00000;
4975 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
4976 phi->init_req(1, _gvn.transform( slow_result ));
4977 r->init_req(1, iftrue);
4978
4979 // Else fall through
4980 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4981 set_control(iffalse);
4982
4983 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
4984 r->init_req(2, iffalse);
4985
4986 // Post merge
4987 set_control(_gvn.transform(r));
4988 record_for_igvn(r);
4989
4990 C->set_has_split_ifs(true); // Has chance for split-if optimization
4991 result = phi;
4992 assert(result->bottom_type()->isa_int(), "must be");
4993 break;
4994 }
4995
4996 default:
4997 fatal_unexpected_iid(id);
4998 break;
4999 }
5000 set_result(_gvn.transform(result));
5001 return true;
5002 }
5003
5004 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5005 Node* arg = argument(0);
5006 Node* result = nullptr;
5007
5008 switch (id) {
5009 case vmIntrinsics::_floatIsInfinite:
5010 result = new IsInfiniteFNode(arg);
5011 break;
5012 case vmIntrinsics::_floatIsFinite:
5013 result = new IsFiniteFNode(arg);
5014 break;
5015 case vmIntrinsics::_doubleIsInfinite:
5016 result = new IsInfiniteDNode(arg);
5017 break;
5018 case vmIntrinsics::_doubleIsFinite:
5019 result = new IsFiniteDNode(arg);
5020 break;
5021 default:
5022 fatal_unexpected_iid(id);
5023 break;
5024 }
5025 set_result(_gvn.transform(result));
5026 return true;
5027 }
5028
5029 //----------------------inline_unsafe_copyMemory-------------------------
5030 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5031
5032 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5033 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5034 const Type* base_t = gvn.type(base);
5035
5036 bool in_native = (base_t == TypePtr::NULL_PTR);
5037 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5038 bool is_mixed = !in_heap && !in_native;
5039
5040 if (is_mixed) {
5041 return true; // mixed accesses can touch both on-heap and off-heap memory
5042 }
5043 if (in_heap) {
5044 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5045 if (!is_prim_array) {
5046 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5047 // there's not enough type information available to determine proper memory slice for it.
5048 return true;
5049 }
5050 }
5051 return false;
5052 }
5053
5054 bool LibraryCallKit::inline_unsafe_copyMemory() {
5055 if (callee()->is_static()) return false; // caller must have the capability!
5056 null_check_receiver(); // null-check receiver
5057 if (stopped()) return true;
5058
5059 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5060
5061 Node* src_base = argument(1); // type: oop
5062 Node* src_off = ConvL2X(argument(2)); // type: long
5063 Node* dst_base = argument(4); // type: oop
5064 Node* dst_off = ConvL2X(argument(5)); // type: long
5065 Node* size = ConvL2X(argument(7)); // type: long
5066
5067 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5068 "fieldOffset must be byte-scaled");
5069
5070 Node* src_addr = make_unsafe_address(src_base, src_off);
5071 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5072
5073 Node* thread = _gvn.transform(new ThreadLocalNode());
5074 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5075 BasicType doing_unsafe_access_bt = T_BYTE;
5076 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5077
5078 // update volatile field
5079 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5080
5081 int flags = RC_LEAF | RC_NO_FP;
5082
5083 const TypePtr* dst_type = TypePtr::BOTTOM;
5084
5085 // Adjust memory effects of the runtime call based on input values.
5086 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5087 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5088 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5089
5090 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5091 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5092 flags |= RC_NARROW_MEM; // narrow in memory
5093 }
5094 }
5095
5096 // Call it. Note that the length argument is not scaled.
5097 make_runtime_call(flags,
5098 OptoRuntime::fast_arraycopy_Type(),
5099 StubRoutines::unsafe_arraycopy(),
5100 "unsafe_arraycopy",
5101 dst_type,
5102 src_addr, dst_addr, size XTOP);
5103
5104 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5105
5106 return true;
5107 }
5108
5109 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5110 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5111 bool LibraryCallKit::inline_unsafe_setMemory() {
5112 if (callee()->is_static()) return false; // caller must have the capability!
5113 null_check_receiver(); // null-check receiver
5114 if (stopped()) return true;
5115
5116 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5117
5118 Node* dst_base = argument(1); // type: oop
5119 Node* dst_off = ConvL2X(argument(2)); // type: long
5120 Node* size = ConvL2X(argument(4)); // type: long
5121 Node* byte = argument(6); // type: byte
5122
5123 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5124 "fieldOffset must be byte-scaled");
5125
5126 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5127
5128 Node* thread = _gvn.transform(new ThreadLocalNode());
5129 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5130 BasicType doing_unsafe_access_bt = T_BYTE;
5131 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5132
5133 // update volatile field
5134 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5135
5136 int flags = RC_LEAF | RC_NO_FP;
5137
5138 const TypePtr* dst_type = TypePtr::BOTTOM;
5139
5140 // Adjust memory effects of the runtime call based on input values.
5141 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5142 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5143
5144 flags |= RC_NARROW_MEM; // narrow in memory
5145 }
5146
5147 // Call it. Note that the length argument is not scaled.
5148 make_runtime_call(flags,
5149 OptoRuntime::make_setmemory_Type(),
5150 StubRoutines::unsafe_setmemory(),
5151 "unsafe_setmemory",
5152 dst_type,
5153 dst_addr, size XTOP, byte);
5154
5155 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, Compile::AliasIdxRaw, MemNode::unordered);
5156
5157 return true;
5158 }
5159
5160 #undef XTOP
5161
5162 //------------------------clone_coping-----------------------------------
5163 // Helper function for inline_native_clone.
5164 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5165 assert(obj_size != nullptr, "");
5166 Node* raw_obj = alloc_obj->in(1);
5167 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5168
5169 AllocateNode* alloc = nullptr;
5170 if (ReduceBulkZeroing &&
5171 // If we are implementing an array clone without knowing its source type
5172 // (can happen when compiling the array-guarded branch of a reflective
5173 // Object.clone() invocation), initialize the array within the allocation.
5174 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5175 // to a runtime clone call that assumes fully initialized source arrays.
5176 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5177 // We will be completely responsible for initializing this object -
5178 // mark Initialize node as complete.
5179 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5180 // The object was just allocated - there should be no any stores!
5181 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5182 // Mark as complete_with_arraycopy so that on AllocateNode
5183 // expansion, we know this AllocateNode is initialized by an array
5184 // copy and a StoreStore barrier exists after the array copy.
5185 alloc->initialization()->set_complete_with_arraycopy();
5186 }
5187
5188 Node* size = _gvn.transform(obj_size);
5189 access_clone(obj, alloc_obj, size, is_array);
5190
5191 // Do not let reads from the cloned object float above the arraycopy.
5192 if (alloc != nullptr) {
5193 // Do not let stores that initialize this object be reordered with
5194 // a subsequent store that would make this object accessible by
5195 // other threads.
5196 // Record what AllocateNode this StoreStore protects so that
5197 // escape analysis can go from the MemBarStoreStoreNode to the
5198 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5199 // based on the escape status of the AllocateNode.
5200 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5201 } else {
5202 insert_mem_bar(Op_MemBarCPUOrder);
5203 }
5204 }
5205
5206 //------------------------inline_native_clone----------------------------
5207 // protected native Object java.lang.Object.clone();
5208 //
5209 // Here are the simple edge cases:
5210 // null receiver => normal trap
5211 // virtual and clone was overridden => slow path to out-of-line clone
5212 // not cloneable or finalizer => slow path to out-of-line Object.clone
5213 //
5214 // The general case has two steps, allocation and copying.
5215 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5216 //
5217 // Copying also has two cases, oop arrays and everything else.
5218 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5219 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5220 //
5221 // These steps fold up nicely if and when the cloned object's klass
5222 // can be sharply typed as an object array, a type array, or an instance.
5223 //
5224 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5225 PhiNode* result_val;
5226
5227 // Set the reexecute bit for the interpreter to reexecute
5228 // the bytecode that invokes Object.clone if deoptimization happens.
5229 { PreserveReexecuteState preexecs(this);
5230 jvms()->set_should_reexecute(true);
5231
5232 Node* obj = null_check_receiver();
5233 if (stopped()) return true;
5234
5235 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5236
5237 // If we are going to clone an instance, we need its exact type to
5238 // know the number and types of fields to convert the clone to
5239 // loads/stores. Maybe a speculative type can help us.
5240 if (!obj_type->klass_is_exact() &&
5241 obj_type->speculative_type() != nullptr &&
5242 obj_type->speculative_type()->is_instance_klass()) {
5243 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5244 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5245 !spec_ik->has_injected_fields()) {
5246 if (!obj_type->isa_instptr() ||
5247 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5248 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5249 }
5250 }
5251 }
5252
5253 // Conservatively insert a memory barrier on all memory slices.
5254 // Do not let writes into the original float below the clone.
5255 insert_mem_bar(Op_MemBarCPUOrder);
5256
5257 // paths into result_reg:
5258 enum {
5259 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5260 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5261 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5262 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5263 PATH_LIMIT
5264 };
5265 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5266 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5267 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5268 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5269 record_for_igvn(result_reg);
5270
5271 Node* obj_klass = load_object_klass(obj);
5272 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr);
5273 if (array_ctl != nullptr) {
5274 // It's an array.
5275 PreserveJVMState pjvms(this);
5276 set_control(array_ctl);
5277 Node* obj_length = load_array_length(obj);
5278 Node* array_size = nullptr; // Size of the array without object alignment padding.
5279 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5280
5281 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5282 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5283 // If it is an oop array, it requires very special treatment,
5284 // because gc barriers are required when accessing the array.
5285 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5286 if (is_obja != nullptr) {
5287 PreserveJVMState pjvms2(this);
5288 set_control(is_obja);
5289 // Generate a direct call to the right arraycopy function(s).
5290 // Clones are always tightly coupled.
5291 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5292 ac->set_clone_oop_array();
5293 Node* n = _gvn.transform(ac);
5294 assert(n == ac, "cannot disappear");
5295 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5296
5297 result_reg->init_req(_objArray_path, control());
5298 result_val->init_req(_objArray_path, alloc_obj);
5299 result_i_o ->set_req(_objArray_path, i_o());
5300 result_mem ->set_req(_objArray_path, reset_memory());
5301 }
5302 }
5303 // Otherwise, there are no barriers to worry about.
5304 // (We can dispense with card marks if we know the allocation
5305 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5306 // causes the non-eden paths to take compensating steps to
5307 // simulate a fresh allocation, so that no further
5308 // card marks are required in compiled code to initialize
5309 // the object.)
5310
5311 if (!stopped()) {
5312 copy_to_clone(obj, alloc_obj, array_size, true);
5313
5314 // Present the results of the copy.
5315 result_reg->init_req(_array_path, control());
5316 result_val->init_req(_array_path, alloc_obj);
5317 result_i_o ->set_req(_array_path, i_o());
5318 result_mem ->set_req(_array_path, reset_memory());
5319 }
5320 }
5321
5322 // We only go to the instance fast case code if we pass a number of guards.
5323 // The paths which do not pass are accumulated in the slow_region.
5324 RegionNode* slow_region = new RegionNode(1);
5325 record_for_igvn(slow_region);
5326 if (!stopped()) {
5327 // It's an instance (we did array above). Make the slow-path tests.
5328 // If this is a virtual call, we generate a funny guard. We grab
5329 // the vtable entry corresponding to clone() from the target object.
5330 // If the target method which we are calling happens to be the
5331 // Object clone() method, we pass the guard. We do not need this
5332 // guard for non-virtual calls; the caller is known to be the native
5333 // Object clone().
5334 if (is_virtual) {
5335 generate_virtual_guard(obj_klass, slow_region);
5336 }
5337
5338 // The object must be easily cloneable and must not have a finalizer.
5339 // Both of these conditions may be checked in a single test.
5340 // We could optimize the test further, but we don't care.
5341 generate_misc_flags_guard(obj_klass,
5342 // Test both conditions:
5343 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5344 // Must be cloneable but not finalizer:
5345 KlassFlags::_misc_is_cloneable_fast,
5346 slow_region);
5347 }
5348
5349 if (!stopped()) {
5350 // It's an instance, and it passed the slow-path tests.
5351 PreserveJVMState pjvms(this);
5352 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5353 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5354 // is reexecuted if deoptimization occurs and there could be problems when merging
5355 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5356 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5357
5358 copy_to_clone(obj, alloc_obj, obj_size, false);
5359
5360 // Present the results of the slow call.
5361 result_reg->init_req(_instance_path, control());
5362 result_val->init_req(_instance_path, alloc_obj);
5363 result_i_o ->set_req(_instance_path, i_o());
5364 result_mem ->set_req(_instance_path, reset_memory());
5365 }
5366
5367 // Generate code for the slow case. We make a call to clone().
5368 set_control(_gvn.transform(slow_region));
5369 if (!stopped()) {
5370 PreserveJVMState pjvms(this);
5371 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5372 // We need to deoptimize on exception (see comment above)
5373 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5374 // this->control() comes from set_results_for_java_call
5375 result_reg->init_req(_slow_path, control());
5376 result_val->init_req(_slow_path, slow_result);
5377 result_i_o ->set_req(_slow_path, i_o());
5378 result_mem ->set_req(_slow_path, reset_memory());
5379 }
5380
5381 // Return the combined state.
5382 set_control( _gvn.transform(result_reg));
5383 set_i_o( _gvn.transform(result_i_o));
5384 set_all_memory( _gvn.transform(result_mem));
5385 } // original reexecute is set back here
5386
5387 set_result(_gvn.transform(result_val));
5388 return true;
5389 }
5390
5391 // If we have a tightly coupled allocation, the arraycopy may take care
5392 // of the array initialization. If one of the guards we insert between
5393 // the allocation and the arraycopy causes a deoptimization, an
5394 // uninitialized array will escape the compiled method. To prevent that
5395 // we set the JVM state for uncommon traps between the allocation and
5396 // the arraycopy to the state before the allocation so, in case of
5397 // deoptimization, we'll reexecute the allocation and the
5398 // initialization.
5399 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5400 if (alloc != nullptr) {
5401 ciMethod* trap_method = alloc->jvms()->method();
5402 int trap_bci = alloc->jvms()->bci();
5403
5404 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5405 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5406 // Make sure there's no store between the allocation and the
5407 // arraycopy otherwise visible side effects could be rexecuted
5408 // in case of deoptimization and cause incorrect execution.
5409 bool no_interfering_store = true;
5410 Node* mem = alloc->in(TypeFunc::Memory);
5411 if (mem->is_MergeMem()) {
5412 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5413 Node* n = mms.memory();
5414 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5415 assert(n->is_Store(), "what else?");
5416 no_interfering_store = false;
5417 break;
5418 }
5419 }
5420 } else {
5421 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5422 Node* n = mms.memory();
5423 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5424 assert(n->is_Store(), "what else?");
5425 no_interfering_store = false;
5426 break;
5427 }
5428 }
5429 }
5430
5431 if (no_interfering_store) {
5432 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5433
5434 JVMState* saved_jvms = jvms();
5435 saved_reexecute_sp = _reexecute_sp;
5436
5437 set_jvms(sfpt->jvms());
5438 _reexecute_sp = jvms()->sp();
5439
5440 return saved_jvms;
5441 }
5442 }
5443 }
5444 return nullptr;
5445 }
5446
5447 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5448 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5449 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5450 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5451 uint size = alloc->req();
5452 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5453 old_jvms->set_map(sfpt);
5454 for (uint i = 0; i < size; i++) {
5455 sfpt->init_req(i, alloc->in(i));
5456 }
5457 // re-push array length for deoptimization
5458 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5459 old_jvms->set_sp(old_jvms->sp()+1);
5460 old_jvms->set_monoff(old_jvms->monoff()+1);
5461 old_jvms->set_scloff(old_jvms->scloff()+1);
5462 old_jvms->set_endoff(old_jvms->endoff()+1);
5463 old_jvms->set_should_reexecute(true);
5464
5465 sfpt->set_i_o(map()->i_o());
5466 sfpt->set_memory(map()->memory());
5467 sfpt->set_control(map()->control());
5468 return sfpt;
5469 }
5470
5471 // In case of a deoptimization, we restart execution at the
5472 // allocation, allocating a new array. We would leave an uninitialized
5473 // array in the heap that GCs wouldn't expect. Move the allocation
5474 // after the traps so we don't allocate the array if we
5475 // deoptimize. This is possible because tightly_coupled_allocation()
5476 // guarantees there's no observer of the allocated array at this point
5477 // and the control flow is simple enough.
5478 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5479 int saved_reexecute_sp, uint new_idx) {
5480 if (saved_jvms_before_guards != nullptr && !stopped()) {
5481 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5482
5483 assert(alloc != nullptr, "only with a tightly coupled allocation");
5484 // restore JVM state to the state at the arraycopy
5485 saved_jvms_before_guards->map()->set_control(map()->control());
5486 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5487 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5488 // If we've improved the types of some nodes (null check) while
5489 // emitting the guards, propagate them to the current state
5490 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5491 set_jvms(saved_jvms_before_guards);
5492 _reexecute_sp = saved_reexecute_sp;
5493
5494 // Remove the allocation from above the guards
5495 CallProjections callprojs;
5496 alloc->extract_projections(&callprojs, true);
5497 InitializeNode* init = alloc->initialization();
5498 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5499 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5500 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5501
5502 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5503 // the allocation (i.e. is only valid if the allocation succeeds):
5504 // 1) replace CastIINode with AllocateArrayNode's length here
5505 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5506 //
5507 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5508 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5509 Node* init_control = init->proj_out(TypeFunc::Control);
5510 Node* alloc_length = alloc->Ideal_length();
5511 #ifdef ASSERT
5512 Node* prev_cast = nullptr;
5513 #endif
5514 for (uint i = 0; i < init_control->outcnt(); i++) {
5515 Node* init_out = init_control->raw_out(i);
5516 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5517 #ifdef ASSERT
5518 if (prev_cast == nullptr) {
5519 prev_cast = init_out;
5520 } else {
5521 if (prev_cast->cmp(*init_out) == false) {
5522 prev_cast->dump();
5523 init_out->dump();
5524 assert(false, "not equal CastIINode");
5525 }
5526 }
5527 #endif
5528 C->gvn_replace_by(init_out, alloc_length);
5529 }
5530 }
5531 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5532
5533 // move the allocation here (after the guards)
5534 _gvn.hash_delete(alloc);
5535 alloc->set_req(TypeFunc::Control, control());
5536 alloc->set_req(TypeFunc::I_O, i_o());
5537 Node *mem = reset_memory();
5538 set_all_memory(mem);
5539 alloc->set_req(TypeFunc::Memory, mem);
5540 set_control(init->proj_out_or_null(TypeFunc::Control));
5541 set_i_o(callprojs.fallthrough_ioproj);
5542
5543 // Update memory as done in GraphKit::set_output_for_allocation()
5544 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5545 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5546 if (ary_type->isa_aryptr() && length_type != nullptr) {
5547 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5548 }
5549 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5550 int elemidx = C->get_alias_index(telemref);
5551 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5552 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5553
5554 Node* allocx = _gvn.transform(alloc);
5555 assert(allocx == alloc, "where has the allocation gone?");
5556 assert(dest->is_CheckCastPP(), "not an allocation result?");
5557
5558 _gvn.hash_delete(dest);
5559 dest->set_req(0, control());
5560 Node* destx = _gvn.transform(dest);
5561 assert(destx == dest, "where has the allocation result gone?");
5562
5563 array_ideal_length(alloc, ary_type, true);
5564 }
5565 }
5566
5567 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5568 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5569 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5570 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5571 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5572 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5573 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5574 JVMState* saved_jvms_before_guards) {
5575 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5576 // There is at least one unrelated uncommon trap which needs to be replaced.
5577 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5578
5579 JVMState* saved_jvms = jvms();
5580 const int saved_reexecute_sp = _reexecute_sp;
5581 set_jvms(sfpt->jvms());
5582 _reexecute_sp = jvms()->sp();
5583
5584 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5585
5586 // Restore state
5587 set_jvms(saved_jvms);
5588 _reexecute_sp = saved_reexecute_sp;
5589 }
5590 }
5591
5592 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5593 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5594 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5595 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5596 while (if_proj->is_IfProj()) {
5597 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5598 if (uncommon_trap != nullptr) {
5599 create_new_uncommon_trap(uncommon_trap);
5600 }
5601 assert(if_proj->in(0)->is_If(), "must be If");
5602 if_proj = if_proj->in(0)->in(0);
5603 }
5604 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5605 "must have reached control projection of init node");
5606 }
5607
5608 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5609 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5610 assert(trap_request != 0, "no valid UCT trap request");
5611 PreserveJVMState pjvms(this);
5612 set_control(uncommon_trap_call->in(0));
5613 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5614 Deoptimization::trap_request_action(trap_request));
5615 assert(stopped(), "Should be stopped");
5616 _gvn.hash_delete(uncommon_trap_call);
5617 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5618 }
5619
5620 // Common checks for array sorting intrinsics arguments.
5621 // Returns `true` if checks passed.
5622 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5623 // check address of the class
5624 if (elementType == nullptr || elementType->is_top()) {
5625 return false; // dead path
5626 }
5627 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5628 if (elem_klass == nullptr) {
5629 return false; // dead path
5630 }
5631 // java_mirror_type() returns non-null for compile-time Class constants only
5632 ciType* elem_type = elem_klass->java_mirror_type();
5633 if (elem_type == nullptr) {
5634 return false;
5635 }
5636 bt = elem_type->basic_type();
5637 // Disable the intrinsic if the CPU does not support SIMD sort
5638 if (!Matcher::supports_simd_sort(bt)) {
5639 return false;
5640 }
5641 // check address of the array
5642 if (obj == nullptr || obj->is_top()) {
5643 return false; // dead path
5644 }
5645 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5646 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5647 return false; // failed input validation
5648 }
5649 return true;
5650 }
5651
5652 //------------------------------inline_array_partition-----------------------
5653 bool LibraryCallKit::inline_array_partition() {
5654 address stubAddr = StubRoutines::select_array_partition_function();
5655 if (stubAddr == nullptr) {
5656 return false; // Intrinsic's stub is not implemented on this platform
5657 }
5658 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5659
5660 // no receiver because it is a static method
5661 Node* elementType = argument(0);
5662 Node* obj = argument(1);
5663 Node* offset = argument(2); // long
5664 Node* fromIndex = argument(4);
5665 Node* toIndex = argument(5);
5666 Node* indexPivot1 = argument(6);
5667 Node* indexPivot2 = argument(7);
5668 // PartitionOperation: argument(8) is ignored
5669
5670 Node* pivotIndices = nullptr;
5671 BasicType bt = T_ILLEGAL;
5672
5673 if (!check_array_sort_arguments(elementType, obj, bt)) {
5674 return false;
5675 }
5676 null_check(obj);
5677 // If obj is dead, only null-path is taken.
5678 if (stopped()) {
5679 return true;
5680 }
5681 // Set the original stack and the reexecute bit for the interpreter to reexecute
5682 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5683 { PreserveReexecuteState preexecs(this);
5684 jvms()->set_should_reexecute(true);
5685
5686 Node* obj_adr = make_unsafe_address(obj, offset);
5687
5688 // create the pivotIndices array of type int and size = 2
5689 Node* size = intcon(2);
5690 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5691 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5692 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5693 guarantee(alloc != nullptr, "created above");
5694 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5695
5696 // pass the basic type enum to the stub
5697 Node* elemType = intcon(bt);
5698
5699 // Call the stub
5700 const char *stubName = "array_partition_stub";
5701 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5702 stubAddr, stubName, TypePtr::BOTTOM,
5703 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5704 indexPivot1, indexPivot2);
5705
5706 } // original reexecute is set back here
5707
5708 if (!stopped()) {
5709 set_result(pivotIndices);
5710 }
5711
5712 return true;
5713 }
5714
5715
5716 //------------------------------inline_array_sort-----------------------
5717 bool LibraryCallKit::inline_array_sort() {
5718 address stubAddr = StubRoutines::select_arraysort_function();
5719 if (stubAddr == nullptr) {
5720 return false; // Intrinsic's stub is not implemented on this platform
5721 }
5722 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5723
5724 // no receiver because it is a static method
5725 Node* elementType = argument(0);
5726 Node* obj = argument(1);
5727 Node* offset = argument(2); // long
5728 Node* fromIndex = argument(4);
5729 Node* toIndex = argument(5);
5730 // SortOperation: argument(6) is ignored
5731
5732 BasicType bt = T_ILLEGAL;
5733
5734 if (!check_array_sort_arguments(elementType, obj, bt)) {
5735 return false;
5736 }
5737 null_check(obj);
5738 // If obj is dead, only null-path is taken.
5739 if (stopped()) {
5740 return true;
5741 }
5742 Node* obj_adr = make_unsafe_address(obj, offset);
5743
5744 // pass the basic type enum to the stub
5745 Node* elemType = intcon(bt);
5746
5747 // Call the stub.
5748 const char *stubName = "arraysort_stub";
5749 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5750 stubAddr, stubName, TypePtr::BOTTOM,
5751 obj_adr, elemType, fromIndex, toIndex);
5752
5753 return true;
5754 }
5755
5756
5757 //------------------------------inline_arraycopy-----------------------
5758 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5759 // Object dest, int destPos,
5760 // int length);
5761 bool LibraryCallKit::inline_arraycopy() {
5762 // Get the arguments.
5763 Node* src = argument(0); // type: oop
5764 Node* src_offset = argument(1); // type: int
5765 Node* dest = argument(2); // type: oop
5766 Node* dest_offset = argument(3); // type: int
5767 Node* length = argument(4); // type: int
5768
5769 uint new_idx = C->unique();
5770
5771 // Check for allocation before we add nodes that would confuse
5772 // tightly_coupled_allocation()
5773 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5774
5775 int saved_reexecute_sp = -1;
5776 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5777 // See arraycopy_restore_alloc_state() comment
5778 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5779 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5780 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5781 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5782
5783 // The following tests must be performed
5784 // (1) src and dest are arrays.
5785 // (2) src and dest arrays must have elements of the same BasicType
5786 // (3) src and dest must not be null.
5787 // (4) src_offset must not be negative.
5788 // (5) dest_offset must not be negative.
5789 // (6) length must not be negative.
5790 // (7) src_offset + length must not exceed length of src.
5791 // (8) dest_offset + length must not exceed length of dest.
5792 // (9) each element of an oop array must be assignable
5793
5794 // (3) src and dest must not be null.
5795 // always do this here because we need the JVM state for uncommon traps
5796 Node* null_ctl = top();
5797 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5798 assert(null_ctl->is_top(), "no null control here");
5799 dest = null_check(dest, T_ARRAY);
5800
5801 if (!can_emit_guards) {
5802 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5803 // guards but the arraycopy node could still take advantage of a
5804 // tightly allocated allocation. tightly_coupled_allocation() is
5805 // called again to make sure it takes the null check above into
5806 // account: the null check is mandatory and if it caused an
5807 // uncommon trap to be emitted then the allocation can't be
5808 // considered tightly coupled in this context.
5809 alloc = tightly_coupled_allocation(dest);
5810 }
5811
5812 bool validated = false;
5813
5814 const Type* src_type = _gvn.type(src);
5815 const Type* dest_type = _gvn.type(dest);
5816 const TypeAryPtr* top_src = src_type->isa_aryptr();
5817 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5818
5819 // Do we have the type of src?
5820 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5821 // Do we have the type of dest?
5822 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5823 // Is the type for src from speculation?
5824 bool src_spec = false;
5825 // Is the type for dest from speculation?
5826 bool dest_spec = false;
5827
5828 if ((!has_src || !has_dest) && can_emit_guards) {
5829 // We don't have sufficient type information, let's see if
5830 // speculative types can help. We need to have types for both src
5831 // and dest so that it pays off.
5832
5833 // Do we already have or could we have type information for src
5834 bool could_have_src = has_src;
5835 // Do we already have or could we have type information for dest
5836 bool could_have_dest = has_dest;
5837
5838 ciKlass* src_k = nullptr;
5839 if (!has_src) {
5840 src_k = src_type->speculative_type_not_null();
5841 if (src_k != nullptr && src_k->is_array_klass()) {
5842 could_have_src = true;
5843 }
5844 }
5845
5846 ciKlass* dest_k = nullptr;
5847 if (!has_dest) {
5848 dest_k = dest_type->speculative_type_not_null();
5849 if (dest_k != nullptr && dest_k->is_array_klass()) {
5850 could_have_dest = true;
5851 }
5852 }
5853
5854 if (could_have_src && could_have_dest) {
5855 // This is going to pay off so emit the required guards
5856 if (!has_src) {
5857 src = maybe_cast_profiled_obj(src, src_k, true);
5858 src_type = _gvn.type(src);
5859 top_src = src_type->isa_aryptr();
5860 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5861 src_spec = true;
5862 }
5863 if (!has_dest) {
5864 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5865 dest_type = _gvn.type(dest);
5866 top_dest = dest_type->isa_aryptr();
5867 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5868 dest_spec = true;
5869 }
5870 }
5871 }
5872
5873 if (has_src && has_dest && can_emit_guards) {
5874 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5875 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5876 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5877 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5878
5879 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5880 // If both arrays are object arrays then having the exact types
5881 // for both will remove the need for a subtype check at runtime
5882 // before the call and may make it possible to pick a faster copy
5883 // routine (without a subtype check on every element)
5884 // Do we have the exact type of src?
5885 bool could_have_src = src_spec;
5886 // Do we have the exact type of dest?
5887 bool could_have_dest = dest_spec;
5888 ciKlass* src_k = nullptr;
5889 ciKlass* dest_k = nullptr;
5890 if (!src_spec) {
5891 src_k = src_type->speculative_type_not_null();
5892 if (src_k != nullptr && src_k->is_array_klass()) {
5893 could_have_src = true;
5894 }
5895 }
5896 if (!dest_spec) {
5897 dest_k = dest_type->speculative_type_not_null();
5898 if (dest_k != nullptr && dest_k->is_array_klass()) {
5899 could_have_dest = true;
5900 }
5901 }
5902 if (could_have_src && could_have_dest) {
5903 // If we can have both exact types, emit the missing guards
5904 if (could_have_src && !src_spec) {
5905 src = maybe_cast_profiled_obj(src, src_k, true);
5906 }
5907 if (could_have_dest && !dest_spec) {
5908 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5909 }
5910 }
5911 }
5912 }
5913
5914 ciMethod* trap_method = method();
5915 int trap_bci = bci();
5916 if (saved_jvms_before_guards != nullptr) {
5917 trap_method = alloc->jvms()->method();
5918 trap_bci = alloc->jvms()->bci();
5919 }
5920
5921 bool negative_length_guard_generated = false;
5922
5923 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5924 can_emit_guards &&
5925 !src->is_top() && !dest->is_top()) {
5926 // validate arguments: enables transformation the ArrayCopyNode
5927 validated = true;
5928
5929 RegionNode* slow_region = new RegionNode(1);
5930 record_for_igvn(slow_region);
5931
5932 // (1) src and dest are arrays.
5933 generate_non_array_guard(load_object_klass(src), slow_region);
5934 generate_non_array_guard(load_object_klass(dest), slow_region);
5935
5936 // (2) src and dest arrays must have elements of the same BasicType
5937 // done at macro expansion or at Ideal transformation time
5938
5939 // (4) src_offset must not be negative.
5940 generate_negative_guard(src_offset, slow_region);
5941
5942 // (5) dest_offset must not be negative.
5943 generate_negative_guard(dest_offset, slow_region);
5944
5945 // (7) src_offset + length must not exceed length of src.
5946 generate_limit_guard(src_offset, length,
5947 load_array_length(src),
5948 slow_region);
5949
5950 // (8) dest_offset + length must not exceed length of dest.
5951 generate_limit_guard(dest_offset, length,
5952 load_array_length(dest),
5953 slow_region);
5954
5955 // (6) length must not be negative.
5956 // This is also checked in generate_arraycopy() during macro expansion, but
5957 // we also have to check it here for the case where the ArrayCopyNode will
5958 // be eliminated by Escape Analysis.
5959 if (EliminateAllocations) {
5960 generate_negative_guard(length, slow_region);
5961 negative_length_guard_generated = true;
5962 }
5963
5964 // (9) each element of an oop array must be assignable
5965 Node* dest_klass = load_object_klass(dest);
5966 if (src != dest) {
5967 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
5968
5969 if (not_subtype_ctrl != top()) {
5970 PreserveJVMState pjvms(this);
5971 set_control(not_subtype_ctrl);
5972 uncommon_trap(Deoptimization::Reason_intrinsic,
5973 Deoptimization::Action_make_not_entrant);
5974 assert(stopped(), "Should be stopped");
5975 }
5976 }
5977 {
5978 PreserveJVMState pjvms(this);
5979 set_control(_gvn.transform(slow_region));
5980 uncommon_trap(Deoptimization::Reason_intrinsic,
5981 Deoptimization::Action_make_not_entrant);
5982 assert(stopped(), "Should be stopped");
5983 }
5984
5985 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
5986 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
5987 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
5988 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
5989 }
5990
5991 if (stopped()) {
5992 return true;
5993 }
5994
5995 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
5996 // Create LoadRange and LoadKlass nodes for use during macro expansion here
5997 // so the compiler has a chance to eliminate them: during macro expansion,
5998 // we have to set their control (CastPP nodes are eliminated).
5999 load_object_klass(src), load_object_klass(dest),
6000 load_array_length(src), load_array_length(dest));
6001
6002 ac->set_arraycopy(validated);
6003
6004 Node* n = _gvn.transform(ac);
6005 if (n == ac) {
6006 ac->connect_outputs(this);
6007 } else {
6008 assert(validated, "shouldn't transform if all arguments not validated");
6009 set_all_memory(n);
6010 }
6011 clear_upper_avx();
6012
6013
6014 return true;
6015 }
6016
6017
6018 // Helper function which determines if an arraycopy immediately follows
6019 // an allocation, with no intervening tests or other escapes for the object.
6020 AllocateArrayNode*
6021 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6022 if (stopped()) return nullptr; // no fast path
6023 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6024
6025 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6026 if (alloc == nullptr) return nullptr;
6027
6028 Node* rawmem = memory(Compile::AliasIdxRaw);
6029 // Is the allocation's memory state untouched?
6030 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6031 // Bail out if there have been raw-memory effects since the allocation.
6032 // (Example: There might have been a call or safepoint.)
6033 return nullptr;
6034 }
6035 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6036 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6037 return nullptr;
6038 }
6039
6040 // There must be no unexpected observers of this allocation.
6041 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6042 Node* obs = ptr->fast_out(i);
6043 if (obs != this->map()) {
6044 return nullptr;
6045 }
6046 }
6047
6048 // This arraycopy must unconditionally follow the allocation of the ptr.
6049 Node* alloc_ctl = ptr->in(0);
6050 Node* ctl = control();
6051 while (ctl != alloc_ctl) {
6052 // There may be guards which feed into the slow_region.
6053 // Any other control flow means that we might not get a chance
6054 // to finish initializing the allocated object.
6055 // Various low-level checks bottom out in uncommon traps. These
6056 // are considered safe since we've already checked above that
6057 // there is no unexpected observer of this allocation.
6058 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6059 assert(ctl->in(0)->is_If(), "must be If");
6060 ctl = ctl->in(0)->in(0);
6061 } else {
6062 return nullptr;
6063 }
6064 }
6065
6066 // If we get this far, we have an allocation which immediately
6067 // precedes the arraycopy, and we can take over zeroing the new object.
6068 // The arraycopy will finish the initialization, and provide
6069 // a new control state to which we will anchor the destination pointer.
6070
6071 return alloc;
6072 }
6073
6074 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6075 if (node->is_IfProj()) {
6076 Node* other_proj = node->as_IfProj()->other_if_proj();
6077 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6078 Node* obs = other_proj->fast_out(j);
6079 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6080 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6081 return obs->as_CallStaticJava();
6082 }
6083 }
6084 }
6085 return nullptr;
6086 }
6087
6088 //-------------inline_encodeISOArray-----------------------------------
6089 // encode char[] to byte[] in ISO_8859_1 or ASCII
6090 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6091 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6092 // no receiver since it is static method
6093 Node *src = argument(0);
6094 Node *src_offset = argument(1);
6095 Node *dst = argument(2);
6096 Node *dst_offset = argument(3);
6097 Node *length = argument(4);
6098
6099 src = must_be_not_null(src, true);
6100 dst = must_be_not_null(dst, true);
6101
6102 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6103 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6104 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6105 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6106 // failed array check
6107 return false;
6108 }
6109
6110 // Figure out the size and type of the elements we will be copying.
6111 BasicType src_elem = src_type->elem()->array_element_basic_type();
6112 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6113 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6114 return false;
6115 }
6116
6117 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6118 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6119 // 'src_start' points to src array + scaled offset
6120 // 'dst_start' points to dst array + scaled offset
6121
6122 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6123 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6124 enc = _gvn.transform(enc);
6125 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6126 set_memory(res_mem, mtype);
6127 set_result(enc);
6128 clear_upper_avx();
6129
6130 return true;
6131 }
6132
6133 //-------------inline_multiplyToLen-----------------------------------
6134 bool LibraryCallKit::inline_multiplyToLen() {
6135 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6136
6137 address stubAddr = StubRoutines::multiplyToLen();
6138 if (stubAddr == nullptr) {
6139 return false; // Intrinsic's stub is not implemented on this platform
6140 }
6141 const char* stubName = "multiplyToLen";
6142
6143 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6144
6145 // no receiver because it is a static method
6146 Node* x = argument(0);
6147 Node* xlen = argument(1);
6148 Node* y = argument(2);
6149 Node* ylen = argument(3);
6150 Node* z = argument(4);
6151
6152 x = must_be_not_null(x, true);
6153 y = must_be_not_null(y, true);
6154
6155 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6156 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6157 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6158 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6159 // failed array check
6160 return false;
6161 }
6162
6163 BasicType x_elem = x_type->elem()->array_element_basic_type();
6164 BasicType y_elem = y_type->elem()->array_element_basic_type();
6165 if (x_elem != T_INT || y_elem != T_INT) {
6166 return false;
6167 }
6168
6169 Node* x_start = array_element_address(x, intcon(0), x_elem);
6170 Node* y_start = array_element_address(y, intcon(0), y_elem);
6171 // 'x_start' points to x array + scaled xlen
6172 // 'y_start' points to y array + scaled ylen
6173
6174 Node* z_start = array_element_address(z, intcon(0), T_INT);
6175
6176 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6177 OptoRuntime::multiplyToLen_Type(),
6178 stubAddr, stubName, TypePtr::BOTTOM,
6179 x_start, xlen, y_start, ylen, z_start);
6180
6181 C->set_has_split_ifs(true); // Has chance for split-if optimization
6182 set_result(z);
6183 return true;
6184 }
6185
6186 //-------------inline_squareToLen------------------------------------
6187 bool LibraryCallKit::inline_squareToLen() {
6188 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6189
6190 address stubAddr = StubRoutines::squareToLen();
6191 if (stubAddr == nullptr) {
6192 return false; // Intrinsic's stub is not implemented on this platform
6193 }
6194 const char* stubName = "squareToLen";
6195
6196 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6197
6198 Node* x = argument(0);
6199 Node* len = argument(1);
6200 Node* z = argument(2);
6201 Node* zlen = argument(3);
6202
6203 x = must_be_not_null(x, true);
6204 z = must_be_not_null(z, true);
6205
6206 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6207 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6208 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6209 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6210 // failed array check
6211 return false;
6212 }
6213
6214 BasicType x_elem = x_type->elem()->array_element_basic_type();
6215 BasicType z_elem = z_type->elem()->array_element_basic_type();
6216 if (x_elem != T_INT || z_elem != T_INT) {
6217 return false;
6218 }
6219
6220
6221 Node* x_start = array_element_address(x, intcon(0), x_elem);
6222 Node* z_start = array_element_address(z, intcon(0), z_elem);
6223
6224 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6225 OptoRuntime::squareToLen_Type(),
6226 stubAddr, stubName, TypePtr::BOTTOM,
6227 x_start, len, z_start, zlen);
6228
6229 set_result(z);
6230 return true;
6231 }
6232
6233 //-------------inline_mulAdd------------------------------------------
6234 bool LibraryCallKit::inline_mulAdd() {
6235 assert(UseMulAddIntrinsic, "not implemented on this platform");
6236
6237 address stubAddr = StubRoutines::mulAdd();
6238 if (stubAddr == nullptr) {
6239 return false; // Intrinsic's stub is not implemented on this platform
6240 }
6241 const char* stubName = "mulAdd";
6242
6243 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6244
6245 Node* out = argument(0);
6246 Node* in = argument(1);
6247 Node* offset = argument(2);
6248 Node* len = argument(3);
6249 Node* k = argument(4);
6250
6251 in = must_be_not_null(in, true);
6252 out = must_be_not_null(out, true);
6253
6254 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6255 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6256 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6257 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6258 // failed array check
6259 return false;
6260 }
6261
6262 BasicType out_elem = out_type->elem()->array_element_basic_type();
6263 BasicType in_elem = in_type->elem()->array_element_basic_type();
6264 if (out_elem != T_INT || in_elem != T_INT) {
6265 return false;
6266 }
6267
6268 Node* outlen = load_array_length(out);
6269 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6270 Node* out_start = array_element_address(out, intcon(0), out_elem);
6271 Node* in_start = array_element_address(in, intcon(0), in_elem);
6272
6273 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6274 OptoRuntime::mulAdd_Type(),
6275 stubAddr, stubName, TypePtr::BOTTOM,
6276 out_start,in_start, new_offset, len, k);
6277 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6278 set_result(result);
6279 return true;
6280 }
6281
6282 //-------------inline_montgomeryMultiply-----------------------------------
6283 bool LibraryCallKit::inline_montgomeryMultiply() {
6284 address stubAddr = StubRoutines::montgomeryMultiply();
6285 if (stubAddr == nullptr) {
6286 return false; // Intrinsic's stub is not implemented on this platform
6287 }
6288
6289 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6290 const char* stubName = "montgomery_multiply";
6291
6292 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6293
6294 Node* a = argument(0);
6295 Node* b = argument(1);
6296 Node* n = argument(2);
6297 Node* len = argument(3);
6298 Node* inv = argument(4);
6299 Node* m = argument(6);
6300
6301 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6302 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6303 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6304 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6305 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6306 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6307 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6308 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6309 // failed array check
6310 return false;
6311 }
6312
6313 BasicType a_elem = a_type->elem()->array_element_basic_type();
6314 BasicType b_elem = b_type->elem()->array_element_basic_type();
6315 BasicType n_elem = n_type->elem()->array_element_basic_type();
6316 BasicType m_elem = m_type->elem()->array_element_basic_type();
6317 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6318 return false;
6319 }
6320
6321 // Make the call
6322 {
6323 Node* a_start = array_element_address(a, intcon(0), a_elem);
6324 Node* b_start = array_element_address(b, intcon(0), b_elem);
6325 Node* n_start = array_element_address(n, intcon(0), n_elem);
6326 Node* m_start = array_element_address(m, intcon(0), m_elem);
6327
6328 Node* call = make_runtime_call(RC_LEAF,
6329 OptoRuntime::montgomeryMultiply_Type(),
6330 stubAddr, stubName, TypePtr::BOTTOM,
6331 a_start, b_start, n_start, len, inv, top(),
6332 m_start);
6333 set_result(m);
6334 }
6335
6336 return true;
6337 }
6338
6339 bool LibraryCallKit::inline_montgomerySquare() {
6340 address stubAddr = StubRoutines::montgomerySquare();
6341 if (stubAddr == nullptr) {
6342 return false; // Intrinsic's stub is not implemented on this platform
6343 }
6344
6345 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6346 const char* stubName = "montgomery_square";
6347
6348 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6349
6350 Node* a = argument(0);
6351 Node* n = argument(1);
6352 Node* len = argument(2);
6353 Node* inv = argument(3);
6354 Node* m = argument(5);
6355
6356 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6357 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6358 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6359 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6360 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6361 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6362 // failed array check
6363 return false;
6364 }
6365
6366 BasicType a_elem = a_type->elem()->array_element_basic_type();
6367 BasicType n_elem = n_type->elem()->array_element_basic_type();
6368 BasicType m_elem = m_type->elem()->array_element_basic_type();
6369 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6370 return false;
6371 }
6372
6373 // Make the call
6374 {
6375 Node* a_start = array_element_address(a, intcon(0), a_elem);
6376 Node* n_start = array_element_address(n, intcon(0), n_elem);
6377 Node* m_start = array_element_address(m, intcon(0), m_elem);
6378
6379 Node* call = make_runtime_call(RC_LEAF,
6380 OptoRuntime::montgomerySquare_Type(),
6381 stubAddr, stubName, TypePtr::BOTTOM,
6382 a_start, n_start, len, inv, top(),
6383 m_start);
6384 set_result(m);
6385 }
6386
6387 return true;
6388 }
6389
6390 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6391 address stubAddr = nullptr;
6392 const char* stubName = nullptr;
6393
6394 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6395 if (stubAddr == nullptr) {
6396 return false; // Intrinsic's stub is not implemented on this platform
6397 }
6398
6399 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6400
6401 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6402
6403 Node* newArr = argument(0);
6404 Node* oldArr = argument(1);
6405 Node* newIdx = argument(2);
6406 Node* shiftCount = argument(3);
6407 Node* numIter = argument(4);
6408
6409 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6410 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6411 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6412 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6413 return false;
6414 }
6415
6416 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6417 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6418 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6419 return false;
6420 }
6421
6422 // Make the call
6423 {
6424 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6425 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6426
6427 Node* call = make_runtime_call(RC_LEAF,
6428 OptoRuntime::bigIntegerShift_Type(),
6429 stubAddr,
6430 stubName,
6431 TypePtr::BOTTOM,
6432 newArr_start,
6433 oldArr_start,
6434 newIdx,
6435 shiftCount,
6436 numIter);
6437 }
6438
6439 return true;
6440 }
6441
6442 //-------------inline_vectorizedMismatch------------------------------
6443 bool LibraryCallKit::inline_vectorizedMismatch() {
6444 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6445
6446 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6447 Node* obja = argument(0); // Object
6448 Node* aoffset = argument(1); // long
6449 Node* objb = argument(3); // Object
6450 Node* boffset = argument(4); // long
6451 Node* length = argument(6); // int
6452 Node* scale = argument(7); // int
6453
6454 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6455 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6456 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6457 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6458 scale == top()) {
6459 return false; // failed input validation
6460 }
6461
6462 Node* obja_adr = make_unsafe_address(obja, aoffset);
6463 Node* objb_adr = make_unsafe_address(objb, boffset);
6464
6465 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6466 //
6467 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6468 // if (length <= inline_limit) {
6469 // inline_path:
6470 // vmask = VectorMaskGen length
6471 // vload1 = LoadVectorMasked obja, vmask
6472 // vload2 = LoadVectorMasked objb, vmask
6473 // result1 = VectorCmpMasked vload1, vload2, vmask
6474 // } else {
6475 // call_stub_path:
6476 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6477 // }
6478 // exit_block:
6479 // return Phi(result1, result2);
6480 //
6481 enum { inline_path = 1, // input is small enough to process it all at once
6482 stub_path = 2, // input is too large; call into the VM
6483 PATH_LIMIT = 3
6484 };
6485
6486 Node* exit_block = new RegionNode(PATH_LIMIT);
6487 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6488 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6489
6490 Node* call_stub_path = control();
6491
6492 BasicType elem_bt = T_ILLEGAL;
6493
6494 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6495 if (scale_t->is_con()) {
6496 switch (scale_t->get_con()) {
6497 case 0: elem_bt = T_BYTE; break;
6498 case 1: elem_bt = T_SHORT; break;
6499 case 2: elem_bt = T_INT; break;
6500 case 3: elem_bt = T_LONG; break;
6501
6502 default: elem_bt = T_ILLEGAL; break; // not supported
6503 }
6504 }
6505
6506 int inline_limit = 0;
6507 bool do_partial_inline = false;
6508
6509 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6510 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6511 do_partial_inline = inline_limit >= 16;
6512 }
6513
6514 if (do_partial_inline) {
6515 assert(elem_bt != T_ILLEGAL, "sanity");
6516
6517 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6518 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6519 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6520
6521 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6522 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6523 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6524
6525 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6526
6527 if (!stopped()) {
6528 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6529
6530 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6531 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6532 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6533 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6534
6535 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6536 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6537 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6538 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6539
6540 exit_block->init_req(inline_path, control());
6541 memory_phi->init_req(inline_path, map()->memory());
6542 result_phi->init_req(inline_path, result);
6543
6544 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6545 clear_upper_avx();
6546 }
6547 }
6548 }
6549
6550 if (call_stub_path != nullptr) {
6551 set_control(call_stub_path);
6552
6553 Node* call = make_runtime_call(RC_LEAF,
6554 OptoRuntime::vectorizedMismatch_Type(),
6555 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6556 obja_adr, objb_adr, length, scale);
6557
6558 exit_block->init_req(stub_path, control());
6559 memory_phi->init_req(stub_path, map()->memory());
6560 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6561 }
6562
6563 exit_block = _gvn.transform(exit_block);
6564 memory_phi = _gvn.transform(memory_phi);
6565 result_phi = _gvn.transform(result_phi);
6566
6567 set_control(exit_block);
6568 set_all_memory(memory_phi);
6569 set_result(result_phi);
6570
6571 return true;
6572 }
6573
6574 //------------------------------inline_vectorizedHashcode----------------------------
6575 bool LibraryCallKit::inline_vectorizedHashCode() {
6576 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6577
6578 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6579 Node* array = argument(0);
6580 Node* offset = argument(1);
6581 Node* length = argument(2);
6582 Node* initialValue = argument(3);
6583 Node* basic_type = argument(4);
6584
6585 if (basic_type == top()) {
6586 return false; // failed input validation
6587 }
6588
6589 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6590 if (!basic_type_t->is_con()) {
6591 return false; // Only intrinsify if mode argument is constant
6592 }
6593
6594 array = must_be_not_null(array, true);
6595
6596 BasicType bt = (BasicType)basic_type_t->get_con();
6597
6598 // Resolve address of first element
6599 Node* array_start = array_element_address(array, offset, bt);
6600
6601 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6602 array_start, length, initialValue, basic_type)));
6603 clear_upper_avx();
6604
6605 return true;
6606 }
6607
6608 /**
6609 * Calculate CRC32 for byte.
6610 * int java.util.zip.CRC32.update(int crc, int b)
6611 */
6612 bool LibraryCallKit::inline_updateCRC32() {
6613 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6614 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6615 // no receiver since it is static method
6616 Node* crc = argument(0); // type: int
6617 Node* b = argument(1); // type: int
6618
6619 /*
6620 * int c = ~ crc;
6621 * b = timesXtoThe32[(b ^ c) & 0xFF];
6622 * b = b ^ (c >>> 8);
6623 * crc = ~b;
6624 */
6625
6626 Node* M1 = intcon(-1);
6627 crc = _gvn.transform(new XorINode(crc, M1));
6628 Node* result = _gvn.transform(new XorINode(crc, b));
6629 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6630
6631 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6632 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6633 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6634 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6635
6636 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6637 result = _gvn.transform(new XorINode(crc, result));
6638 result = _gvn.transform(new XorINode(result, M1));
6639 set_result(result);
6640 return true;
6641 }
6642
6643 /**
6644 * Calculate CRC32 for byte[] array.
6645 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6646 */
6647 bool LibraryCallKit::inline_updateBytesCRC32() {
6648 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6649 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6650 // no receiver since it is static method
6651 Node* crc = argument(0); // type: int
6652 Node* src = argument(1); // type: oop
6653 Node* offset = argument(2); // type: int
6654 Node* length = argument(3); // type: int
6655
6656 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6657 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6658 // failed array check
6659 return false;
6660 }
6661
6662 // Figure out the size and type of the elements we will be copying.
6663 BasicType src_elem = src_type->elem()->array_element_basic_type();
6664 if (src_elem != T_BYTE) {
6665 return false;
6666 }
6667
6668 // 'src_start' points to src array + scaled offset
6669 src = must_be_not_null(src, true);
6670 Node* src_start = array_element_address(src, offset, src_elem);
6671
6672 // We assume that range check is done by caller.
6673 // TODO: generate range check (offset+length < src.length) in debug VM.
6674
6675 // Call the stub.
6676 address stubAddr = StubRoutines::updateBytesCRC32();
6677 const char *stubName = "updateBytesCRC32";
6678
6679 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6680 stubAddr, stubName, TypePtr::BOTTOM,
6681 crc, src_start, length);
6682 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6683 set_result(result);
6684 return true;
6685 }
6686
6687 /**
6688 * Calculate CRC32 for ByteBuffer.
6689 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6690 */
6691 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6692 assert(UseCRC32Intrinsics, "need AVX and LCMUL instructions support");
6693 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6694 // no receiver since it is static method
6695 Node* crc = argument(0); // type: int
6696 Node* src = argument(1); // type: long
6697 Node* offset = argument(3); // type: int
6698 Node* length = argument(4); // type: int
6699
6700 src = ConvL2X(src); // adjust Java long to machine word
6701 Node* base = _gvn.transform(new CastX2PNode(src));
6702 offset = ConvI2X(offset);
6703
6704 // 'src_start' points to src array + scaled offset
6705 Node* src_start = basic_plus_adr(top(), base, offset);
6706
6707 // Call the stub.
6708 address stubAddr = StubRoutines::updateBytesCRC32();
6709 const char *stubName = "updateBytesCRC32";
6710
6711 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6712 stubAddr, stubName, TypePtr::BOTTOM,
6713 crc, src_start, length);
6714 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6715 set_result(result);
6716 return true;
6717 }
6718
6719 //------------------------------get_table_from_crc32c_class-----------------------
6720 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6721 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6722 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6723
6724 return table;
6725 }
6726
6727 //------------------------------inline_updateBytesCRC32C-----------------------
6728 //
6729 // Calculate CRC32C for byte[] array.
6730 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6731 //
6732 bool LibraryCallKit::inline_updateBytesCRC32C() {
6733 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6734 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6735 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6736 // no receiver since it is a static method
6737 Node* crc = argument(0); // type: int
6738 Node* src = argument(1); // type: oop
6739 Node* offset = argument(2); // type: int
6740 Node* end = argument(3); // type: int
6741
6742 Node* length = _gvn.transform(new SubINode(end, offset));
6743
6744 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6745 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6746 // failed array check
6747 return false;
6748 }
6749
6750 // Figure out the size and type of the elements we will be copying.
6751 BasicType src_elem = src_type->elem()->array_element_basic_type();
6752 if (src_elem != T_BYTE) {
6753 return false;
6754 }
6755
6756 // 'src_start' points to src array + scaled offset
6757 src = must_be_not_null(src, true);
6758 Node* src_start = array_element_address(src, offset, src_elem);
6759
6760 // static final int[] byteTable in class CRC32C
6761 Node* table = get_table_from_crc32c_class(callee()->holder());
6762 table = must_be_not_null(table, true);
6763 Node* table_start = array_element_address(table, intcon(0), T_INT);
6764
6765 // We assume that range check is done by caller.
6766 // TODO: generate range check (offset+length < src.length) in debug VM.
6767
6768 // Call the stub.
6769 address stubAddr = StubRoutines::updateBytesCRC32C();
6770 const char *stubName = "updateBytesCRC32C";
6771
6772 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6773 stubAddr, stubName, TypePtr::BOTTOM,
6774 crc, src_start, length, table_start);
6775 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6776 set_result(result);
6777 return true;
6778 }
6779
6780 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6781 //
6782 // Calculate CRC32C for DirectByteBuffer.
6783 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6784 //
6785 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6786 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6787 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6788 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6789 // no receiver since it is a static method
6790 Node* crc = argument(0); // type: int
6791 Node* src = argument(1); // type: long
6792 Node* offset = argument(3); // type: int
6793 Node* end = argument(4); // type: int
6794
6795 Node* length = _gvn.transform(new SubINode(end, offset));
6796
6797 src = ConvL2X(src); // adjust Java long to machine word
6798 Node* base = _gvn.transform(new CastX2PNode(src));
6799 offset = ConvI2X(offset);
6800
6801 // 'src_start' points to src array + scaled offset
6802 Node* src_start = basic_plus_adr(top(), base, offset);
6803
6804 // static final int[] byteTable in class CRC32C
6805 Node* table = get_table_from_crc32c_class(callee()->holder());
6806 table = must_be_not_null(table, true);
6807 Node* table_start = array_element_address(table, intcon(0), T_INT);
6808
6809 // Call the stub.
6810 address stubAddr = StubRoutines::updateBytesCRC32C();
6811 const char *stubName = "updateBytesCRC32C";
6812
6813 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6814 stubAddr, stubName, TypePtr::BOTTOM,
6815 crc, src_start, length, table_start);
6816 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6817 set_result(result);
6818 return true;
6819 }
6820
6821 //------------------------------inline_updateBytesAdler32----------------------
6822 //
6823 // Calculate Adler32 checksum for byte[] array.
6824 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6825 //
6826 bool LibraryCallKit::inline_updateBytesAdler32() {
6827 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6828 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6829 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6830 // no receiver since it is static method
6831 Node* crc = argument(0); // type: int
6832 Node* src = argument(1); // type: oop
6833 Node* offset = argument(2); // type: int
6834 Node* length = argument(3); // type: int
6835
6836 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6837 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6838 // failed array check
6839 return false;
6840 }
6841
6842 // Figure out the size and type of the elements we will be copying.
6843 BasicType src_elem = src_type->elem()->array_element_basic_type();
6844 if (src_elem != T_BYTE) {
6845 return false;
6846 }
6847
6848 // 'src_start' points to src array + scaled offset
6849 Node* src_start = array_element_address(src, offset, src_elem);
6850
6851 // We assume that range check is done by caller.
6852 // TODO: generate range check (offset+length < src.length) in debug VM.
6853
6854 // Call the stub.
6855 address stubAddr = StubRoutines::updateBytesAdler32();
6856 const char *stubName = "updateBytesAdler32";
6857
6858 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6859 stubAddr, stubName, TypePtr::BOTTOM,
6860 crc, src_start, length);
6861 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6862 set_result(result);
6863 return true;
6864 }
6865
6866 //------------------------------inline_updateByteBufferAdler32---------------
6867 //
6868 // Calculate Adler32 checksum for DirectByteBuffer.
6869 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6870 //
6871 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6872 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6873 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6874 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6875 // no receiver since it is static method
6876 Node* crc = argument(0); // type: int
6877 Node* src = argument(1); // type: long
6878 Node* offset = argument(3); // type: int
6879 Node* length = argument(4); // type: int
6880
6881 src = ConvL2X(src); // adjust Java long to machine word
6882 Node* base = _gvn.transform(new CastX2PNode(src));
6883 offset = ConvI2X(offset);
6884
6885 // 'src_start' points to src array + scaled offset
6886 Node* src_start = basic_plus_adr(top(), base, offset);
6887
6888 // Call the stub.
6889 address stubAddr = StubRoutines::updateBytesAdler32();
6890 const char *stubName = "updateBytesAdler32";
6891
6892 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6893 stubAddr, stubName, TypePtr::BOTTOM,
6894 crc, src_start, length);
6895
6896 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6897 set_result(result);
6898 return true;
6899 }
6900
6901 //----------------------------inline_reference_get----------------------------
6902 // public T java.lang.ref.Reference.get();
6903 bool LibraryCallKit::inline_reference_get() {
6904 const int referent_offset = java_lang_ref_Reference::referent_offset();
6905
6906 // Get the argument:
6907 Node* reference_obj = null_check_receiver();
6908 if (stopped()) return true;
6909
6910 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6911 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6912 decorators, /*is_static*/ false, nullptr);
6913 if (result == nullptr) return false;
6914
6915 // Add memory barrier to prevent commoning reads from this field
6916 // across safepoint since GC can change its value.
6917 insert_mem_bar(Op_MemBarCPUOrder);
6918
6919 set_result(result);
6920 return true;
6921 }
6922
6923 //----------------------------inline_reference_refersTo0----------------------------
6924 // bool java.lang.ref.Reference.refersTo0();
6925 // bool java.lang.ref.PhantomReference.refersTo0();
6926 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6927 // Get arguments:
6928 Node* reference_obj = null_check_receiver();
6929 Node* other_obj = argument(1);
6930 if (stopped()) return true;
6931
6932 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6933 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
6934 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6935 decorators, /*is_static*/ false, nullptr);
6936 if (referent == nullptr) return false;
6937
6938 // Add memory barrier to prevent commoning reads from this field
6939 // across safepoint since GC can change its value.
6940 insert_mem_bar(Op_MemBarCPUOrder);
6941
6942 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
6943 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
6944 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
6945
6946 RegionNode* region = new RegionNode(3);
6947 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
6948
6949 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
6950 region->init_req(1, if_true);
6951 phi->init_req(1, intcon(1));
6952
6953 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
6954 region->init_req(2, if_false);
6955 phi->init_req(2, intcon(0));
6956
6957 set_control(_gvn.transform(region));
6958 record_for_igvn(region);
6959 set_result(_gvn.transform(phi));
6960 return true;
6961 }
6962
6963
6964 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
6965 DecoratorSet decorators, bool is_static,
6966 ciInstanceKlass* fromKls) {
6967 if (fromKls == nullptr) {
6968 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
6969 assert(tinst != nullptr, "obj is null");
6970 assert(tinst->is_loaded(), "obj is not loaded");
6971 fromKls = tinst->instance_klass();
6972 } else {
6973 assert(is_static, "only for static field access");
6974 }
6975 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
6976 ciSymbol::make(fieldTypeString),
6977 is_static);
6978
6979 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
6980 if (field == nullptr) return (Node *) nullptr;
6981
6982 if (is_static) {
6983 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
6984 fromObj = makecon(tip);
6985 }
6986
6987 // Next code copied from Parse::do_get_xxx():
6988
6989 // Compute address and memory type.
6990 int offset = field->offset_in_bytes();
6991 bool is_vol = field->is_volatile();
6992 ciType* field_klass = field->type();
6993 assert(field_klass->is_loaded(), "should be loaded");
6994 const TypePtr* adr_type = C->alias_type(field)->adr_type();
6995 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
6996 BasicType bt = field->layout_type();
6997
6998 // Build the resultant type of the load
6999 const Type *type;
7000 if (bt == T_OBJECT) {
7001 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7002 } else {
7003 type = Type::get_const_basic_type(bt);
7004 }
7005
7006 if (is_vol) {
7007 decorators |= MO_SEQ_CST;
7008 }
7009
7010 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7011 }
7012
7013 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7014 bool is_exact /* true */, bool is_static /* false */,
7015 ciInstanceKlass * fromKls /* nullptr */) {
7016 if (fromKls == nullptr) {
7017 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7018 assert(tinst != nullptr, "obj is null");
7019 assert(tinst->is_loaded(), "obj is not loaded");
7020 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7021 fromKls = tinst->instance_klass();
7022 }
7023 else {
7024 assert(is_static, "only for static field access");
7025 }
7026 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7027 ciSymbol::make(fieldTypeString),
7028 is_static);
7029
7030 assert(field != nullptr, "undefined field");
7031 assert(!field->is_volatile(), "not defined for volatile fields");
7032
7033 if (is_static) {
7034 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7035 fromObj = makecon(tip);
7036 }
7037
7038 // Next code copied from Parse::do_get_xxx():
7039
7040 // Compute address and memory type.
7041 int offset = field->offset_in_bytes();
7042 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7043
7044 return adr;
7045 }
7046
7047 //------------------------------inline_aescrypt_Block-----------------------
7048 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7049 address stubAddr = nullptr;
7050 const char *stubName;
7051 assert(UseAES, "need AES instruction support");
7052
7053 switch(id) {
7054 case vmIntrinsics::_aescrypt_encryptBlock:
7055 stubAddr = StubRoutines::aescrypt_encryptBlock();
7056 stubName = "aescrypt_encryptBlock";
7057 break;
7058 case vmIntrinsics::_aescrypt_decryptBlock:
7059 stubAddr = StubRoutines::aescrypt_decryptBlock();
7060 stubName = "aescrypt_decryptBlock";
7061 break;
7062 default:
7063 break;
7064 }
7065 if (stubAddr == nullptr) return false;
7066
7067 Node* aescrypt_object = argument(0);
7068 Node* src = argument(1);
7069 Node* src_offset = argument(2);
7070 Node* dest = argument(3);
7071 Node* dest_offset = argument(4);
7072
7073 src = must_be_not_null(src, true);
7074 dest = must_be_not_null(dest, true);
7075
7076 // (1) src and dest are arrays.
7077 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7078 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7079 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7080 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7081
7082 // for the quick and dirty code we will skip all the checks.
7083 // we are just trying to get the call to be generated.
7084 Node* src_start = src;
7085 Node* dest_start = dest;
7086 if (src_offset != nullptr || dest_offset != nullptr) {
7087 assert(src_offset != nullptr && dest_offset != nullptr, "");
7088 src_start = array_element_address(src, src_offset, T_BYTE);
7089 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7090 }
7091
7092 // now need to get the start of its expanded key array
7093 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7094 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7095 if (k_start == nullptr) return false;
7096
7097 // Call the stub.
7098 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7099 stubAddr, stubName, TypePtr::BOTTOM,
7100 src_start, dest_start, k_start);
7101
7102 return true;
7103 }
7104
7105 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7106 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7107 address stubAddr = nullptr;
7108 const char *stubName = nullptr;
7109
7110 assert(UseAES, "need AES instruction support");
7111
7112 switch(id) {
7113 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7114 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7115 stubName = "cipherBlockChaining_encryptAESCrypt";
7116 break;
7117 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7118 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7119 stubName = "cipherBlockChaining_decryptAESCrypt";
7120 break;
7121 default:
7122 break;
7123 }
7124 if (stubAddr == nullptr) return false;
7125
7126 Node* cipherBlockChaining_object = argument(0);
7127 Node* src = argument(1);
7128 Node* src_offset = argument(2);
7129 Node* len = argument(3);
7130 Node* dest = argument(4);
7131 Node* dest_offset = argument(5);
7132
7133 src = must_be_not_null(src, false);
7134 dest = must_be_not_null(dest, false);
7135
7136 // (1) src and dest are arrays.
7137 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7138 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7139 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7140 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7141
7142 // checks are the responsibility of the caller
7143 Node* src_start = src;
7144 Node* dest_start = dest;
7145 if (src_offset != nullptr || dest_offset != nullptr) {
7146 assert(src_offset != nullptr && dest_offset != nullptr, "");
7147 src_start = array_element_address(src, src_offset, T_BYTE);
7148 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7149 }
7150
7151 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7152 // (because of the predicated logic executed earlier).
7153 // so we cast it here safely.
7154 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7155
7156 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7157 if (embeddedCipherObj == nullptr) return false;
7158
7159 // cast it to what we know it will be at runtime
7160 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7161 assert(tinst != nullptr, "CBC obj is null");
7162 assert(tinst->is_loaded(), "CBC obj is not loaded");
7163 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7164 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7165
7166 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7167 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7168 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7169 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7170 aescrypt_object = _gvn.transform(aescrypt_object);
7171
7172 // we need to get the start of the aescrypt_object's expanded key array
7173 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7174 if (k_start == nullptr) return false;
7175
7176 // similarly, get the start address of the r vector
7177 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7178 if (objRvec == nullptr) return false;
7179 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7180
7181 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7182 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7183 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7184 stubAddr, stubName, TypePtr::BOTTOM,
7185 src_start, dest_start, k_start, r_start, len);
7186
7187 // return cipher length (int)
7188 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7189 set_result(retvalue);
7190 return true;
7191 }
7192
7193 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7194 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7195 address stubAddr = nullptr;
7196 const char *stubName = nullptr;
7197
7198 assert(UseAES, "need AES instruction support");
7199
7200 switch (id) {
7201 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7202 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7203 stubName = "electronicCodeBook_encryptAESCrypt";
7204 break;
7205 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7206 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7207 stubName = "electronicCodeBook_decryptAESCrypt";
7208 break;
7209 default:
7210 break;
7211 }
7212
7213 if (stubAddr == nullptr) return false;
7214
7215 Node* electronicCodeBook_object = argument(0);
7216 Node* src = argument(1);
7217 Node* src_offset = argument(2);
7218 Node* len = argument(3);
7219 Node* dest = argument(4);
7220 Node* dest_offset = argument(5);
7221
7222 // (1) src and dest are arrays.
7223 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7224 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7225 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7226 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7227
7228 // checks are the responsibility of the caller
7229 Node* src_start = src;
7230 Node* dest_start = dest;
7231 if (src_offset != nullptr || dest_offset != nullptr) {
7232 assert(src_offset != nullptr && dest_offset != nullptr, "");
7233 src_start = array_element_address(src, src_offset, T_BYTE);
7234 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7235 }
7236
7237 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7238 // (because of the predicated logic executed earlier).
7239 // so we cast it here safely.
7240 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7241
7242 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7243 if (embeddedCipherObj == nullptr) return false;
7244
7245 // cast it to what we know it will be at runtime
7246 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7247 assert(tinst != nullptr, "ECB obj is null");
7248 assert(tinst->is_loaded(), "ECB obj is not loaded");
7249 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7250 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7251
7252 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7253 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7254 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7255 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7256 aescrypt_object = _gvn.transform(aescrypt_object);
7257
7258 // we need to get the start of the aescrypt_object's expanded key array
7259 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7260 if (k_start == nullptr) return false;
7261
7262 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7263 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7264 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7265 stubAddr, stubName, TypePtr::BOTTOM,
7266 src_start, dest_start, k_start, len);
7267
7268 // return cipher length (int)
7269 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7270 set_result(retvalue);
7271 return true;
7272 }
7273
7274 //------------------------------inline_counterMode_AESCrypt-----------------------
7275 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7276 assert(UseAES, "need AES instruction support");
7277 if (!UseAESCTRIntrinsics) return false;
7278
7279 address stubAddr = nullptr;
7280 const char *stubName = nullptr;
7281 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7282 stubAddr = StubRoutines::counterMode_AESCrypt();
7283 stubName = "counterMode_AESCrypt";
7284 }
7285 if (stubAddr == nullptr) return false;
7286
7287 Node* counterMode_object = argument(0);
7288 Node* src = argument(1);
7289 Node* src_offset = argument(2);
7290 Node* len = argument(3);
7291 Node* dest = argument(4);
7292 Node* dest_offset = argument(5);
7293
7294 // (1) src and dest are arrays.
7295 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7296 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7297 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7298 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7299
7300 // checks are the responsibility of the caller
7301 Node* src_start = src;
7302 Node* dest_start = dest;
7303 if (src_offset != nullptr || dest_offset != nullptr) {
7304 assert(src_offset != nullptr && dest_offset != nullptr, "");
7305 src_start = array_element_address(src, src_offset, T_BYTE);
7306 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7307 }
7308
7309 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7310 // (because of the predicated logic executed earlier).
7311 // so we cast it here safely.
7312 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7313 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7314 if (embeddedCipherObj == nullptr) return false;
7315 // cast it to what we know it will be at runtime
7316 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7317 assert(tinst != nullptr, "CTR obj is null");
7318 assert(tinst->is_loaded(), "CTR obj is not loaded");
7319 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7320 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7321 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7322 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7323 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7324 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7325 aescrypt_object = _gvn.transform(aescrypt_object);
7326 // we need to get the start of the aescrypt_object's expanded key array
7327 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7328 if (k_start == nullptr) return false;
7329 // similarly, get the start address of the r vector
7330 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7331 if (obj_counter == nullptr) return false;
7332 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7333
7334 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7335 if (saved_encCounter == nullptr) return false;
7336 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7337 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7338
7339 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7340 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7341 OptoRuntime::counterMode_aescrypt_Type(),
7342 stubAddr, stubName, TypePtr::BOTTOM,
7343 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7344
7345 // return cipher length (int)
7346 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7347 set_result(retvalue);
7348 return true;
7349 }
7350
7351 //------------------------------get_key_start_from_aescrypt_object-----------------------
7352 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7353 #if defined(PPC64) || defined(S390)
7354 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7355 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7356 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7357 // The ppc64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7358 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7359 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7360 if (objSessionK == nullptr) {
7361 return (Node *) nullptr;
7362 }
7363 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7364 #else
7365 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7366 #endif // PPC64
7367 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7368 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7369
7370 // now have the array, need to get the start address of the K array
7371 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7372 return k_start;
7373 }
7374
7375 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7376 // Return node representing slow path of predicate check.
7377 // the pseudo code we want to emulate with this predicate is:
7378 // for encryption:
7379 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7380 // for decryption:
7381 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7382 // note cipher==plain is more conservative than the original java code but that's OK
7383 //
7384 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7385 // The receiver was checked for null already.
7386 Node* objCBC = argument(0);
7387
7388 Node* src = argument(1);
7389 Node* dest = argument(4);
7390
7391 // Load embeddedCipher field of CipherBlockChaining object.
7392 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7393
7394 // get AESCrypt klass for instanceOf check
7395 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7396 // will have same classloader as CipherBlockChaining object
7397 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7398 assert(tinst != nullptr, "CBCobj is null");
7399 assert(tinst->is_loaded(), "CBCobj is not loaded");
7400
7401 // we want to do an instanceof comparison against the AESCrypt class
7402 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7403 if (!klass_AESCrypt->is_loaded()) {
7404 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7405 Node* ctrl = control();
7406 set_control(top()); // no regular fast path
7407 return ctrl;
7408 }
7409
7410 src = must_be_not_null(src, true);
7411 dest = must_be_not_null(dest, true);
7412
7413 // Resolve oops to stable for CmpP below.
7414 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7415
7416 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7417 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7418 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7419
7420 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7421
7422 // for encryption, we are done
7423 if (!decrypting)
7424 return instof_false; // even if it is null
7425
7426 // for decryption, we need to add a further check to avoid
7427 // taking the intrinsic path when cipher and plain are the same
7428 // see the original java code for why.
7429 RegionNode* region = new RegionNode(3);
7430 region->init_req(1, instof_false);
7431
7432 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7433 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7434 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7435 region->init_req(2, src_dest_conjoint);
7436
7437 record_for_igvn(region);
7438 return _gvn.transform(region);
7439 }
7440
7441 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7442 // Return node representing slow path of predicate check.
7443 // the pseudo code we want to emulate with this predicate is:
7444 // for encryption:
7445 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7446 // for decryption:
7447 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7448 // note cipher==plain is more conservative than the original java code but that's OK
7449 //
7450 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7451 // The receiver was checked for null already.
7452 Node* objECB = argument(0);
7453
7454 // Load embeddedCipher field of ElectronicCodeBook object.
7455 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7456
7457 // get AESCrypt klass for instanceOf check
7458 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7459 // will have same classloader as ElectronicCodeBook object
7460 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7461 assert(tinst != nullptr, "ECBobj is null");
7462 assert(tinst->is_loaded(), "ECBobj is not loaded");
7463
7464 // we want to do an instanceof comparison against the AESCrypt class
7465 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7466 if (!klass_AESCrypt->is_loaded()) {
7467 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7468 Node* ctrl = control();
7469 set_control(top()); // no regular fast path
7470 return ctrl;
7471 }
7472 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7473
7474 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7475 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7476 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7477
7478 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7479
7480 // for encryption, we are done
7481 if (!decrypting)
7482 return instof_false; // even if it is null
7483
7484 // for decryption, we need to add a further check to avoid
7485 // taking the intrinsic path when cipher and plain are the same
7486 // see the original java code for why.
7487 RegionNode* region = new RegionNode(3);
7488 region->init_req(1, instof_false);
7489 Node* src = argument(1);
7490 Node* dest = argument(4);
7491 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7492 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7493 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7494 region->init_req(2, src_dest_conjoint);
7495
7496 record_for_igvn(region);
7497 return _gvn.transform(region);
7498 }
7499
7500 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7501 // Return node representing slow path of predicate check.
7502 // the pseudo code we want to emulate with this predicate is:
7503 // for encryption:
7504 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7505 // for decryption:
7506 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7507 // note cipher==plain is more conservative than the original java code but that's OK
7508 //
7509
7510 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7511 // The receiver was checked for null already.
7512 Node* objCTR = argument(0);
7513
7514 // Load embeddedCipher field of CipherBlockChaining object.
7515 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7516
7517 // get AESCrypt klass for instanceOf check
7518 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7519 // will have same classloader as CipherBlockChaining object
7520 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7521 assert(tinst != nullptr, "CTRobj is null");
7522 assert(tinst->is_loaded(), "CTRobj is not loaded");
7523
7524 // we want to do an instanceof comparison against the AESCrypt class
7525 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7526 if (!klass_AESCrypt->is_loaded()) {
7527 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7528 Node* ctrl = control();
7529 set_control(top()); // no regular fast path
7530 return ctrl;
7531 }
7532
7533 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7534 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7535 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7536 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7537 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7538
7539 return instof_false; // even if it is null
7540 }
7541
7542 //------------------------------inline_ghash_processBlocks
7543 bool LibraryCallKit::inline_ghash_processBlocks() {
7544 address stubAddr;
7545 const char *stubName;
7546 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7547
7548 stubAddr = StubRoutines::ghash_processBlocks();
7549 stubName = "ghash_processBlocks";
7550
7551 Node* data = argument(0);
7552 Node* offset = argument(1);
7553 Node* len = argument(2);
7554 Node* state = argument(3);
7555 Node* subkeyH = argument(4);
7556
7557 state = must_be_not_null(state, true);
7558 subkeyH = must_be_not_null(subkeyH, true);
7559 data = must_be_not_null(data, true);
7560
7561 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7562 assert(state_start, "state is null");
7563 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7564 assert(subkeyH_start, "subkeyH is null");
7565 Node* data_start = array_element_address(data, offset, T_BYTE);
7566 assert(data_start, "data is null");
7567
7568 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7569 OptoRuntime::ghash_processBlocks_Type(),
7570 stubAddr, stubName, TypePtr::BOTTOM,
7571 state_start, subkeyH_start, data_start, len);
7572 return true;
7573 }
7574
7575 //------------------------------inline_chacha20Block
7576 bool LibraryCallKit::inline_chacha20Block() {
7577 address stubAddr;
7578 const char *stubName;
7579 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7580
7581 stubAddr = StubRoutines::chacha20Block();
7582 stubName = "chacha20Block";
7583
7584 Node* state = argument(0);
7585 Node* result = argument(1);
7586
7587 state = must_be_not_null(state, true);
7588 result = must_be_not_null(result, true);
7589
7590 Node* state_start = array_element_address(state, intcon(0), T_INT);
7591 assert(state_start, "state is null");
7592 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7593 assert(result_start, "result is null");
7594
7595 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7596 OptoRuntime::chacha20Block_Type(),
7597 stubAddr, stubName, TypePtr::BOTTOM,
7598 state_start, result_start);
7599 // return key stream length (int)
7600 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7601 set_result(retvalue);
7602 return true;
7603 }
7604
7605 bool LibraryCallKit::inline_base64_encodeBlock() {
7606 address stubAddr;
7607 const char *stubName;
7608 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7609 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
7610 stubAddr = StubRoutines::base64_encodeBlock();
7611 stubName = "encodeBlock";
7612
7613 if (!stubAddr) return false;
7614 Node* base64obj = argument(0);
7615 Node* src = argument(1);
7616 Node* offset = argument(2);
7617 Node* len = argument(3);
7618 Node* dest = argument(4);
7619 Node* dp = argument(5);
7620 Node* isURL = argument(6);
7621
7622 src = must_be_not_null(src, true);
7623 dest = must_be_not_null(dest, true);
7624
7625 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7626 assert(src_start, "source array is null");
7627 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7628 assert(dest_start, "destination array is null");
7629
7630 Node* base64 = make_runtime_call(RC_LEAF,
7631 OptoRuntime::base64_encodeBlock_Type(),
7632 stubAddr, stubName, TypePtr::BOTTOM,
7633 src_start, offset, len, dest_start, dp, isURL);
7634 return true;
7635 }
7636
7637 bool LibraryCallKit::inline_base64_decodeBlock() {
7638 address stubAddr;
7639 const char *stubName;
7640 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
7641 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
7642 stubAddr = StubRoutines::base64_decodeBlock();
7643 stubName = "decodeBlock";
7644
7645 if (!stubAddr) return false;
7646 Node* base64obj = argument(0);
7647 Node* src = argument(1);
7648 Node* src_offset = argument(2);
7649 Node* len = argument(3);
7650 Node* dest = argument(4);
7651 Node* dest_offset = argument(5);
7652 Node* isURL = argument(6);
7653 Node* isMIME = argument(7);
7654
7655 src = must_be_not_null(src, true);
7656 dest = must_be_not_null(dest, true);
7657
7658 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
7659 assert(src_start, "source array is null");
7660 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
7661 assert(dest_start, "destination array is null");
7662
7663 Node* call = make_runtime_call(RC_LEAF,
7664 OptoRuntime::base64_decodeBlock_Type(),
7665 stubAddr, stubName, TypePtr::BOTTOM,
7666 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
7667 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7668 set_result(result);
7669 return true;
7670 }
7671
7672 bool LibraryCallKit::inline_poly1305_processBlocks() {
7673 address stubAddr;
7674 const char *stubName;
7675 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
7676 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
7677 stubAddr = StubRoutines::poly1305_processBlocks();
7678 stubName = "poly1305_processBlocks";
7679
7680 if (!stubAddr) return false;
7681 null_check_receiver(); // null-check receiver
7682 if (stopped()) return true;
7683
7684 Node* input = argument(1);
7685 Node* input_offset = argument(2);
7686 Node* len = argument(3);
7687 Node* alimbs = argument(4);
7688 Node* rlimbs = argument(5);
7689
7690 input = must_be_not_null(input, true);
7691 alimbs = must_be_not_null(alimbs, true);
7692 rlimbs = must_be_not_null(rlimbs, true);
7693
7694 Node* input_start = array_element_address(input, input_offset, T_BYTE);
7695 assert(input_start, "input array is null");
7696 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
7697 assert(acc_start, "acc array is null");
7698 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
7699 assert(r_start, "r array is null");
7700
7701 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7702 OptoRuntime::poly1305_processBlocks_Type(),
7703 stubAddr, stubName, TypePtr::BOTTOM,
7704 input_start, len, acc_start, r_start);
7705 return true;
7706 }
7707
7708 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
7709 address stubAddr;
7710 const char *stubName;
7711 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7712 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
7713 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
7714 stubName = "intpoly_montgomeryMult_P256";
7715
7716 if (!stubAddr) return false;
7717 null_check_receiver(); // null-check receiver
7718 if (stopped()) return true;
7719
7720 Node* a = argument(1);
7721 Node* b = argument(2);
7722 Node* r = argument(3);
7723
7724 a = must_be_not_null(a, true);
7725 b = must_be_not_null(b, true);
7726 r = must_be_not_null(r, true);
7727
7728 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7729 assert(a_start, "a array is NULL");
7730 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7731 assert(b_start, "b array is NULL");
7732 Node* r_start = array_element_address(r, intcon(0), T_LONG);
7733 assert(r_start, "r array is NULL");
7734
7735 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7736 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
7737 stubAddr, stubName, TypePtr::BOTTOM,
7738 a_start, b_start, r_start);
7739 return true;
7740 }
7741
7742 bool LibraryCallKit::inline_intpoly_assign() {
7743 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
7744 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
7745 const char *stubName = "intpoly_assign";
7746 address stubAddr = StubRoutines::intpoly_assign();
7747 if (!stubAddr) return false;
7748
7749 Node* set = argument(0);
7750 Node* a = argument(1);
7751 Node* b = argument(2);
7752 Node* arr_length = load_array_length(a);
7753
7754 a = must_be_not_null(a, true);
7755 b = must_be_not_null(b, true);
7756
7757 Node* a_start = array_element_address(a, intcon(0), T_LONG);
7758 assert(a_start, "a array is NULL");
7759 Node* b_start = array_element_address(b, intcon(0), T_LONG);
7760 assert(b_start, "b array is NULL");
7761
7762 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
7763 OptoRuntime::intpoly_assign_Type(),
7764 stubAddr, stubName, TypePtr::BOTTOM,
7765 set, a_start, b_start, arr_length);
7766 return true;
7767 }
7768
7769 //------------------------------inline_digestBase_implCompress-----------------------
7770 //
7771 // Calculate MD5 for single-block byte[] array.
7772 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
7773 //
7774 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
7775 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
7776 //
7777 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
7778 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
7779 //
7780 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
7781 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
7782 //
7783 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
7784 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
7785 //
7786 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
7787 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
7788
7789 Node* digestBase_obj = argument(0);
7790 Node* src = argument(1); // type oop
7791 Node* ofs = argument(2); // type int
7792
7793 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7794 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7795 // failed array check
7796 return false;
7797 }
7798 // Figure out the size and type of the elements we will be copying.
7799 BasicType src_elem = src_type->elem()->array_element_basic_type();
7800 if (src_elem != T_BYTE) {
7801 return false;
7802 }
7803 // 'src_start' points to src array + offset
7804 src = must_be_not_null(src, true);
7805 Node* src_start = array_element_address(src, ofs, src_elem);
7806 Node* state = nullptr;
7807 Node* block_size = nullptr;
7808 address stubAddr;
7809 const char *stubName;
7810
7811 switch(id) {
7812 case vmIntrinsics::_md5_implCompress:
7813 assert(UseMD5Intrinsics, "need MD5 instruction support");
7814 state = get_state_from_digest_object(digestBase_obj, T_INT);
7815 stubAddr = StubRoutines::md5_implCompress();
7816 stubName = "md5_implCompress";
7817 break;
7818 case vmIntrinsics::_sha_implCompress:
7819 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
7820 state = get_state_from_digest_object(digestBase_obj, T_INT);
7821 stubAddr = StubRoutines::sha1_implCompress();
7822 stubName = "sha1_implCompress";
7823 break;
7824 case vmIntrinsics::_sha2_implCompress:
7825 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
7826 state = get_state_from_digest_object(digestBase_obj, T_INT);
7827 stubAddr = StubRoutines::sha256_implCompress();
7828 stubName = "sha256_implCompress";
7829 break;
7830 case vmIntrinsics::_sha5_implCompress:
7831 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
7832 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7833 stubAddr = StubRoutines::sha512_implCompress();
7834 stubName = "sha512_implCompress";
7835 break;
7836 case vmIntrinsics::_sha3_implCompress:
7837 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
7838 state = get_state_from_digest_object(digestBase_obj, T_LONG);
7839 stubAddr = StubRoutines::sha3_implCompress();
7840 stubName = "sha3_implCompress";
7841 block_size = get_block_size_from_digest_object(digestBase_obj);
7842 if (block_size == nullptr) return false;
7843 break;
7844 default:
7845 fatal_unexpected_iid(id);
7846 return false;
7847 }
7848 if (state == nullptr) return false;
7849
7850 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
7851 if (stubAddr == nullptr) return false;
7852
7853 // Call the stub.
7854 Node* call;
7855 if (block_size == nullptr) {
7856 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
7857 stubAddr, stubName, TypePtr::BOTTOM,
7858 src_start, state);
7859 } else {
7860 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
7861 stubAddr, stubName, TypePtr::BOTTOM,
7862 src_start, state, block_size);
7863 }
7864
7865 return true;
7866 }
7867
7868 //------------------------------inline_digestBase_implCompressMB-----------------------
7869 //
7870 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
7871 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
7872 //
7873 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
7874 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
7875 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
7876 assert((uint)predicate < 5, "sanity");
7877 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
7878
7879 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
7880 Node* src = argument(1); // byte[] array
7881 Node* ofs = argument(2); // type int
7882 Node* limit = argument(3); // type int
7883
7884 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7885 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7886 // failed array check
7887 return false;
7888 }
7889 // Figure out the size and type of the elements we will be copying.
7890 BasicType src_elem = src_type->elem()->array_element_basic_type();
7891 if (src_elem != T_BYTE) {
7892 return false;
7893 }
7894 // 'src_start' points to src array + offset
7895 src = must_be_not_null(src, false);
7896 Node* src_start = array_element_address(src, ofs, src_elem);
7897
7898 const char* klass_digestBase_name = nullptr;
7899 const char* stub_name = nullptr;
7900 address stub_addr = nullptr;
7901 BasicType elem_type = T_INT;
7902
7903 switch (predicate) {
7904 case 0:
7905 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
7906 klass_digestBase_name = "sun/security/provider/MD5";
7907 stub_name = "md5_implCompressMB";
7908 stub_addr = StubRoutines::md5_implCompressMB();
7909 }
7910 break;
7911 case 1:
7912 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
7913 klass_digestBase_name = "sun/security/provider/SHA";
7914 stub_name = "sha1_implCompressMB";
7915 stub_addr = StubRoutines::sha1_implCompressMB();
7916 }
7917 break;
7918 case 2:
7919 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
7920 klass_digestBase_name = "sun/security/provider/SHA2";
7921 stub_name = "sha256_implCompressMB";
7922 stub_addr = StubRoutines::sha256_implCompressMB();
7923 }
7924 break;
7925 case 3:
7926 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
7927 klass_digestBase_name = "sun/security/provider/SHA5";
7928 stub_name = "sha512_implCompressMB";
7929 stub_addr = StubRoutines::sha512_implCompressMB();
7930 elem_type = T_LONG;
7931 }
7932 break;
7933 case 4:
7934 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
7935 klass_digestBase_name = "sun/security/provider/SHA3";
7936 stub_name = "sha3_implCompressMB";
7937 stub_addr = StubRoutines::sha3_implCompressMB();
7938 elem_type = T_LONG;
7939 }
7940 break;
7941 default:
7942 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
7943 }
7944 if (klass_digestBase_name != nullptr) {
7945 assert(stub_addr != nullptr, "Stub is generated");
7946 if (stub_addr == nullptr) return false;
7947
7948 // get DigestBase klass to lookup for SHA klass
7949 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
7950 assert(tinst != nullptr, "digestBase_obj is not instance???");
7951 assert(tinst->is_loaded(), "DigestBase is not loaded");
7952
7953 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
7954 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
7955 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
7956 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
7957 }
7958 return false;
7959 }
7960
7961 //------------------------------inline_digestBase_implCompressMB-----------------------
7962 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
7963 BasicType elem_type, address stubAddr, const char *stubName,
7964 Node* src_start, Node* ofs, Node* limit) {
7965 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
7966 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7967 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
7968 digest_obj = _gvn.transform(digest_obj);
7969
7970 Node* state = get_state_from_digest_object(digest_obj, elem_type);
7971 if (state == nullptr) return false;
7972
7973 Node* block_size = nullptr;
7974 if (strcmp("sha3_implCompressMB", stubName) == 0) {
7975 block_size = get_block_size_from_digest_object(digest_obj);
7976 if (block_size == nullptr) return false;
7977 }
7978
7979 // Call the stub.
7980 Node* call;
7981 if (block_size == nullptr) {
7982 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7983 OptoRuntime::digestBase_implCompressMB_Type(false),
7984 stubAddr, stubName, TypePtr::BOTTOM,
7985 src_start, state, ofs, limit);
7986 } else {
7987 call = make_runtime_call(RC_LEAF|RC_NO_FP,
7988 OptoRuntime::digestBase_implCompressMB_Type(true),
7989 stubAddr, stubName, TypePtr::BOTTOM,
7990 src_start, state, block_size, ofs, limit);
7991 }
7992
7993 // return ofs (int)
7994 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7995 set_result(result);
7996
7997 return true;
7998 }
7999
8000 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8001 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8002 assert(UseAES, "need AES instruction support");
8003 address stubAddr = nullptr;
8004 const char *stubName = nullptr;
8005 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8006 stubName = "galoisCounterMode_AESCrypt";
8007
8008 if (stubAddr == nullptr) return false;
8009
8010 Node* in = argument(0);
8011 Node* inOfs = argument(1);
8012 Node* len = argument(2);
8013 Node* ct = argument(3);
8014 Node* ctOfs = argument(4);
8015 Node* out = argument(5);
8016 Node* outOfs = argument(6);
8017 Node* gctr_object = argument(7);
8018 Node* ghash_object = argument(8);
8019
8020 // (1) in, ct and out are arrays.
8021 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8022 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8023 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8024 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8025 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8026 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8027
8028 // checks are the responsibility of the caller
8029 Node* in_start = in;
8030 Node* ct_start = ct;
8031 Node* out_start = out;
8032 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8033 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8034 in_start = array_element_address(in, inOfs, T_BYTE);
8035 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8036 out_start = array_element_address(out, outOfs, T_BYTE);
8037 }
8038
8039 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8040 // (because of the predicated logic executed earlier).
8041 // so we cast it here safely.
8042 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8043 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8044 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8045 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8046 Node* state = load_field_from_object(ghash_object, "state", "[J");
8047
8048 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8049 return false;
8050 }
8051 // cast it to what we know it will be at runtime
8052 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8053 assert(tinst != nullptr, "GCTR obj is null");
8054 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8055 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8056 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8057 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8058 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8059 const TypeOopPtr* xtype = aklass->as_instance_type();
8060 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8061 aescrypt_object = _gvn.transform(aescrypt_object);
8062 // we need to get the start of the aescrypt_object's expanded key array
8063 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8064 if (k_start == nullptr) return false;
8065 // similarly, get the start address of the r vector
8066 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8067 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8068 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8069
8070
8071 // Call the stub, passing params
8072 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8073 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8074 stubAddr, stubName, TypePtr::BOTTOM,
8075 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8076
8077 // return cipher length (int)
8078 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8079 set_result(retvalue);
8080
8081 return true;
8082 }
8083
8084 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8085 // Return node representing slow path of predicate check.
8086 // the pseudo code we want to emulate with this predicate is:
8087 // for encryption:
8088 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8089 // for decryption:
8090 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8091 // note cipher==plain is more conservative than the original java code but that's OK
8092 //
8093
8094 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8095 // The receiver was checked for null already.
8096 Node* objGCTR = argument(7);
8097 // Load embeddedCipher field of GCTR object.
8098 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8099 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8100
8101 // get AESCrypt klass for instanceOf check
8102 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8103 // will have same classloader as CipherBlockChaining object
8104 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8105 assert(tinst != nullptr, "GCTR obj is null");
8106 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8107
8108 // we want to do an instanceof comparison against the AESCrypt class
8109 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8110 if (!klass_AESCrypt->is_loaded()) {
8111 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8112 Node* ctrl = control();
8113 set_control(top()); // no regular fast path
8114 return ctrl;
8115 }
8116
8117 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8118 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8119 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8120 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8121 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8122
8123 return instof_false; // even if it is null
8124 }
8125
8126 //------------------------------get_state_from_digest_object-----------------------
8127 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8128 const char* state_type;
8129 switch (elem_type) {
8130 case T_BYTE: state_type = "[B"; break;
8131 case T_INT: state_type = "[I"; break;
8132 case T_LONG: state_type = "[J"; break;
8133 default: ShouldNotReachHere();
8134 }
8135 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8136 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8137 if (digest_state == nullptr) return (Node *) nullptr;
8138
8139 // now have the array, need to get the start address of the state array
8140 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8141 return state;
8142 }
8143
8144 //------------------------------get_block_size_from_sha3_object----------------------------------
8145 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8146 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8147 assert (block_size != nullptr, "sanity");
8148 return block_size;
8149 }
8150
8151 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8152 // Return node representing slow path of predicate check.
8153 // the pseudo code we want to emulate with this predicate is:
8154 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8155 //
8156 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8157 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8158 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8159 assert((uint)predicate < 5, "sanity");
8160
8161 // The receiver was checked for null already.
8162 Node* digestBaseObj = argument(0);
8163
8164 // get DigestBase klass for instanceOf check
8165 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8166 assert(tinst != nullptr, "digestBaseObj is null");
8167 assert(tinst->is_loaded(), "DigestBase is not loaded");
8168
8169 const char* klass_name = nullptr;
8170 switch (predicate) {
8171 case 0:
8172 if (UseMD5Intrinsics) {
8173 // we want to do an instanceof comparison against the MD5 class
8174 klass_name = "sun/security/provider/MD5";
8175 }
8176 break;
8177 case 1:
8178 if (UseSHA1Intrinsics) {
8179 // we want to do an instanceof comparison against the SHA class
8180 klass_name = "sun/security/provider/SHA";
8181 }
8182 break;
8183 case 2:
8184 if (UseSHA256Intrinsics) {
8185 // we want to do an instanceof comparison against the SHA2 class
8186 klass_name = "sun/security/provider/SHA2";
8187 }
8188 break;
8189 case 3:
8190 if (UseSHA512Intrinsics) {
8191 // we want to do an instanceof comparison against the SHA5 class
8192 klass_name = "sun/security/provider/SHA5";
8193 }
8194 break;
8195 case 4:
8196 if (UseSHA3Intrinsics) {
8197 // we want to do an instanceof comparison against the SHA3 class
8198 klass_name = "sun/security/provider/SHA3";
8199 }
8200 break;
8201 default:
8202 fatal("unknown SHA intrinsic predicate: %d", predicate);
8203 }
8204
8205 ciKlass* klass = nullptr;
8206 if (klass_name != nullptr) {
8207 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8208 }
8209 if ((klass == nullptr) || !klass->is_loaded()) {
8210 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8211 Node* ctrl = control();
8212 set_control(top()); // no intrinsic path
8213 return ctrl;
8214 }
8215 ciInstanceKlass* instklass = klass->as_instance_klass();
8216
8217 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8218 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8219 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8220 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8221
8222 return instof_false; // even if it is null
8223 }
8224
8225 //-------------inline_fma-----------------------------------
8226 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8227 Node *a = nullptr;
8228 Node *b = nullptr;
8229 Node *c = nullptr;
8230 Node* result = nullptr;
8231 switch (id) {
8232 case vmIntrinsics::_fmaD:
8233 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8234 // no receiver since it is static method
8235 a = round_double_node(argument(0));
8236 b = round_double_node(argument(2));
8237 c = round_double_node(argument(4));
8238 result = _gvn.transform(new FmaDNode(control(), a, b, c));
8239 break;
8240 case vmIntrinsics::_fmaF:
8241 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8242 a = argument(0);
8243 b = argument(1);
8244 c = argument(2);
8245 result = _gvn.transform(new FmaFNode(control(), a, b, c));
8246 break;
8247 default:
8248 fatal_unexpected_iid(id); break;
8249 }
8250 set_result(result);
8251 return true;
8252 }
8253
8254 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8255 // argument(0) is receiver
8256 Node* codePoint = argument(1);
8257 Node* n = nullptr;
8258
8259 switch (id) {
8260 case vmIntrinsics::_isDigit :
8261 n = new DigitNode(control(), codePoint);
8262 break;
8263 case vmIntrinsics::_isLowerCase :
8264 n = new LowerCaseNode(control(), codePoint);
8265 break;
8266 case vmIntrinsics::_isUpperCase :
8267 n = new UpperCaseNode(control(), codePoint);
8268 break;
8269 case vmIntrinsics::_isWhitespace :
8270 n = new WhitespaceNode(control(), codePoint);
8271 break;
8272 default:
8273 fatal_unexpected_iid(id);
8274 }
8275
8276 set_result(_gvn.transform(n));
8277 return true;
8278 }
8279
8280 //------------------------------inline_fp_min_max------------------------------
8281 bool LibraryCallKit::inline_fp_min_max(vmIntrinsics::ID id) {
8282 /* DISABLED BECAUSE METHOD DATA ISN'T COLLECTED PER CALL-SITE, SEE JDK-8015416.
8283
8284 // The intrinsic should be used only when the API branches aren't predictable,
8285 // the last one performing the most important comparison. The following heuristic
8286 // uses the branch statistics to eventually bail out if necessary.
8287
8288 ciMethodData *md = callee()->method_data();
8289
8290 if ( md != nullptr && md->is_mature() && md->invocation_count() > 0 ) {
8291 ciCallProfile cp = caller()->call_profile_at_bci(bci());
8292
8293 if ( ((double)cp.count()) / ((double)md->invocation_count()) < 0.8 ) {
8294 // Bail out if the call-site didn't contribute enough to the statistics.
8295 return false;
8296 }
8297
8298 uint taken = 0, not_taken = 0;
8299
8300 for (ciProfileData *p = md->first_data(); md->is_valid(p); p = md->next_data(p)) {
8301 if (p->is_BranchData()) {
8302 taken = ((ciBranchData*)p)->taken();
8303 not_taken = ((ciBranchData*)p)->not_taken();
8304 }
8305 }
8306
8307 double balance = (((double)taken) - ((double)not_taken)) / ((double)md->invocation_count());
8308 balance = balance < 0 ? -balance : balance;
8309 if ( balance > 0.2 ) {
8310 // Bail out if the most important branch is predictable enough.
8311 return false;
8312 }
8313 }
8314 */
8315
8316 Node *a = nullptr;
8317 Node *b = nullptr;
8318 Node *n = nullptr;
8319 switch (id) {
8320 case vmIntrinsics::_maxF:
8321 case vmIntrinsics::_minF:
8322 case vmIntrinsics::_maxF_strict:
8323 case vmIntrinsics::_minF_strict:
8324 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
8325 a = argument(0);
8326 b = argument(1);
8327 break;
8328 case vmIntrinsics::_maxD:
8329 case vmIntrinsics::_minD:
8330 case vmIntrinsics::_maxD_strict:
8331 case vmIntrinsics::_minD_strict:
8332 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
8333 a = round_double_node(argument(0));
8334 b = round_double_node(argument(2));
8335 break;
8336 default:
8337 fatal_unexpected_iid(id);
8338 break;
8339 }
8340 switch (id) {
8341 case vmIntrinsics::_maxF:
8342 case vmIntrinsics::_maxF_strict:
8343 n = new MaxFNode(a, b);
8344 break;
8345 case vmIntrinsics::_minF:
8346 case vmIntrinsics::_minF_strict:
8347 n = new MinFNode(a, b);
8348 break;
8349 case vmIntrinsics::_maxD:
8350 case vmIntrinsics::_maxD_strict:
8351 n = new MaxDNode(a, b);
8352 break;
8353 case vmIntrinsics::_minD:
8354 case vmIntrinsics::_minD_strict:
8355 n = new MinDNode(a, b);
8356 break;
8357 default:
8358 fatal_unexpected_iid(id);
8359 break;
8360 }
8361 set_result(_gvn.transform(n));
8362 return true;
8363 }
8364
8365 bool LibraryCallKit::inline_profileBoolean() {
8366 Node* counts = argument(1);
8367 const TypeAryPtr* ary = nullptr;
8368 ciArray* aobj = nullptr;
8369 if (counts->is_Con()
8370 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8371 && (aobj = ary->const_oop()->as_array()) != nullptr
8372 && (aobj->length() == 2)) {
8373 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8374 jint false_cnt = aobj->element_value(0).as_int();
8375 jint true_cnt = aobj->element_value(1).as_int();
8376
8377 if (C->log() != nullptr) {
8378 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8379 false_cnt, true_cnt);
8380 }
8381
8382 if (false_cnt + true_cnt == 0) {
8383 // According to profile, never executed.
8384 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8385 Deoptimization::Action_reinterpret);
8386 return true;
8387 }
8388
8389 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8390 // is a number of each value occurrences.
8391 Node* result = argument(0);
8392 if (false_cnt == 0 || true_cnt == 0) {
8393 // According to profile, one value has been never seen.
8394 int expected_val = (false_cnt == 0) ? 1 : 0;
8395
8396 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8397 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8398
8399 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8400 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8401 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8402
8403 { // Slow path: uncommon trap for never seen value and then reexecute
8404 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8405 // the value has been seen at least once.
8406 PreserveJVMState pjvms(this);
8407 PreserveReexecuteState preexecs(this);
8408 jvms()->set_should_reexecute(true);
8409
8410 set_control(slow_path);
8411 set_i_o(i_o());
8412
8413 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8414 Deoptimization::Action_reinterpret);
8415 }
8416 // The guard for never seen value enables sharpening of the result and
8417 // returning a constant. It allows to eliminate branches on the same value
8418 // later on.
8419 set_control(fast_path);
8420 result = intcon(expected_val);
8421 }
8422 // Stop profiling.
8423 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8424 // By replacing method body with profile data (represented as ProfileBooleanNode
8425 // on IR level) we effectively disable profiling.
8426 // It enables full speed execution once optimized code is generated.
8427 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8428 C->record_for_igvn(profile);
8429 set_result(profile);
8430 return true;
8431 } else {
8432 // Continue profiling.
8433 // Profile data isn't available at the moment. So, execute method's bytecode version.
8434 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8435 // is compiled and counters aren't available since corresponding MethodHandle
8436 // isn't a compile-time constant.
8437 return false;
8438 }
8439 }
8440
8441 bool LibraryCallKit::inline_isCompileConstant() {
8442 Node* n = argument(0);
8443 set_result(n->is_Con() ? intcon(1) : intcon(0));
8444 return true;
8445 }
8446
8447 //------------------------------- inline_getObjectSize --------------------------------------
8448 //
8449 // Calculate the runtime size of the object/array.
8450 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8451 //
8452 bool LibraryCallKit::inline_getObjectSize() {
8453 Node* obj = argument(3);
8454 Node* klass_node = load_object_klass(obj);
8455
8456 jint layout_con = Klass::_lh_neutral_value;
8457 Node* layout_val = get_layout_helper(klass_node, layout_con);
8458 int layout_is_con = (layout_val == nullptr);
8459
8460 if (layout_is_con) {
8461 // Layout helper is constant, can figure out things at compile time.
8462
8463 if (Klass::layout_helper_is_instance(layout_con)) {
8464 // Instance case: layout_con contains the size itself.
8465 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8466 set_result(size);
8467 } else {
8468 // Array case: size is round(header + element_size*arraylength).
8469 // Since arraylength is different for every array instance, we have to
8470 // compute the whole thing at runtime.
8471
8472 Node* arr_length = load_array_length(obj);
8473
8474 int round_mask = MinObjAlignmentInBytes - 1;
8475 int hsize = Klass::layout_helper_header_size(layout_con);
8476 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8477
8478 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8479 round_mask = 0; // strength-reduce it if it goes away completely
8480 }
8481 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8482 Node* header_size = intcon(hsize + round_mask);
8483
8484 Node* lengthx = ConvI2X(arr_length);
8485 Node* headerx = ConvI2X(header_size);
8486
8487 Node* abody = lengthx;
8488 if (eshift != 0) {
8489 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8490 }
8491 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8492 if (round_mask != 0) {
8493 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8494 }
8495 size = ConvX2L(size);
8496 set_result(size);
8497 }
8498 } else {
8499 // Layout helper is not constant, need to test for array-ness at runtime.
8500
8501 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8502 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8503 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8504 record_for_igvn(result_reg);
8505
8506 Node* array_ctl = generate_array_guard(klass_node, nullptr);
8507 if (array_ctl != nullptr) {
8508 // Array case: size is round(header + element_size*arraylength).
8509 // Since arraylength is different for every array instance, we have to
8510 // compute the whole thing at runtime.
8511
8512 PreserveJVMState pjvms(this);
8513 set_control(array_ctl);
8514 Node* arr_length = load_array_length(obj);
8515
8516 int round_mask = MinObjAlignmentInBytes - 1;
8517 Node* mask = intcon(round_mask);
8518
8519 Node* hss = intcon(Klass::_lh_header_size_shift);
8520 Node* hsm = intcon(Klass::_lh_header_size_mask);
8521 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8522 header_size = _gvn.transform(new AndINode(header_size, hsm));
8523 header_size = _gvn.transform(new AddINode(header_size, mask));
8524
8525 // There is no need to mask or shift this value.
8526 // The semantics of LShiftINode include an implicit mask to 0x1F.
8527 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8528 Node* elem_shift = layout_val;
8529
8530 Node* lengthx = ConvI2X(arr_length);
8531 Node* headerx = ConvI2X(header_size);
8532
8533 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
8534 Node* size = _gvn.transform(new AddXNode(headerx, abody));
8535 if (round_mask != 0) {
8536 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
8537 }
8538 size = ConvX2L(size);
8539
8540 result_reg->init_req(_array_path, control());
8541 result_val->init_req(_array_path, size);
8542 }
8543
8544 if (!stopped()) {
8545 // Instance case: the layout helper gives us instance size almost directly,
8546 // but we need to mask out the _lh_instance_slow_path_bit.
8547 Node* size = ConvI2X(layout_val);
8548 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
8549 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
8550 size = _gvn.transform(new AndXNode(size, mask));
8551 size = ConvX2L(size);
8552
8553 result_reg->init_req(_instance_path, control());
8554 result_val->init_req(_instance_path, size);
8555 }
8556
8557 set_result(result_reg, result_val);
8558 }
8559
8560 return true;
8561 }
8562
8563 //------------------------------- inline_blackhole --------------------------------------
8564 //
8565 // Make sure all arguments to this node are alive.
8566 // This matches methods that were requested to be blackholed through compile commands.
8567 //
8568 bool LibraryCallKit::inline_blackhole() {
8569 assert(callee()->is_static(), "Should have been checked before: only static methods here");
8570 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
8571 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
8572
8573 // Blackhole node pinches only the control, not memory. This allows
8574 // the blackhole to be pinned in the loop that computes blackholed
8575 // values, but have no other side effects, like breaking the optimizations
8576 // across the blackhole.
8577
8578 Node* bh = _gvn.transform(new BlackholeNode(control()));
8579 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
8580
8581 // Bind call arguments as blackhole arguments to keep them alive
8582 uint nargs = callee()->arg_size();
8583 for (uint i = 0; i < nargs; i++) {
8584 bh->add_req(argument(i));
8585 }
8586
8587 return true;
8588 }