1 /*
2 * Copyright (c) 1999, 2025, 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 "asm/macroAssembler.hpp"
26 #include "ci/ciArrayKlass.hpp"
27 #include "ci/ciFlatArrayKlass.hpp"
28 #include "ci/ciInstanceKlass.hpp"
29 #include "ci/ciSymbols.hpp"
30 #include "ci/ciUtilities.inline.hpp"
31 #include "classfile/vmIntrinsics.hpp"
32 #include "compiler/compileBroker.hpp"
33 #include "compiler/compileLog.hpp"
34 #include "gc/shared/barrierSet.hpp"
35 #include "gc/shared/c2/barrierSetC2.hpp"
36 #include "jfr/support/jfrIntrinsics.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "oops/accessDecorators.hpp"
39 #include "oops/klass.inline.hpp"
40 #include "oops/layoutKind.hpp"
41 #include "oops/objArrayKlass.hpp"
42 #include "opto/addnode.hpp"
43 #include "opto/arraycopynode.hpp"
44 #include "opto/c2compiler.hpp"
45 #include "opto/castnode.hpp"
46 #include "opto/cfgnode.hpp"
47 #include "opto/convertnode.hpp"
48 #include "opto/countbitsnode.hpp"
49 #include "opto/graphKit.hpp"
50 #include "opto/idealKit.hpp"
51 #include "opto/inlinetypenode.hpp"
52 #include "opto/library_call.hpp"
53 #include "opto/mathexactnode.hpp"
54 #include "opto/mulnode.hpp"
55 #include "opto/narrowptrnode.hpp"
56 #include "opto/opaquenode.hpp"
57 #include "opto/opcodes.hpp"
58 #include "opto/parse.hpp"
59 #include "opto/rootnode.hpp"
60 #include "opto/runtime.hpp"
61 #include "opto/subnode.hpp"
62 #include "opto/type.hpp"
63 #include "opto/vectornode.hpp"
64 #include "prims/jvmtiExport.hpp"
65 #include "prims/jvmtiThreadState.hpp"
66 #include "prims/unsafe.hpp"
67 #include "runtime/jniHandles.inline.hpp"
68 #include "runtime/objectMonitor.hpp"
69 #include "runtime/sharedRuntime.hpp"
70 #include "runtime/stubRoutines.hpp"
71 #include "utilities/globalDefinitions.hpp"
72 #include "utilities/macros.hpp"
73 #include "utilities/powerOfTwo.hpp"
74
75 //---------------------------make_vm_intrinsic----------------------------
76 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
77 vmIntrinsicID id = m->intrinsic_id();
78 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
79
80 if (!m->is_loaded()) {
81 // Do not attempt to inline unloaded methods.
82 return nullptr;
83 }
84
85 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
86 bool is_available = false;
87
88 {
89 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
90 // the compiler must transition to '_thread_in_vm' state because both
91 // methods access VM-internal data.
92 VM_ENTRY_MARK;
93 methodHandle mh(THREAD, m->get_Method());
94 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
95 if (is_available && is_virtual) {
96 is_available = vmIntrinsics::does_virtual_dispatch(id);
97 }
98 }
99
100 if (is_available) {
101 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
102 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
103 return new LibraryIntrinsic(m, is_virtual,
104 vmIntrinsics::predicates_needed(id),
105 vmIntrinsics::does_virtual_dispatch(id),
106 id);
107 } else {
108 return nullptr;
109 }
110 }
111
112 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
113 LibraryCallKit kit(jvms, this);
114 Compile* C = kit.C;
115 int nodes = C->unique();
116 #ifndef PRODUCT
117 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
118 char buf[1000];
119 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
120 tty->print_cr("Intrinsic %s", str);
121 }
122 #endif
123 ciMethod* callee = kit.callee();
124 const int bci = kit.bci();
125 #ifdef ASSERT
126 Node* ctrl = kit.control();
127 #endif
128 // Try to inline the intrinsic.
129 if (callee->check_intrinsic_candidate() &&
130 kit.try_to_inline(_last_predicate)) {
131 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
132 : "(intrinsic)";
133 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
134 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
135 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
136 if (C->log()) {
137 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
138 vmIntrinsics::name_at(intrinsic_id()),
139 (is_virtual() ? " virtual='1'" : ""),
140 C->unique() - nodes);
141 }
142 // Push the result from the inlined method onto the stack.
143 kit.push_result();
144 return kit.transfer_exceptions_into_jvms();
145 }
146
147 // The intrinsic bailed out
148 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
149 assert(jvms->map() == kit.map(), "Out of sync JVM state");
150 if (jvms->has_method()) {
151 // Not a root compile.
152 const char* msg;
153 if (callee->intrinsic_candidate()) {
154 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
155 } else {
156 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
157 : "failed to inline (intrinsic), method not annotated";
158 }
159 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
160 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
161 } else {
162 // Root compile
163 ResourceMark rm;
164 stringStream msg_stream;
165 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
166 vmIntrinsics::name_at(intrinsic_id()),
167 is_virtual() ? " (virtual)" : "", bci);
168 const char *msg = msg_stream.freeze();
169 log_debug(jit, inlining)("%s", msg);
170 if (C->print_intrinsics() || C->print_inlining()) {
171 tty->print("%s", msg);
172 }
173 }
174 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
175
176 return nullptr;
177 }
178
179 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
180 LibraryCallKit kit(jvms, this);
181 Compile* C = kit.C;
182 int nodes = C->unique();
183 _last_predicate = predicate;
184 #ifndef PRODUCT
185 assert(is_predicated() && predicate < predicates_count(), "sanity");
186 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
187 char buf[1000];
188 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
189 tty->print_cr("Predicate for intrinsic %s", str);
190 }
191 #endif
192 ciMethod* callee = kit.callee();
193 const int bci = kit.bci();
194
195 Node* slow_ctl = kit.try_to_predicate(predicate);
196 if (!kit.failing()) {
197 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
198 : "(intrinsic, predicate)";
199 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
200 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
201
202 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
203 if (C->log()) {
204 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
205 vmIntrinsics::name_at(intrinsic_id()),
206 (is_virtual() ? " virtual='1'" : ""),
207 C->unique() - nodes);
208 }
209 return slow_ctl; // Could be null if the check folds.
210 }
211
212 // The intrinsic bailed out
213 if (jvms->has_method()) {
214 // Not a root compile.
215 const char* msg = "failed to generate predicate for intrinsic";
216 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 } else {
219 // Root compile
220 ResourceMark rm;
221 stringStream msg_stream;
222 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
223 vmIntrinsics::name_at(intrinsic_id()),
224 is_virtual() ? " (virtual)" : "", bci);
225 const char *msg = msg_stream.freeze();
226 log_debug(jit, inlining)("%s", msg);
227 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
228 }
229 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
230 return nullptr;
231 }
232
233 bool LibraryCallKit::try_to_inline(int predicate) {
234 // Handle symbolic names for otherwise undistinguished boolean switches:
235 const bool is_store = true;
236 const bool is_compress = true;
237 const bool is_static = true;
238 const bool is_volatile = true;
239
240 if (!jvms()->has_method()) {
241 // Root JVMState has a null method.
242 assert(map()->memory()->Opcode() == Op_Parm, "");
243 // Insert the memory aliasing node
244 set_all_memory(reset_memory());
245 }
246 assert(merged_memory(), "");
247
248 switch (intrinsic_id()) {
249 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
250 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
251 case vmIntrinsics::_getClass: return inline_native_getClass();
252
253 case vmIntrinsics::_ceil:
254 case vmIntrinsics::_floor:
255 case vmIntrinsics::_rint:
256 case vmIntrinsics::_dsin:
257 case vmIntrinsics::_dcos:
258 case vmIntrinsics::_dtan:
259 case vmIntrinsics::_dsinh:
260 case vmIntrinsics::_dtanh:
261 case vmIntrinsics::_dcbrt:
262 case vmIntrinsics::_dabs:
263 case vmIntrinsics::_fabs:
264 case vmIntrinsics::_iabs:
265 case vmIntrinsics::_labs:
266 case vmIntrinsics::_datan2:
267 case vmIntrinsics::_dsqrt:
268 case vmIntrinsics::_dsqrt_strict:
269 case vmIntrinsics::_dexp:
270 case vmIntrinsics::_dlog:
271 case vmIntrinsics::_dlog10:
272 case vmIntrinsics::_dpow:
273 case vmIntrinsics::_dcopySign:
274 case vmIntrinsics::_fcopySign:
275 case vmIntrinsics::_dsignum:
276 case vmIntrinsics::_roundF:
277 case vmIntrinsics::_roundD:
278 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
279
280 case vmIntrinsics::_notify:
281 case vmIntrinsics::_notifyAll:
282 return inline_notify(intrinsic_id());
283
284 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
285 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
286 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
287 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
288 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
289 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
290 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
291 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
292 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
293 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
294 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
295 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
296 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
297 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
298
299 case vmIntrinsics::_arraycopy: return inline_arraycopy();
300
301 case vmIntrinsics::_arraySort: return inline_array_sort();
302 case vmIntrinsics::_arrayPartition: return inline_array_partition();
303
304 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
305 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
306 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
307 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
308
309 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
310 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
311 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
312 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
313 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
314 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
315 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
316 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
317
318 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
319
320 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
321
322 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
323 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
324 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
325 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
326
327 case vmIntrinsics::_compressStringC:
328 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
329 case vmIntrinsics::_inflateStringC:
330 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
331
332 case vmIntrinsics::_makePrivateBuffer: return inline_unsafe_make_private_buffer();
333 case vmIntrinsics::_finishPrivateBuffer: return inline_unsafe_finish_private_buffer();
334 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
335 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
336 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
337 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
338 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
339 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
340 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
341 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
342 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
343 case vmIntrinsics::_getValue: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false, true);
344
345 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
346 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
347 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
348 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
349 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
350 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
351 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
352 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
353 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
354 case vmIntrinsics::_putValue: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false, true);
355
356 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
357 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
358 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
359 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
360 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
361 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
362 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
363 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
364 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
365
366 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
367 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
368 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
369 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
370 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
371 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
372 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
373 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
374 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
375
376 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
377 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
378 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
379 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
380
381 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
382 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
383 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
384 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
385
386 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
387 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
388 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
389 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
390 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
391 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
392 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
393 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
394 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
395
396 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
397 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
398 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
399 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
400 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
401 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
402 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
403 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
404 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
405
406 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
407 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
408 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
409 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
410 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
411 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
412 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
413 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
414 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
415
416 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
417 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
418 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
419 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
420 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
421 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
422 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
423 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
424 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
425
426 case vmIntrinsics::_getFlatValue: return inline_unsafe_flat_access(!is_store, Relaxed);
427 case vmIntrinsics::_putFlatValue: return inline_unsafe_flat_access( is_store, Relaxed);
428
429 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
430 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
431 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
432 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
433 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
434
435 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
436 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
437 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
438 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
439 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
440 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
441 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
442 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
443 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
444 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
445 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
446 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
447 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
448 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
449 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
450 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
451 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
452 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
453 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
454 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
455
456 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
457 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
458 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
459 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
460 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
461 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
462 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
463 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
464 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
465 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
466 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
467 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
468 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
469 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
470 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
471
472 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
473 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
474 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
475 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
476
477 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
478 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
479 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
480 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
481 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
482
483 case vmIntrinsics::_loadFence:
484 case vmIntrinsics::_storeFence:
485 case vmIntrinsics::_storeStoreFence:
486 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
487
488 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
489
490 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
491 case vmIntrinsics::_currentThread: return inline_native_currentThread();
492 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
493
494 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
495 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
496
497 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
498 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
499
500 #if INCLUDE_JVMTI
501 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
502 "notifyJvmtiStart", true, false);
503 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
504 "notifyJvmtiEnd", false, true);
505 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
506 "notifyJvmtiMount", false, false);
507 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
508 "notifyJvmtiUnmount", false, false);
509 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
510 #endif
511
512 #ifdef JFR_HAVE_INTRINSICS
513 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
514 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
515 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
516 #endif
517 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
518 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
519 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
520 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
521 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
522 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
523 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
524 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
525 case vmIntrinsics::_getLength: return inline_native_getLength();
526 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
527 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
528 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
529 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
530 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
531 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
532 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
533
534 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
535 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
536 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
537 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
538 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
539
540 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
541
542 case vmIntrinsics::_isInstance:
543 case vmIntrinsics::_isHidden:
544 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
545
546 case vmIntrinsics::_floatToRawIntBits:
547 case vmIntrinsics::_floatToIntBits:
548 case vmIntrinsics::_intBitsToFloat:
549 case vmIntrinsics::_doubleToRawLongBits:
550 case vmIntrinsics::_doubleToLongBits:
551 case vmIntrinsics::_longBitsToDouble:
552 case vmIntrinsics::_floatToFloat16:
553 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
554 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
555 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
556 case vmIntrinsics::_floatIsFinite:
557 case vmIntrinsics::_floatIsInfinite:
558 case vmIntrinsics::_doubleIsFinite:
559 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
560
561 case vmIntrinsics::_numberOfLeadingZeros_i:
562 case vmIntrinsics::_numberOfLeadingZeros_l:
563 case vmIntrinsics::_numberOfTrailingZeros_i:
564 case vmIntrinsics::_numberOfTrailingZeros_l:
565 case vmIntrinsics::_bitCount_i:
566 case vmIntrinsics::_bitCount_l:
567 case vmIntrinsics::_reverse_i:
568 case vmIntrinsics::_reverse_l:
569 case vmIntrinsics::_reverseBytes_i:
570 case vmIntrinsics::_reverseBytes_l:
571 case vmIntrinsics::_reverseBytes_s:
572 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
573
574 case vmIntrinsics::_compress_i:
575 case vmIntrinsics::_compress_l:
576 case vmIntrinsics::_expand_i:
577 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
578
579 case vmIntrinsics::_compareUnsigned_i:
580 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
581
582 case vmIntrinsics::_divideUnsigned_i:
583 case vmIntrinsics::_divideUnsigned_l:
584 case vmIntrinsics::_remainderUnsigned_i:
585 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
586
587 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
588
589 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
590 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
591 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
592 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
593 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
594
595 case vmIntrinsics::_Class_cast: return inline_Class_cast();
596
597 case vmIntrinsics::_aescrypt_encryptBlock:
598 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
599
600 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
601 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
602 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
603
604 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
605 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
606 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
607
608 case vmIntrinsics::_counterMode_AESCrypt:
609 return inline_counterMode_AESCrypt(intrinsic_id());
610
611 case vmIntrinsics::_galoisCounterMode_AESCrypt:
612 return inline_galoisCounterMode_AESCrypt();
613
614 case vmIntrinsics::_md5_implCompress:
615 case vmIntrinsics::_sha_implCompress:
616 case vmIntrinsics::_sha2_implCompress:
617 case vmIntrinsics::_sha5_implCompress:
618 case vmIntrinsics::_sha3_implCompress:
619 return inline_digestBase_implCompress(intrinsic_id());
620 case vmIntrinsics::_double_keccak:
621 return inline_double_keccak();
622
623 case vmIntrinsics::_digestBase_implCompressMB:
624 return inline_digestBase_implCompressMB(predicate);
625
626 case vmIntrinsics::_multiplyToLen:
627 return inline_multiplyToLen();
628
629 case vmIntrinsics::_squareToLen:
630 return inline_squareToLen();
631
632 case vmIntrinsics::_mulAdd:
633 return inline_mulAdd();
634
635 case vmIntrinsics::_montgomeryMultiply:
636 return inline_montgomeryMultiply();
637 case vmIntrinsics::_montgomerySquare:
638 return inline_montgomerySquare();
639
640 case vmIntrinsics::_bigIntegerRightShiftWorker:
641 return inline_bigIntegerShift(true);
642 case vmIntrinsics::_bigIntegerLeftShiftWorker:
643 return inline_bigIntegerShift(false);
644
645 case vmIntrinsics::_vectorizedMismatch:
646 return inline_vectorizedMismatch();
647
648 case vmIntrinsics::_ghash_processBlocks:
649 return inline_ghash_processBlocks();
650 case vmIntrinsics::_chacha20Block:
651 return inline_chacha20Block();
652 case vmIntrinsics::_kyberNtt:
653 return inline_kyberNtt();
654 case vmIntrinsics::_kyberInverseNtt:
655 return inline_kyberInverseNtt();
656 case vmIntrinsics::_kyberNttMult:
657 return inline_kyberNttMult();
658 case vmIntrinsics::_kyberAddPoly_2:
659 return inline_kyberAddPoly_2();
660 case vmIntrinsics::_kyberAddPoly_3:
661 return inline_kyberAddPoly_3();
662 case vmIntrinsics::_kyber12To16:
663 return inline_kyber12To16();
664 case vmIntrinsics::_kyberBarrettReduce:
665 return inline_kyberBarrettReduce();
666 case vmIntrinsics::_dilithiumAlmostNtt:
667 return inline_dilithiumAlmostNtt();
668 case vmIntrinsics::_dilithiumAlmostInverseNtt:
669 return inline_dilithiumAlmostInverseNtt();
670 case vmIntrinsics::_dilithiumNttMult:
671 return inline_dilithiumNttMult();
672 case vmIntrinsics::_dilithiumMontMulByConstant:
673 return inline_dilithiumMontMulByConstant();
674 case vmIntrinsics::_dilithiumDecomposePoly:
675 return inline_dilithiumDecomposePoly();
676 case vmIntrinsics::_base64_encodeBlock:
677 return inline_base64_encodeBlock();
678 case vmIntrinsics::_base64_decodeBlock:
679 return inline_base64_decodeBlock();
680 case vmIntrinsics::_poly1305_processBlocks:
681 return inline_poly1305_processBlocks();
682 case vmIntrinsics::_intpoly_montgomeryMult_P256:
683 return inline_intpoly_montgomeryMult_P256();
684 case vmIntrinsics::_intpoly_assign:
685 return inline_intpoly_assign();
686 case vmIntrinsics::_encodeISOArray:
687 case vmIntrinsics::_encodeByteISOArray:
688 return inline_encodeISOArray(false);
689 case vmIntrinsics::_encodeAsciiArray:
690 return inline_encodeISOArray(true);
691
692 case vmIntrinsics::_updateCRC32:
693 return inline_updateCRC32();
694 case vmIntrinsics::_updateBytesCRC32:
695 return inline_updateBytesCRC32();
696 case vmIntrinsics::_updateByteBufferCRC32:
697 return inline_updateByteBufferCRC32();
698
699 case vmIntrinsics::_updateBytesCRC32C:
700 return inline_updateBytesCRC32C();
701 case vmIntrinsics::_updateDirectByteBufferCRC32C:
702 return inline_updateDirectByteBufferCRC32C();
703
704 case vmIntrinsics::_updateBytesAdler32:
705 return inline_updateBytesAdler32();
706 case vmIntrinsics::_updateByteBufferAdler32:
707 return inline_updateByteBufferAdler32();
708
709 case vmIntrinsics::_profileBoolean:
710 return inline_profileBoolean();
711 case vmIntrinsics::_isCompileConstant:
712 return inline_isCompileConstant();
713
714 case vmIntrinsics::_countPositives:
715 return inline_countPositives();
716
717 case vmIntrinsics::_fmaD:
718 case vmIntrinsics::_fmaF:
719 return inline_fma(intrinsic_id());
720
721 case vmIntrinsics::_isDigit:
722 case vmIntrinsics::_isLowerCase:
723 case vmIntrinsics::_isUpperCase:
724 case vmIntrinsics::_isWhitespace:
725 return inline_character_compare(intrinsic_id());
726
727 case vmIntrinsics::_min:
728 case vmIntrinsics::_max:
729 case vmIntrinsics::_min_strict:
730 case vmIntrinsics::_max_strict:
731 case vmIntrinsics::_minL:
732 case vmIntrinsics::_maxL:
733 case vmIntrinsics::_minF:
734 case vmIntrinsics::_maxF:
735 case vmIntrinsics::_minD:
736 case vmIntrinsics::_maxD:
737 case vmIntrinsics::_minF_strict:
738 case vmIntrinsics::_maxF_strict:
739 case vmIntrinsics::_minD_strict:
740 case vmIntrinsics::_maxD_strict:
741 return inline_min_max(intrinsic_id());
742
743 case vmIntrinsics::_VectorUnaryOp:
744 return inline_vector_nary_operation(1);
745 case vmIntrinsics::_VectorBinaryOp:
746 return inline_vector_nary_operation(2);
747 case vmIntrinsics::_VectorUnaryLibOp:
748 return inline_vector_call(1);
749 case vmIntrinsics::_VectorBinaryLibOp:
750 return inline_vector_call(2);
751 case vmIntrinsics::_VectorTernaryOp:
752 return inline_vector_nary_operation(3);
753 case vmIntrinsics::_VectorFromBitsCoerced:
754 return inline_vector_frombits_coerced();
755 case vmIntrinsics::_VectorMaskOp:
756 return inline_vector_mask_operation();
757 case vmIntrinsics::_VectorLoadOp:
758 return inline_vector_mem_operation(/*is_store=*/false);
759 case vmIntrinsics::_VectorLoadMaskedOp:
760 return inline_vector_mem_masked_operation(/*is_store*/false);
761 case vmIntrinsics::_VectorStoreOp:
762 return inline_vector_mem_operation(/*is_store=*/true);
763 case vmIntrinsics::_VectorStoreMaskedOp:
764 return inline_vector_mem_masked_operation(/*is_store=*/true);
765 case vmIntrinsics::_VectorGatherOp:
766 return inline_vector_gather_scatter(/*is_scatter*/ false);
767 case vmIntrinsics::_VectorScatterOp:
768 return inline_vector_gather_scatter(/*is_scatter*/ true);
769 case vmIntrinsics::_VectorReductionCoerced:
770 return inline_vector_reduction();
771 case vmIntrinsics::_VectorTest:
772 return inline_vector_test();
773 case vmIntrinsics::_VectorBlend:
774 return inline_vector_blend();
775 case vmIntrinsics::_VectorRearrange:
776 return inline_vector_rearrange();
777 case vmIntrinsics::_VectorSelectFrom:
778 return inline_vector_select_from();
779 case vmIntrinsics::_VectorCompare:
780 return inline_vector_compare();
781 case vmIntrinsics::_VectorBroadcastInt:
782 return inline_vector_broadcast_int();
783 case vmIntrinsics::_VectorConvert:
784 return inline_vector_convert();
785 case vmIntrinsics::_VectorInsert:
786 return inline_vector_insert();
787 case vmIntrinsics::_VectorExtract:
788 return inline_vector_extract();
789 case vmIntrinsics::_VectorCompressExpand:
790 return inline_vector_compress_expand();
791 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
792 return inline_vector_select_from_two_vectors();
793 case vmIntrinsics::_IndexVector:
794 return inline_index_vector();
795 case vmIntrinsics::_IndexPartiallyInUpperRange:
796 return inline_index_partially_in_upper_range();
797
798 case vmIntrinsics::_getObjectSize:
799 return inline_getObjectSize();
800
801 case vmIntrinsics::_blackhole:
802 return inline_blackhole();
803
804 default:
805 // If you get here, it may be that someone has added a new intrinsic
806 // to the list in vmIntrinsics.hpp without implementing it here.
807 #ifndef PRODUCT
808 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
809 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
810 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
811 }
812 #endif
813 return false;
814 }
815 }
816
817 Node* LibraryCallKit::try_to_predicate(int predicate) {
818 if (!jvms()->has_method()) {
819 // Root JVMState has a null method.
820 assert(map()->memory()->Opcode() == Op_Parm, "");
821 // Insert the memory aliasing node
822 set_all_memory(reset_memory());
823 }
824 assert(merged_memory(), "");
825
826 switch (intrinsic_id()) {
827 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
828 return inline_cipherBlockChaining_AESCrypt_predicate(false);
829 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
830 return inline_cipherBlockChaining_AESCrypt_predicate(true);
831 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
832 return inline_electronicCodeBook_AESCrypt_predicate(false);
833 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
834 return inline_electronicCodeBook_AESCrypt_predicate(true);
835 case vmIntrinsics::_counterMode_AESCrypt:
836 return inline_counterMode_AESCrypt_predicate();
837 case vmIntrinsics::_digestBase_implCompressMB:
838 return inline_digestBase_implCompressMB_predicate(predicate);
839 case vmIntrinsics::_galoisCounterMode_AESCrypt:
840 return inline_galoisCounterMode_AESCrypt_predicate();
841
842 default:
843 // If you get here, it may be that someone has added a new intrinsic
844 // to the list in vmIntrinsics.hpp without implementing it here.
845 #ifndef PRODUCT
846 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
847 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
848 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
849 }
850 #endif
851 Node* slow_ctl = control();
852 set_control(top()); // No fast path intrinsic
853 return slow_ctl;
854 }
855 }
856
857 //------------------------------set_result-------------------------------
858 // Helper function for finishing intrinsics.
859 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
860 record_for_igvn(region);
861 set_control(_gvn.transform(region));
862 set_result( _gvn.transform(value));
863 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
864 }
865
866 //------------------------------generate_guard---------------------------
867 // Helper function for generating guarded fast-slow graph structures.
868 // The given 'test', if true, guards a slow path. If the test fails
869 // then a fast path can be taken. (We generally hope it fails.)
870 // In all cases, GraphKit::control() is updated to the fast path.
871 // The returned value represents the control for the slow path.
872 // The return value is never 'top'; it is either a valid control
873 // or null if it is obvious that the slow path can never be taken.
874 // Also, if region and the slow control are not null, the slow edge
875 // is appended to the region.
876 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
877 if (stopped()) {
878 // Already short circuited.
879 return nullptr;
880 }
881
882 // Build an if node and its projections.
883 // If test is true we take the slow path, which we assume is uncommon.
884 if (_gvn.type(test) == TypeInt::ZERO) {
885 // The slow branch is never taken. No need to build this guard.
886 return nullptr;
887 }
888
889 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
890
891 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
892 if (if_slow == top()) {
893 // The slow branch is never taken. No need to build this guard.
894 return nullptr;
895 }
896
897 if (region != nullptr)
898 region->add_req(if_slow);
899
900 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
901 set_control(if_fast);
902
903 return if_slow;
904 }
905
906 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
907 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
908 }
909 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
910 return generate_guard(test, region, PROB_FAIR);
911 }
912
913 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
914 Node* *pos_index) {
915 if (stopped())
916 return nullptr; // already stopped
917 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
918 return nullptr; // index is already adequately typed
919 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
920 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
921 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
922 if (is_neg != nullptr && pos_index != nullptr) {
923 // Emulate effect of Parse::adjust_map_after_if.
924 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
925 (*pos_index) = _gvn.transform(ccast);
926 }
927 return is_neg;
928 }
929
930 // Make sure that 'position' is a valid limit index, in [0..length].
931 // There are two equivalent plans for checking this:
932 // A. (offset + copyLength) unsigned<= arrayLength
933 // B. offset <= (arrayLength - copyLength)
934 // We require that all of the values above, except for the sum and
935 // difference, are already known to be non-negative.
936 // Plan A is robust in the face of overflow, if offset and copyLength
937 // are both hugely positive.
938 //
939 // Plan B is less direct and intuitive, but it does not overflow at
940 // all, since the difference of two non-negatives is always
941 // representable. Whenever Java methods must perform the equivalent
942 // check they generally use Plan B instead of Plan A.
943 // For the moment we use Plan A.
944 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
945 Node* subseq_length,
946 Node* array_length,
947 RegionNode* region) {
948 if (stopped())
949 return nullptr; // already stopped
950 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
951 if (zero_offset && subseq_length->eqv_uncast(array_length))
952 return nullptr; // common case of whole-array copy
953 Node* last = subseq_length;
954 if (!zero_offset) // last += offset
955 last = _gvn.transform(new AddINode(last, offset));
956 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
957 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
958 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
959 return is_over;
960 }
961
962 // Emit range checks for the given String.value byte array
963 void LibraryCallKit::generate_string_range_check(Node* array,
964 Node* offset,
965 Node* count,
966 bool char_count,
967 bool halt_on_oob) {
968 if (stopped()) {
969 return; // already stopped
970 }
971 RegionNode* bailout = new RegionNode(1);
972 record_for_igvn(bailout);
973 if (char_count) {
974 // Convert char count to byte count
975 count = _gvn.transform(new LShiftINode(count, intcon(1)));
976 }
977
978 // Offset and count must not be negative
979 generate_negative_guard(offset, bailout);
980 generate_negative_guard(count, bailout);
981 // Offset + count must not exceed length of array
982 generate_limit_guard(offset, count, load_array_length(array), bailout);
983
984 if (bailout->req() > 1) {
985 if (halt_on_oob) {
986 bailout = _gvn.transform(bailout)->as_Region();
987 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
988 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
989 C->root()->add_req(halt);
990 } else {
991 PreserveJVMState pjvms(this);
992 set_control(_gvn.transform(bailout));
993 uncommon_trap(Deoptimization::Reason_intrinsic,
994 Deoptimization::Action_maybe_recompile);
995 }
996 }
997 }
998
999 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1000 bool is_immutable) {
1001 ciKlass* thread_klass = env()->Thread_klass();
1002 const Type* thread_type
1003 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1004
1005 Node* thread = _gvn.transform(new ThreadLocalNode());
1006 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
1007 tls_output = thread;
1008
1009 Node* thread_obj_handle
1010 = (is_immutable
1011 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1012 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1013 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1014 thread_obj_handle = _gvn.transform(thread_obj_handle);
1015
1016 DecoratorSet decorators = IN_NATIVE;
1017 if (is_immutable) {
1018 decorators |= C2_IMMUTABLE_MEMORY;
1019 }
1020 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1021 }
1022
1023 //--------------------------generate_current_thread--------------------
1024 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1025 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1026 /*is_immutable*/false);
1027 }
1028
1029 //--------------------------generate_virtual_thread--------------------
1030 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1031 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1032 !C->method()->changes_current_thread());
1033 }
1034
1035 //------------------------------make_string_method_node------------------------
1036 // Helper method for String intrinsic functions. This version is called with
1037 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1038 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1039 // containing the lengths of str1 and str2.
1040 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1041 Node* result = nullptr;
1042 switch (opcode) {
1043 case Op_StrIndexOf:
1044 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1045 str1_start, cnt1, str2_start, cnt2, ae);
1046 break;
1047 case Op_StrComp:
1048 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1049 str1_start, cnt1, str2_start, cnt2, ae);
1050 break;
1051 case Op_StrEquals:
1052 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1053 // Use the constant length if there is one because optimized match rule may exist.
1054 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1055 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1056 break;
1057 default:
1058 ShouldNotReachHere();
1059 return nullptr;
1060 }
1061
1062 // All these intrinsics have checks.
1063 C->set_has_split_ifs(true); // Has chance for split-if optimization
1064 clear_upper_avx();
1065
1066 return _gvn.transform(result);
1067 }
1068
1069 //------------------------------inline_string_compareTo------------------------
1070 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1071 Node* arg1 = argument(0);
1072 Node* arg2 = argument(1);
1073
1074 arg1 = must_be_not_null(arg1, true);
1075 arg2 = must_be_not_null(arg2, true);
1076
1077 // Get start addr and length of first argument
1078 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1079 Node* arg1_cnt = load_array_length(arg1);
1080
1081 // Get start addr and length of second argument
1082 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1083 Node* arg2_cnt = load_array_length(arg2);
1084
1085 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1086 set_result(result);
1087 return true;
1088 }
1089
1090 //------------------------------inline_string_equals------------------------
1091 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1092 Node* arg1 = argument(0);
1093 Node* arg2 = argument(1);
1094
1095 // paths (plus control) merge
1096 RegionNode* region = new RegionNode(3);
1097 Node* phi = new PhiNode(region, TypeInt::BOOL);
1098
1099 if (!stopped()) {
1100
1101 arg1 = must_be_not_null(arg1, true);
1102 arg2 = must_be_not_null(arg2, true);
1103
1104 // Get start addr and length of first argument
1105 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1106 Node* arg1_cnt = load_array_length(arg1);
1107
1108 // Get start addr and length of second argument
1109 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1110 Node* arg2_cnt = load_array_length(arg2);
1111
1112 // Check for arg1_cnt != arg2_cnt
1113 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1114 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1115 Node* if_ne = generate_slow_guard(bol, nullptr);
1116 if (if_ne != nullptr) {
1117 phi->init_req(2, intcon(0));
1118 region->init_req(2, if_ne);
1119 }
1120
1121 // Check for count == 0 is done by assembler code for StrEquals.
1122
1123 if (!stopped()) {
1124 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1125 phi->init_req(1, equals);
1126 region->init_req(1, control());
1127 }
1128 }
1129
1130 // post merge
1131 set_control(_gvn.transform(region));
1132 record_for_igvn(region);
1133
1134 set_result(_gvn.transform(phi));
1135 return true;
1136 }
1137
1138 //------------------------------inline_array_equals----------------------------
1139 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1140 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1141 Node* arg1 = argument(0);
1142 Node* arg2 = argument(1);
1143
1144 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1145 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1146 clear_upper_avx();
1147
1148 return true;
1149 }
1150
1151
1152 //------------------------------inline_countPositives------------------------------
1153 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1154 bool LibraryCallKit::inline_countPositives() {
1155 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1156 return false;
1157 }
1158
1159 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1160 // no receiver since it is static method
1161 Node* ba = argument(0);
1162 Node* offset = argument(1);
1163 Node* len = argument(2);
1164
1165 if (VerifyIntrinsicChecks) {
1166 ba = must_be_not_null(ba, true);
1167 generate_string_range_check(ba, offset, len, false, true);
1168 if (stopped()) {
1169 return true;
1170 }
1171 }
1172
1173 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1174 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1175 set_result(_gvn.transform(result));
1176 clear_upper_avx();
1177 return true;
1178 }
1179
1180 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1181 Node* index = argument(0);
1182 Node* length = bt == T_INT ? argument(1) : argument(2);
1183 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1184 return false;
1185 }
1186
1187 // check that length is positive
1188 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1189 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1190
1191 {
1192 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1193 uncommon_trap(Deoptimization::Reason_intrinsic,
1194 Deoptimization::Action_make_not_entrant);
1195 }
1196
1197 if (stopped()) {
1198 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1199 return true;
1200 }
1201
1202 // length is now known positive, add a cast node to make this explicit
1203 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1204 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1205 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1206 ConstraintCastNode::RegularDependency, bt);
1207 casted_length = _gvn.transform(casted_length);
1208 replace_in_map(length, casted_length);
1209 length = casted_length;
1210
1211 // Use an unsigned comparison for the range check itself
1212 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1213 BoolTest::mask btest = BoolTest::lt;
1214 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1215 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1216 _gvn.set_type(rc, rc->Value(&_gvn));
1217 if (!rc_bool->is_Con()) {
1218 record_for_igvn(rc);
1219 }
1220 set_control(_gvn.transform(new IfTrueNode(rc)));
1221 {
1222 PreserveJVMState pjvms(this);
1223 set_control(_gvn.transform(new IfFalseNode(rc)));
1224 uncommon_trap(Deoptimization::Reason_range_check,
1225 Deoptimization::Action_make_not_entrant);
1226 }
1227
1228 if (stopped()) {
1229 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1230 return true;
1231 }
1232
1233 // index is now known to be >= 0 and < length, cast it
1234 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1235 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1236 ConstraintCastNode::RegularDependency, bt);
1237 result = _gvn.transform(result);
1238 set_result(result);
1239 replace_in_map(index, result);
1240 return true;
1241 }
1242
1243 //------------------------------inline_string_indexOf------------------------
1244 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1245 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1246 return false;
1247 }
1248 Node* src = argument(0);
1249 Node* tgt = argument(1);
1250
1251 // Make the merge point
1252 RegionNode* result_rgn = new RegionNode(4);
1253 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1254
1255 src = must_be_not_null(src, true);
1256 tgt = must_be_not_null(tgt, true);
1257
1258 // Get start addr and length of source string
1259 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1260 Node* src_count = load_array_length(src);
1261
1262 // Get start addr and length of substring
1263 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1264 Node* tgt_count = load_array_length(tgt);
1265
1266 Node* result = nullptr;
1267 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1268
1269 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1270 // Divide src size by 2 if String is UTF16 encoded
1271 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1272 }
1273 if (ae == StrIntrinsicNode::UU) {
1274 // Divide substring size by 2 if String is UTF16 encoded
1275 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1276 }
1277
1278 if (call_opt_stub) {
1279 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1280 StubRoutines::_string_indexof_array[ae],
1281 "stringIndexOf", TypePtr::BOTTOM, src_start,
1282 src_count, tgt_start, tgt_count);
1283 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1284 } else {
1285 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1286 result_rgn, result_phi, ae);
1287 }
1288 if (result != nullptr) {
1289 result_phi->init_req(3, result);
1290 result_rgn->init_req(3, control());
1291 }
1292 set_control(_gvn.transform(result_rgn));
1293 record_for_igvn(result_rgn);
1294 set_result(_gvn.transform(result_phi));
1295
1296 return true;
1297 }
1298
1299 //-----------------------------inline_string_indexOfI-----------------------
1300 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1301 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1302 return false;
1303 }
1304 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1305 return false;
1306 }
1307
1308 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1309 Node* src = argument(0); // byte[]
1310 Node* src_count = argument(1); // char count
1311 Node* tgt = argument(2); // byte[]
1312 Node* tgt_count = argument(3); // char count
1313 Node* from_index = argument(4); // char index
1314
1315 src = must_be_not_null(src, true);
1316 tgt = must_be_not_null(tgt, true);
1317
1318 // Multiply byte array index by 2 if String is UTF16 encoded
1319 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1320 src_count = _gvn.transform(new SubINode(src_count, from_index));
1321 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1322 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1323
1324 // Range checks
1325 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1326 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1327 if (stopped()) {
1328 return true;
1329 }
1330
1331 RegionNode* region = new RegionNode(5);
1332 Node* phi = new PhiNode(region, TypeInt::INT);
1333 Node* result = nullptr;
1334
1335 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1336
1337 if (call_opt_stub) {
1338 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1339 StubRoutines::_string_indexof_array[ae],
1340 "stringIndexOf", TypePtr::BOTTOM, src_start,
1341 src_count, tgt_start, tgt_count);
1342 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1343 } else {
1344 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1345 region, phi, ae);
1346 }
1347 if (result != nullptr) {
1348 // The result is index relative to from_index if substring was found, -1 otherwise.
1349 // Generate code which will fold into cmove.
1350 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1351 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1352
1353 Node* if_lt = generate_slow_guard(bol, nullptr);
1354 if (if_lt != nullptr) {
1355 // result == -1
1356 phi->init_req(3, result);
1357 region->init_req(3, if_lt);
1358 }
1359 if (!stopped()) {
1360 result = _gvn.transform(new AddINode(result, from_index));
1361 phi->init_req(4, result);
1362 region->init_req(4, control());
1363 }
1364 }
1365
1366 set_control(_gvn.transform(region));
1367 record_for_igvn(region);
1368 set_result(_gvn.transform(phi));
1369 clear_upper_avx();
1370
1371 return true;
1372 }
1373
1374 // Create StrIndexOfNode with fast path checks
1375 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1376 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1377 // Check for substr count > string count
1378 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1379 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1380 Node* if_gt = generate_slow_guard(bol, nullptr);
1381 if (if_gt != nullptr) {
1382 phi->init_req(1, intcon(-1));
1383 region->init_req(1, if_gt);
1384 }
1385 if (!stopped()) {
1386 // Check for substr count == 0
1387 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1388 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1389 Node* if_zero = generate_slow_guard(bol, nullptr);
1390 if (if_zero != nullptr) {
1391 phi->init_req(2, intcon(0));
1392 region->init_req(2, if_zero);
1393 }
1394 }
1395 if (!stopped()) {
1396 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1397 }
1398 return nullptr;
1399 }
1400
1401 //-----------------------------inline_string_indexOfChar-----------------------
1402 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1403 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1404 return false;
1405 }
1406 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1407 return false;
1408 }
1409 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1410 Node* src = argument(0); // byte[]
1411 Node* int_ch = argument(1);
1412 Node* from_index = argument(2);
1413 Node* max = argument(3);
1414
1415 src = must_be_not_null(src, true);
1416
1417 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1418 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1419 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1420
1421 // Range checks
1422 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1423
1424 // Check for int_ch >= 0
1425 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1426 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1427 {
1428 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1429 uncommon_trap(Deoptimization::Reason_intrinsic,
1430 Deoptimization::Action_maybe_recompile);
1431 }
1432 if (stopped()) {
1433 return true;
1434 }
1435
1436 RegionNode* region = new RegionNode(3);
1437 Node* phi = new PhiNode(region, TypeInt::INT);
1438
1439 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1440 C->set_has_split_ifs(true); // Has chance for split-if optimization
1441 _gvn.transform(result);
1442
1443 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1444 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1445
1446 Node* if_lt = generate_slow_guard(bol, nullptr);
1447 if (if_lt != nullptr) {
1448 // result == -1
1449 phi->init_req(2, result);
1450 region->init_req(2, if_lt);
1451 }
1452 if (!stopped()) {
1453 result = _gvn.transform(new AddINode(result, from_index));
1454 phi->init_req(1, result);
1455 region->init_req(1, control());
1456 }
1457 set_control(_gvn.transform(region));
1458 record_for_igvn(region);
1459 set_result(_gvn.transform(phi));
1460 clear_upper_avx();
1461
1462 return true;
1463 }
1464 //---------------------------inline_string_copy---------------------
1465 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1466 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1467 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1468 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1469 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1470 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1471 bool LibraryCallKit::inline_string_copy(bool compress) {
1472 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1473 return false;
1474 }
1475 int nargs = 5; // 2 oops, 3 ints
1476 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1477
1478 Node* src = argument(0);
1479 Node* src_offset = argument(1);
1480 Node* dst = argument(2);
1481 Node* dst_offset = argument(3);
1482 Node* length = argument(4);
1483
1484 // Check for allocation before we add nodes that would confuse
1485 // tightly_coupled_allocation()
1486 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1487
1488 // Figure out the size and type of the elements we will be copying.
1489 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1490 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1491 if (src_type == nullptr || dst_type == nullptr) {
1492 return false;
1493 }
1494 BasicType src_elem = src_type->elem()->array_element_basic_type();
1495 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1496 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1497 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1498 "Unsupported array types for inline_string_copy");
1499
1500 src = must_be_not_null(src, true);
1501 dst = must_be_not_null(dst, true);
1502
1503 // Convert char[] offsets to byte[] offsets
1504 bool convert_src = (compress && src_elem == T_BYTE);
1505 bool convert_dst = (!compress && dst_elem == T_BYTE);
1506 if (convert_src) {
1507 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1508 } else if (convert_dst) {
1509 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1510 }
1511
1512 // Range checks
1513 generate_string_range_check(src, src_offset, length, convert_src);
1514 generate_string_range_check(dst, dst_offset, length, convert_dst);
1515 if (stopped()) {
1516 return true;
1517 }
1518
1519 Node* src_start = array_element_address(src, src_offset, src_elem);
1520 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1521 // 'src_start' points to src array + scaled offset
1522 // 'dst_start' points to dst array + scaled offset
1523 Node* count = nullptr;
1524 if (compress) {
1525 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1526 } else {
1527 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1528 }
1529
1530 if (alloc != nullptr) {
1531 if (alloc->maybe_set_complete(&_gvn)) {
1532 // "You break it, you buy it."
1533 InitializeNode* init = alloc->initialization();
1534 assert(init->is_complete(), "we just did this");
1535 init->set_complete_with_arraycopy();
1536 assert(dst->is_CheckCastPP(), "sanity");
1537 assert(dst->in(0)->in(0) == init, "dest pinned");
1538 }
1539 // Do not let stores that initialize this object be reordered with
1540 // a subsequent store that would make this object accessible by
1541 // other threads.
1542 // Record what AllocateNode this StoreStore protects so that
1543 // escape analysis can go from the MemBarStoreStoreNode to the
1544 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1545 // based on the escape status of the AllocateNode.
1546 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1547 }
1548 if (compress) {
1549 set_result(_gvn.transform(count));
1550 }
1551 clear_upper_avx();
1552
1553 return true;
1554 }
1555
1556 #ifdef _LP64
1557 #define XTOP ,top() /*additional argument*/
1558 #else //_LP64
1559 #define XTOP /*no additional argument*/
1560 #endif //_LP64
1561
1562 //------------------------inline_string_toBytesU--------------------------
1563 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1564 bool LibraryCallKit::inline_string_toBytesU() {
1565 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1566 return false;
1567 }
1568 // Get the arguments.
1569 Node* value = argument(0);
1570 Node* offset = argument(1);
1571 Node* length = argument(2);
1572
1573 Node* newcopy = nullptr;
1574
1575 // Set the original stack and the reexecute bit for the interpreter to reexecute
1576 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1577 { PreserveReexecuteState preexecs(this);
1578 jvms()->set_should_reexecute(true);
1579
1580 // Check if a null path was taken unconditionally.
1581 value = null_check(value);
1582
1583 RegionNode* bailout = new RegionNode(1);
1584 record_for_igvn(bailout);
1585
1586 // Range checks
1587 generate_negative_guard(offset, bailout);
1588 generate_negative_guard(length, bailout);
1589 generate_limit_guard(offset, length, load_array_length(value), bailout);
1590 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1591 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1592
1593 if (bailout->req() > 1) {
1594 PreserveJVMState pjvms(this);
1595 set_control(_gvn.transform(bailout));
1596 uncommon_trap(Deoptimization::Reason_intrinsic,
1597 Deoptimization::Action_maybe_recompile);
1598 }
1599 if (stopped()) {
1600 return true;
1601 }
1602
1603 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1604 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1605 newcopy = new_array(klass_node, size, 0); // no arguments to push
1606 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1607 guarantee(alloc != nullptr, "created above");
1608
1609 // Calculate starting addresses.
1610 Node* src_start = array_element_address(value, offset, T_CHAR);
1611 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1612
1613 // Check if dst array address is aligned to HeapWordSize
1614 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1615 // If true, then check if src array address is aligned to HeapWordSize
1616 if (aligned) {
1617 const TypeInt* toffset = gvn().type(offset)->is_int();
1618 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1619 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1620 }
1621
1622 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1623 const char* copyfunc_name = "arraycopy";
1624 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1625 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1626 OptoRuntime::fast_arraycopy_Type(),
1627 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1628 src_start, dst_start, ConvI2X(length) XTOP);
1629 // Do not let reads from the cloned object float above the arraycopy.
1630 if (alloc->maybe_set_complete(&_gvn)) {
1631 // "You break it, you buy it."
1632 InitializeNode* init = alloc->initialization();
1633 assert(init->is_complete(), "we just did this");
1634 init->set_complete_with_arraycopy();
1635 assert(newcopy->is_CheckCastPP(), "sanity");
1636 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1637 }
1638 // Do not let stores that initialize this object be reordered with
1639 // a subsequent store that would make this object accessible by
1640 // other threads.
1641 // Record what AllocateNode this StoreStore protects so that
1642 // escape analysis can go from the MemBarStoreStoreNode to the
1643 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1644 // based on the escape status of the AllocateNode.
1645 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1646 } // original reexecute is set back here
1647
1648 C->set_has_split_ifs(true); // Has chance for split-if optimization
1649 if (!stopped()) {
1650 set_result(newcopy);
1651 }
1652 clear_upper_avx();
1653
1654 return true;
1655 }
1656
1657 //------------------------inline_string_getCharsU--------------------------
1658 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1659 bool LibraryCallKit::inline_string_getCharsU() {
1660 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1661 return false;
1662 }
1663
1664 // Get the arguments.
1665 Node* src = argument(0);
1666 Node* src_begin = argument(1);
1667 Node* src_end = argument(2); // exclusive offset (i < src_end)
1668 Node* dst = argument(3);
1669 Node* dst_begin = argument(4);
1670
1671 // Check for allocation before we add nodes that would confuse
1672 // tightly_coupled_allocation()
1673 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1674
1675 // Check if a null path was taken unconditionally.
1676 src = null_check(src);
1677 dst = null_check(dst);
1678 if (stopped()) {
1679 return true;
1680 }
1681
1682 // Get length and convert char[] offset to byte[] offset
1683 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1684 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1685
1686 // Range checks
1687 generate_string_range_check(src, src_begin, length, true);
1688 generate_string_range_check(dst, dst_begin, length, false);
1689 if (stopped()) {
1690 return true;
1691 }
1692
1693 if (!stopped()) {
1694 // Calculate starting addresses.
1695 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1696 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1697
1698 // Check if array addresses are aligned to HeapWordSize
1699 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1700 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1701 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1702 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1703
1704 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1705 const char* copyfunc_name = "arraycopy";
1706 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1707 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1708 OptoRuntime::fast_arraycopy_Type(),
1709 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1710 src_start, dst_start, ConvI2X(length) XTOP);
1711 // Do not let reads from the cloned object float above the arraycopy.
1712 if (alloc != nullptr) {
1713 if (alloc->maybe_set_complete(&_gvn)) {
1714 // "You break it, you buy it."
1715 InitializeNode* init = alloc->initialization();
1716 assert(init->is_complete(), "we just did this");
1717 init->set_complete_with_arraycopy();
1718 assert(dst->is_CheckCastPP(), "sanity");
1719 assert(dst->in(0)->in(0) == init, "dest pinned");
1720 }
1721 // Do not let stores that initialize this object be reordered with
1722 // a subsequent store that would make this object accessible by
1723 // other threads.
1724 // Record what AllocateNode this StoreStore protects so that
1725 // escape analysis can go from the MemBarStoreStoreNode to the
1726 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1727 // based on the escape status of the AllocateNode.
1728 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1729 } else {
1730 insert_mem_bar(Op_MemBarCPUOrder);
1731 }
1732 }
1733
1734 C->set_has_split_ifs(true); // Has chance for split-if optimization
1735 return true;
1736 }
1737
1738 //----------------------inline_string_char_access----------------------------
1739 // Store/Load char to/from byte[] array.
1740 // static void StringUTF16.putChar(byte[] val, int index, int c)
1741 // static char StringUTF16.getChar(byte[] val, int index)
1742 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1743 Node* value = argument(0);
1744 Node* index = argument(1);
1745 Node* ch = is_store ? argument(2) : nullptr;
1746
1747 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1748 // correctly requires matched array shapes.
1749 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1750 "sanity: byte[] and char[] bases agree");
1751 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1752 "sanity: byte[] and char[] scales agree");
1753
1754 // Bail when getChar over constants is requested: constant folding would
1755 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1756 // Java method would constant fold nicely instead.
1757 if (!is_store && value->is_Con() && index->is_Con()) {
1758 return false;
1759 }
1760
1761 // Save state and restore on bailout
1762 SavedState old_state(this);
1763
1764 value = must_be_not_null(value, true);
1765
1766 Node* adr = array_element_address(value, index, T_CHAR);
1767 if (adr->is_top()) {
1768 return false;
1769 }
1770 old_state.discard();
1771 if (is_store) {
1772 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1773 } else {
1774 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);
1775 set_result(ch);
1776 }
1777 return true;
1778 }
1779
1780
1781 //------------------------------inline_math-----------------------------------
1782 // public static double Math.abs(double)
1783 // public static double Math.sqrt(double)
1784 // public static double Math.log(double)
1785 // public static double Math.log10(double)
1786 // public static double Math.round(double)
1787 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1788 Node* arg = argument(0);
1789 Node* n = nullptr;
1790 switch (id) {
1791 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1792 case vmIntrinsics::_dsqrt:
1793 case vmIntrinsics::_dsqrt_strict:
1794 n = new SqrtDNode(C, control(), arg); break;
1795 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1796 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1797 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1798 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1799 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1800 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1801 default: fatal_unexpected_iid(id); break;
1802 }
1803 set_result(_gvn.transform(n));
1804 return true;
1805 }
1806
1807 //------------------------------inline_math-----------------------------------
1808 // public static float Math.abs(float)
1809 // public static int Math.abs(int)
1810 // public static long Math.abs(long)
1811 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1812 Node* arg = argument(0);
1813 Node* n = nullptr;
1814 switch (id) {
1815 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1816 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1817 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1818 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1819 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1820 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1821 default: fatal_unexpected_iid(id); break;
1822 }
1823 set_result(_gvn.transform(n));
1824 return true;
1825 }
1826
1827 //------------------------------runtime_math-----------------------------
1828 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1829 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1830 "must be (DD)D or (D)D type");
1831
1832 // Inputs
1833 Node* a = argument(0);
1834 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1835
1836 const TypePtr* no_memory_effects = nullptr;
1837 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1838 no_memory_effects,
1839 a, top(), b, b ? top() : nullptr);
1840 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1841 #ifdef ASSERT
1842 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1843 assert(value_top == top(), "second value must be top");
1844 #endif
1845
1846 set_result(value);
1847 return true;
1848 }
1849
1850 //------------------------------inline_math_pow-----------------------------
1851 bool LibraryCallKit::inline_math_pow() {
1852 Node* exp = argument(2);
1853 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1854 if (d != nullptr) {
1855 if (d->getd() == 2.0) {
1856 // Special case: pow(x, 2.0) => x * x
1857 Node* base = argument(0);
1858 set_result(_gvn.transform(new MulDNode(base, base)));
1859 return true;
1860 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1861 // Special case: pow(x, 0.5) => sqrt(x)
1862 Node* base = argument(0);
1863 Node* zero = _gvn.zerocon(T_DOUBLE);
1864
1865 RegionNode* region = new RegionNode(3);
1866 Node* phi = new PhiNode(region, Type::DOUBLE);
1867
1868 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1869 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1870 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1871 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1872 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1873
1874 Node* if_pow = generate_slow_guard(test, nullptr);
1875 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1876 phi->init_req(1, value_sqrt);
1877 region->init_req(1, control());
1878
1879 if (if_pow != nullptr) {
1880 set_control(if_pow);
1881 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1882 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1883 const TypePtr* no_memory_effects = nullptr;
1884 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1885 no_memory_effects, base, top(), exp, top());
1886 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1887 #ifdef ASSERT
1888 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1889 assert(value_top == top(), "second value must be top");
1890 #endif
1891 phi->init_req(2, value_pow);
1892 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1893 }
1894
1895 C->set_has_split_ifs(true); // Has chance for split-if optimization
1896 set_control(_gvn.transform(region));
1897 record_for_igvn(region);
1898 set_result(_gvn.transform(phi));
1899
1900 return true;
1901 }
1902 }
1903
1904 return StubRoutines::dpow() != nullptr ?
1905 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1906 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1907 }
1908
1909 //------------------------------inline_math_native-----------------------------
1910 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1911 switch (id) {
1912 case vmIntrinsics::_dsin:
1913 return StubRoutines::dsin() != nullptr ?
1914 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1915 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1916 case vmIntrinsics::_dcos:
1917 return StubRoutines::dcos() != nullptr ?
1918 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1919 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1920 case vmIntrinsics::_dtan:
1921 return StubRoutines::dtan() != nullptr ?
1922 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1923 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1924 case vmIntrinsics::_dsinh:
1925 return StubRoutines::dsinh() != nullptr ?
1926 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1927 case vmIntrinsics::_dtanh:
1928 return StubRoutines::dtanh() != nullptr ?
1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1930 case vmIntrinsics::_dcbrt:
1931 return StubRoutines::dcbrt() != nullptr ?
1932 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1933 case vmIntrinsics::_dexp:
1934 return StubRoutines::dexp() != nullptr ?
1935 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1936 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1937 case vmIntrinsics::_dlog:
1938 return StubRoutines::dlog() != nullptr ?
1939 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1940 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1941 case vmIntrinsics::_dlog10:
1942 return StubRoutines::dlog10() != nullptr ?
1943 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1944 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1945
1946 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1947 case vmIntrinsics::_ceil:
1948 case vmIntrinsics::_floor:
1949 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1950
1951 case vmIntrinsics::_dsqrt:
1952 case vmIntrinsics::_dsqrt_strict:
1953 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1954 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1955 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1956 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1957 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1958
1959 case vmIntrinsics::_dpow: return inline_math_pow();
1960 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1961 case vmIntrinsics::_fcopySign: return inline_math(id);
1962 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1963 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1964 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1965
1966 // These intrinsics are not yet correctly implemented
1967 case vmIntrinsics::_datan2:
1968 return false;
1969
1970 default:
1971 fatal_unexpected_iid(id);
1972 return false;
1973 }
1974 }
1975
1976 //----------------------------inline_notify-----------------------------------*
1977 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1978 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1979 address func;
1980 if (id == vmIntrinsics::_notify) {
1981 func = OptoRuntime::monitor_notify_Java();
1982 } else {
1983 func = OptoRuntime::monitor_notifyAll_Java();
1984 }
1985 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1986 make_slow_call_ex(call, env()->Throwable_klass(), false);
1987 return true;
1988 }
1989
1990
1991 //----------------------------inline_min_max-----------------------------------
1992 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1993 Node* a = nullptr;
1994 Node* b = nullptr;
1995 Node* n = nullptr;
1996 switch (id) {
1997 case vmIntrinsics::_min:
1998 case vmIntrinsics::_max:
1999 case vmIntrinsics::_minF:
2000 case vmIntrinsics::_maxF:
2001 case vmIntrinsics::_minF_strict:
2002 case vmIntrinsics::_maxF_strict:
2003 case vmIntrinsics::_min_strict:
2004 case vmIntrinsics::_max_strict:
2005 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
2006 a = argument(0);
2007 b = argument(1);
2008 break;
2009 case vmIntrinsics::_minD:
2010 case vmIntrinsics::_maxD:
2011 case vmIntrinsics::_minD_strict:
2012 case vmIntrinsics::_maxD_strict:
2013 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2014 a = argument(0);
2015 b = argument(2);
2016 break;
2017 case vmIntrinsics::_minL:
2018 case vmIntrinsics::_maxL:
2019 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2020 a = argument(0);
2021 b = argument(2);
2022 break;
2023 default:
2024 fatal_unexpected_iid(id);
2025 break;
2026 }
2027
2028 switch (id) {
2029 case vmIntrinsics::_min:
2030 case vmIntrinsics::_min_strict:
2031 n = new MinINode(a, b);
2032 break;
2033 case vmIntrinsics::_max:
2034 case vmIntrinsics::_max_strict:
2035 n = new MaxINode(a, b);
2036 break;
2037 case vmIntrinsics::_minF:
2038 case vmIntrinsics::_minF_strict:
2039 n = new MinFNode(a, b);
2040 break;
2041 case vmIntrinsics::_maxF:
2042 case vmIntrinsics::_maxF_strict:
2043 n = new MaxFNode(a, b);
2044 break;
2045 case vmIntrinsics::_minD:
2046 case vmIntrinsics::_minD_strict:
2047 n = new MinDNode(a, b);
2048 break;
2049 case vmIntrinsics::_maxD:
2050 case vmIntrinsics::_maxD_strict:
2051 n = new MaxDNode(a, b);
2052 break;
2053 case vmIntrinsics::_minL:
2054 n = new MinLNode(_gvn.C, a, b);
2055 break;
2056 case vmIntrinsics::_maxL:
2057 n = new MaxLNode(_gvn.C, a, b);
2058 break;
2059 default:
2060 fatal_unexpected_iid(id);
2061 break;
2062 }
2063
2064 set_result(_gvn.transform(n));
2065 return true;
2066 }
2067
2068 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2069 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2070 env()->ArithmeticException_instance())) {
2071 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2072 // so let's bail out intrinsic rather than risking deopting again.
2073 return false;
2074 }
2075
2076 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2077 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2078 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2079 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2080
2081 {
2082 PreserveJVMState pjvms(this);
2083 PreserveReexecuteState preexecs(this);
2084 jvms()->set_should_reexecute(true);
2085
2086 set_control(slow_path);
2087 set_i_o(i_o());
2088
2089 builtin_throw(Deoptimization::Reason_intrinsic,
2090 env()->ArithmeticException_instance(),
2091 /*allow_too_many_traps*/ false);
2092 }
2093
2094 set_control(fast_path);
2095 set_result(math);
2096 return true;
2097 }
2098
2099 template <typename OverflowOp>
2100 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2101 typedef typename OverflowOp::MathOp MathOp;
2102
2103 MathOp* mathOp = new MathOp(arg1, arg2);
2104 Node* operation = _gvn.transform( mathOp );
2105 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2106 return inline_math_mathExact(operation, ofcheck);
2107 }
2108
2109 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2110 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2111 }
2112
2113 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2114 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2115 }
2116
2117 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2118 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2119 }
2120
2121 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2122 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2123 }
2124
2125 bool LibraryCallKit::inline_math_negateExactI() {
2126 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2127 }
2128
2129 bool LibraryCallKit::inline_math_negateExactL() {
2130 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2131 }
2132
2133 bool LibraryCallKit::inline_math_multiplyExactI() {
2134 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2135 }
2136
2137 bool LibraryCallKit::inline_math_multiplyExactL() {
2138 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2139 }
2140
2141 bool LibraryCallKit::inline_math_multiplyHigh() {
2142 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2143 return true;
2144 }
2145
2146 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2147 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2148 return true;
2149 }
2150
2151 inline int
2152 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2153 const TypePtr* base_type = TypePtr::NULL_PTR;
2154 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2155 if (base_type == nullptr) {
2156 // Unknown type.
2157 return Type::AnyPtr;
2158 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2159 // Since this is a null+long form, we have to switch to a rawptr.
2160 base = _gvn.transform(new CastX2PNode(offset));
2161 offset = MakeConX(0);
2162 return Type::RawPtr;
2163 } else if (base_type->base() == Type::RawPtr) {
2164 return Type::RawPtr;
2165 } else if (base_type->isa_oopptr()) {
2166 // Base is never null => always a heap address.
2167 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2168 return Type::OopPtr;
2169 }
2170 // Offset is small => always a heap address.
2171 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2172 if (offset_type != nullptr &&
2173 base_type->offset() == 0 && // (should always be?)
2174 offset_type->_lo >= 0 &&
2175 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2176 return Type::OopPtr;
2177 } else if (type == T_OBJECT) {
2178 // off heap access to an oop doesn't make any sense. Has to be on
2179 // heap.
2180 return Type::OopPtr;
2181 }
2182 // Otherwise, it might either be oop+off or null+addr.
2183 return Type::AnyPtr;
2184 } else {
2185 // No information:
2186 return Type::AnyPtr;
2187 }
2188 }
2189
2190 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2191 Node* uncasted_base = base;
2192 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2193 if (kind == Type::RawPtr) {
2194 return basic_plus_adr(top(), uncasted_base, offset);
2195 } else if (kind == Type::AnyPtr) {
2196 assert(base == uncasted_base, "unexpected base change");
2197 if (can_cast) {
2198 if (!_gvn.type(base)->speculative_maybe_null() &&
2199 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2200 // According to profiling, this access is always on
2201 // heap. Casting the base to not null and thus avoiding membars
2202 // around the access should allow better optimizations
2203 Node* null_ctl = top();
2204 base = null_check_oop(base, &null_ctl, true, true, true);
2205 assert(null_ctl->is_top(), "no null control here");
2206 return basic_plus_adr(base, offset);
2207 } else if (_gvn.type(base)->speculative_always_null() &&
2208 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2209 // According to profiling, this access is always off
2210 // heap.
2211 base = null_assert(base);
2212 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2213 offset = MakeConX(0);
2214 return basic_plus_adr(top(), raw_base, offset);
2215 }
2216 }
2217 // We don't know if it's an on heap or off heap access. Fall back
2218 // to raw memory access.
2219 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2220 return basic_plus_adr(top(), raw, offset);
2221 } else {
2222 assert(base == uncasted_base, "unexpected base change");
2223 // We know it's an on heap access so base can't be null
2224 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2225 base = must_be_not_null(base, true);
2226 }
2227 return basic_plus_adr(base, offset);
2228 }
2229 }
2230
2231 //--------------------------inline_number_methods-----------------------------
2232 // inline int Integer.numberOfLeadingZeros(int)
2233 // inline int Long.numberOfLeadingZeros(long)
2234 //
2235 // inline int Integer.numberOfTrailingZeros(int)
2236 // inline int Long.numberOfTrailingZeros(long)
2237 //
2238 // inline int Integer.bitCount(int)
2239 // inline int Long.bitCount(long)
2240 //
2241 // inline char Character.reverseBytes(char)
2242 // inline short Short.reverseBytes(short)
2243 // inline int Integer.reverseBytes(int)
2244 // inline long Long.reverseBytes(long)
2245 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2246 Node* arg = argument(0);
2247 Node* n = nullptr;
2248 switch (id) {
2249 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2250 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2251 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2252 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2253 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2254 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2255 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2256 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2257 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2258 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2259 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2260 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2261 default: fatal_unexpected_iid(id); break;
2262 }
2263 set_result(_gvn.transform(n));
2264 return true;
2265 }
2266
2267 //--------------------------inline_bitshuffle_methods-----------------------------
2268 // inline int Integer.compress(int, int)
2269 // inline int Integer.expand(int, int)
2270 // inline long Long.compress(long, long)
2271 // inline long Long.expand(long, long)
2272 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2273 Node* n = nullptr;
2274 switch (id) {
2275 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2276 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2277 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2278 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2279 default: fatal_unexpected_iid(id); break;
2280 }
2281 set_result(_gvn.transform(n));
2282 return true;
2283 }
2284
2285 //--------------------------inline_number_methods-----------------------------
2286 // inline int Integer.compareUnsigned(int, int)
2287 // inline int Long.compareUnsigned(long, long)
2288 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2289 Node* arg1 = argument(0);
2290 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2291 Node* n = nullptr;
2292 switch (id) {
2293 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2294 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2295 default: fatal_unexpected_iid(id); break;
2296 }
2297 set_result(_gvn.transform(n));
2298 return true;
2299 }
2300
2301 //--------------------------inline_unsigned_divmod_methods-----------------------------
2302 // inline int Integer.divideUnsigned(int, int)
2303 // inline int Integer.remainderUnsigned(int, int)
2304 // inline long Long.divideUnsigned(long, long)
2305 // inline long Long.remainderUnsigned(long, long)
2306 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2307 Node* n = nullptr;
2308 switch (id) {
2309 case vmIntrinsics::_divideUnsigned_i: {
2310 zero_check_int(argument(1));
2311 // Compile-time detect of null-exception
2312 if (stopped()) {
2313 return true; // keep the graph constructed so far
2314 }
2315 n = new UDivINode(control(), argument(0), argument(1));
2316 break;
2317 }
2318 case vmIntrinsics::_divideUnsigned_l: {
2319 zero_check_long(argument(2));
2320 // Compile-time detect of null-exception
2321 if (stopped()) {
2322 return true; // keep the graph constructed so far
2323 }
2324 n = new UDivLNode(control(), argument(0), argument(2));
2325 break;
2326 }
2327 case vmIntrinsics::_remainderUnsigned_i: {
2328 zero_check_int(argument(1));
2329 // Compile-time detect of null-exception
2330 if (stopped()) {
2331 return true; // keep the graph constructed so far
2332 }
2333 n = new UModINode(control(), argument(0), argument(1));
2334 break;
2335 }
2336 case vmIntrinsics::_remainderUnsigned_l: {
2337 zero_check_long(argument(2));
2338 // Compile-time detect of null-exception
2339 if (stopped()) {
2340 return true; // keep the graph constructed so far
2341 }
2342 n = new UModLNode(control(), argument(0), argument(2));
2343 break;
2344 }
2345 default: fatal_unexpected_iid(id); break;
2346 }
2347 set_result(_gvn.transform(n));
2348 return true;
2349 }
2350
2351 //----------------------------inline_unsafe_access----------------------------
2352
2353 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2354 // Attempt to infer a sharper value type from the offset and base type.
2355 ciKlass* sharpened_klass = nullptr;
2356 bool null_free = false;
2357
2358 // See if it is an instance field, with an object type.
2359 if (alias_type->field() != nullptr) {
2360 if (alias_type->field()->type()->is_klass()) {
2361 sharpened_klass = alias_type->field()->type()->as_klass();
2362 null_free = alias_type->field()->is_null_free();
2363 }
2364 }
2365
2366 const TypeOopPtr* result = nullptr;
2367 // See if it is a narrow oop array.
2368 if (adr_type->isa_aryptr()) {
2369 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2370 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2371 null_free = adr_type->is_aryptr()->is_null_free();
2372 if (elem_type != nullptr && elem_type->is_loaded()) {
2373 // Sharpen the value type.
2374 result = elem_type;
2375 }
2376 }
2377 }
2378
2379 // The sharpened class might be unloaded if there is no class loader
2380 // contraint in place.
2381 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2382 // Sharpen the value type.
2383 result = TypeOopPtr::make_from_klass(sharpened_klass);
2384 if (null_free) {
2385 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2386 }
2387 }
2388 if (result != nullptr) {
2389 #ifndef PRODUCT
2390 if (C->print_intrinsics() || C->print_inlining()) {
2391 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2392 tty->print(" sharpened value: "); result->dump(); tty->cr();
2393 }
2394 #endif
2395 }
2396 return result;
2397 }
2398
2399 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2400 switch (kind) {
2401 case Relaxed:
2402 return MO_UNORDERED;
2403 case Opaque:
2404 return MO_RELAXED;
2405 case Acquire:
2406 return MO_ACQUIRE;
2407 case Release:
2408 return MO_RELEASE;
2409 case Volatile:
2410 return MO_SEQ_CST;
2411 default:
2412 ShouldNotReachHere();
2413 return 0;
2414 }
2415 }
2416
2417 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2418 _kit(kit),
2419 _sp(kit->sp()),
2420 _jvms(kit->jvms()),
2421 _map(kit->clone_map()),
2422 _discarded(false)
2423 {
2424 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2425 Node* out = kit->control()->fast_out(i);
2426 if (out->is_CFG()) {
2427 _ctrl_succ.push(out);
2428 }
2429 }
2430 }
2431
2432 LibraryCallKit::SavedState::~SavedState() {
2433 if (_discarded) {
2434 _kit->destruct_map_clone(_map);
2435 return;
2436 }
2437 _kit->jvms()->set_map(_map);
2438 _kit->jvms()->set_sp(_sp);
2439 _map->set_jvms(_kit->jvms());
2440 _kit->set_map(_map);
2441 _kit->set_sp(_sp);
2442 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2443 Node* out = _kit->control()->fast_out(i);
2444 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2445 _kit->_gvn.hash_delete(out);
2446 out->set_req(0, _kit->C->top());
2447 _kit->C->record_for_igvn(out);
2448 --i; --imax;
2449 _kit->_gvn.hash_find_insert(out);
2450 }
2451 }
2452 }
2453
2454 void LibraryCallKit::SavedState::discard() {
2455 _discarded = true;
2456 }
2457
2458 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2459 if (callee()->is_static()) return false; // caller must have the capability!
2460 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2461 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2462 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2463 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2464
2465 if (is_reference_type(type)) {
2466 decorators |= ON_UNKNOWN_OOP_REF;
2467 }
2468
2469 if (unaligned) {
2470 decorators |= C2_UNALIGNED;
2471 }
2472
2473 #ifndef PRODUCT
2474 {
2475 ResourceMark rm;
2476 // Check the signatures.
2477 ciSignature* sig = callee()->signature();
2478 #ifdef ASSERT
2479 if (!is_store) {
2480 // Object getReference(Object base, int/long offset), etc.
2481 BasicType rtype = sig->return_type()->basic_type();
2482 assert(rtype == type, "getter must return the expected value");
2483 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2484 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2485 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2486 } else {
2487 // void putReference(Object base, int/long offset, Object x), etc.
2488 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2489 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2490 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2491 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2492 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2493 assert(vtype == type, "putter must accept the expected value");
2494 }
2495 #endif // ASSERT
2496 }
2497 #endif //PRODUCT
2498
2499 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2500
2501 Node* receiver = argument(0); // type: oop
2502
2503 // Build address expression.
2504 Node* heap_base_oop = top();
2505
2506 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2507 Node* base = argument(1); // type: oop
2508 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2509 Node* offset = argument(2); // type: long
2510 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2511 // to be plain byte offsets, which are also the same as those accepted
2512 // by oopDesc::field_addr.
2513 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2514 "fieldOffset must be byte-scaled");
2515
2516 ciInlineKlass* inline_klass = nullptr;
2517 if (is_flat) {
2518 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2519 if (cls == nullptr || cls->const_oop() == nullptr) {
2520 return false;
2521 }
2522 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2523 if (!mirror_type->is_inlinetype()) {
2524 return false;
2525 }
2526 inline_klass = mirror_type->as_inline_klass();
2527 }
2528
2529 if (base->is_InlineType()) {
2530 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2531 InlineTypeNode* vt = base->as_InlineType();
2532 if (offset->is_Con()) {
2533 long off = find_long_con(offset, 0);
2534 ciInlineKlass* vk = vt->type()->inline_klass();
2535 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2536 return false;
2537 }
2538
2539 ciField* field = vk->get_non_flat_field_by_offset(off);
2540 if (field != nullptr) {
2541 BasicType bt = type2field[field->type()->basic_type()];
2542 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2543 bt = T_OBJECT;
2544 }
2545 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2546 Node* value = vt->field_value_by_offset(off, false);
2547 if (value->is_InlineType()) {
2548 value = value->as_InlineType()->adjust_scalarization_depth(this);
2549 }
2550 set_result(value);
2551 return true;
2552 }
2553 }
2554 }
2555 {
2556 // Re-execute the unsafe access if allocation triggers deoptimization.
2557 PreserveReexecuteState preexecs(this);
2558 jvms()->set_should_reexecute(true);
2559 vt = vt->buffer(this);
2560 }
2561 base = vt->get_oop();
2562 }
2563
2564 // 32-bit machines ignore the high half!
2565 offset = ConvL2X(offset);
2566
2567 // Save state and restore on bailout
2568 SavedState old_state(this);
2569
2570 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2571 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2572
2573 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2574 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2575 decorators |= IN_NATIVE; // off-heap primitive access
2576 } else {
2577 return false; // off-heap oop accesses are not supported
2578 }
2579 } else {
2580 heap_base_oop = base; // on-heap or mixed access
2581 }
2582
2583 // Can base be null? Otherwise, always on-heap access.
2584 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2585
2586 if (!can_access_non_heap) {
2587 decorators |= IN_HEAP;
2588 }
2589
2590 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2591
2592 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2593 if (adr_type == TypePtr::NULL_PTR) {
2594 return false; // off-heap access with zero address
2595 }
2596
2597 // Try to categorize the address.
2598 Compile::AliasType* alias_type = C->alias_type(adr_type);
2599 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2600
2601 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2602 alias_type->adr_type() == TypeAryPtr::RANGE) {
2603 return false; // not supported
2604 }
2605
2606 bool mismatched = false;
2607 BasicType bt = T_ILLEGAL;
2608 ciField* field = nullptr;
2609 if (adr_type->isa_instptr()) {
2610 const TypeInstPtr* instptr = adr_type->is_instptr();
2611 ciInstanceKlass* k = instptr->instance_klass();
2612 int off = instptr->offset();
2613 if (instptr->const_oop() != nullptr &&
2614 k == ciEnv::current()->Class_klass() &&
2615 instptr->offset() >= (k->size_helper() * wordSize)) {
2616 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2617 field = k->get_field_by_offset(off, true);
2618 } else {
2619 field = k->get_non_flat_field_by_offset(off);
2620 }
2621 if (field != nullptr) {
2622 bt = type2field[field->type()->basic_type()];
2623 }
2624 if (bt != alias_type->basic_type()) {
2625 // Type mismatch. Is it an access to a nested flat field?
2626 field = k->get_field_by_offset(off, false);
2627 if (field != nullptr) {
2628 bt = type2field[field->type()->basic_type()];
2629 }
2630 }
2631 assert(bt == alias_type->basic_type() || is_flat, "should match");
2632 } else {
2633 bt = alias_type->basic_type();
2634 }
2635
2636 if (bt != T_ILLEGAL) {
2637 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2638 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2639 // Alias type doesn't differentiate between byte[] and boolean[]).
2640 // Use address type to get the element type.
2641 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2642 }
2643 if (is_reference_type(bt, true)) {
2644 // accessing an array field with getReference is not a mismatch
2645 bt = T_OBJECT;
2646 }
2647 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2648 // Don't intrinsify mismatched object accesses
2649 return false;
2650 }
2651 mismatched = (bt != type);
2652 } else if (alias_type->adr_type()->isa_oopptr()) {
2653 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2654 }
2655
2656 if (is_flat) {
2657 if (adr_type->isa_instptr()) {
2658 if (field == nullptr || field->type() != inline_klass) {
2659 mismatched = true;
2660 }
2661 } else if (adr_type->isa_aryptr()) {
2662 const Type* elem = adr_type->is_aryptr()->elem();
2663 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2664 mismatched = true;
2665 }
2666 } else {
2667 mismatched = true;
2668 }
2669 if (is_store) {
2670 const Type* val_t = _gvn.type(val);
2671 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2672 return false;
2673 }
2674 }
2675 }
2676
2677 old_state.discard();
2678 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2679
2680 if (mismatched) {
2681 decorators |= C2_MISMATCHED;
2682 }
2683
2684 // First guess at the value type.
2685 const Type *value_type = Type::get_const_basic_type(type);
2686
2687 // Figure out the memory ordering.
2688 decorators |= mo_decorator_for_access_kind(kind);
2689
2690 if (!is_store) {
2691 if (type == T_OBJECT && !is_flat) {
2692 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2693 if (tjp != nullptr) {
2694 value_type = tjp;
2695 }
2696 }
2697 }
2698
2699 receiver = null_check(receiver);
2700 if (stopped()) {
2701 return true;
2702 }
2703 // Heap pointers get a null-check from the interpreter,
2704 // as a courtesy. However, this is not guaranteed by Unsafe,
2705 // and it is not possible to fully distinguish unintended nulls
2706 // from intended ones in this API.
2707
2708 if (!is_store) {
2709 Node* p = nullptr;
2710 // Try to constant fold a load from a constant field
2711
2712 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2713 // final or stable field
2714 p = make_constant_from_field(field, heap_base_oop);
2715 }
2716
2717 if (p == nullptr) { // Could not constant fold the load
2718 if (is_flat) {
2719 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, adr_type, false, false, true);
2720 } else {
2721 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2722 const TypeOopPtr* ptr = value_type->make_oopptr();
2723 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2724 // Load a non-flattened inline type from memory
2725 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2726 }
2727 }
2728 // Normalize the value returned by getBoolean in the following cases
2729 if (type == T_BOOLEAN &&
2730 (mismatched ||
2731 heap_base_oop == top() || // - heap_base_oop is null or
2732 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2733 // and the unsafe access is made to large offset
2734 // (i.e., larger than the maximum offset necessary for any
2735 // field access)
2736 ) {
2737 IdealKit ideal = IdealKit(this);
2738 #define __ ideal.
2739 IdealVariable normalized_result(ideal);
2740 __ declarations_done();
2741 __ set(normalized_result, p);
2742 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2743 __ set(normalized_result, ideal.ConI(1));
2744 ideal.end_if();
2745 final_sync(ideal);
2746 p = __ value(normalized_result);
2747 #undef __
2748 }
2749 }
2750 if (type == T_ADDRESS) {
2751 p = gvn().transform(new CastP2XNode(nullptr, p));
2752 p = ConvX2UL(p);
2753 }
2754 // The load node has the control of the preceding MemBarCPUOrder. All
2755 // following nodes will have the control of the MemBarCPUOrder inserted at
2756 // the end of this method. So, pushing the load onto the stack at a later
2757 // point is fine.
2758 set_result(p);
2759 } else {
2760 if (bt == T_ADDRESS) {
2761 // Repackage the long as a pointer.
2762 val = ConvL2X(val);
2763 val = gvn().transform(new CastX2PNode(val));
2764 }
2765 if (is_flat) {
2766 val->as_InlineType()->store_flat(this, base, adr, false, false, true, decorators);
2767 } else {
2768 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2769 }
2770 }
2771
2772 return true;
2773 }
2774
2775 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2776 #ifdef ASSERT
2777 {
2778 ResourceMark rm;
2779 // Check the signatures.
2780 ciSignature* sig = callee()->signature();
2781 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2782 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2783 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2784 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2785 if (is_store) {
2786 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2787 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2788 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2789 } else {
2790 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2791 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2792 }
2793 }
2794 #endif // ASSERT
2795
2796 assert(kind == Relaxed, "Only plain accesses for now");
2797 if (callee()->is_static()) {
2798 // caller must have the capability!
2799 return false;
2800 }
2801 C->set_has_unsafe_access(true);
2802
2803 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2804 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2805 // parameter valueType is not a constant
2806 return false;
2807 }
2808 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2809 if (!mirror_type->is_inlinetype()) {
2810 // Dead code
2811 return false;
2812 }
2813 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2814
2815 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2816 if (layout_type == nullptr || !layout_type->is_con()) {
2817 // parameter layoutKind is not a constant
2818 return false;
2819 }
2820 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2821 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2822 "invalid layoutKind %d", layout_type->get_con());
2823 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2824 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NON_ATOMIC_FLAT ||
2825 layout == LayoutKind::ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2826 "unexpected layoutKind %d", layout_type->get_con());
2827
2828 null_check(argument(0));
2829 if (stopped()) {
2830 return true;
2831 }
2832
2833 Node* base = must_be_not_null(argument(1), true);
2834 Node* offset = argument(2);
2835 const Type* base_type = _gvn.type(base);
2836
2837 Node* ptr;
2838 bool immutable_memory = false;
2839 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2840 if (base_type->isa_instptr()) {
2841 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2842 if (offset_type == nullptr || !offset_type->is_con()) {
2843 // Offset into a non-array should be a constant
2844 decorators |= C2_MISMATCHED;
2845 } else {
2846 int offset_con = checked_cast<int>(offset_type->get_con());
2847 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2848 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2849 if (field == nullptr) {
2850 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2851 decorators |= C2_MISMATCHED;
2852 } else {
2853 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2854 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2855 immutable_memory = field->is_strict() && field->is_final();
2856
2857 if (base->is_InlineType()) {
2858 assert(!is_store, "Cannot store into a non-larval value object");
2859 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2860 return true;
2861 }
2862 }
2863 }
2864
2865 if (base->is_InlineType()) {
2866 assert(!is_store, "Cannot store into a non-larval value object");
2867 base = base->as_InlineType()->buffer(this, true);
2868 }
2869 ptr = basic_plus_adr(base, ConvL2X(offset));
2870 } else if (base_type->isa_aryptr()) {
2871 decorators |= IS_ARRAY;
2872 if (layout == LayoutKind::REFERENCE) {
2873 if (!base_type->is_aryptr()->is_not_flat()) {
2874 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2875 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::StrongDependency));
2876 replace_in_map(base, new_base);
2877 base = new_base;
2878 }
2879 ptr = basic_plus_adr(base, ConvL2X(offset));
2880 } else {
2881 if (UseArrayFlattening) {
2882 // Flat array must have an exact type
2883 bool is_null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2884 bool is_atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2885 Node* new_base = cast_to_flat_array(base, value_klass, is_null_free, !is_null_free, is_atomic);
2886 replace_in_map(base, new_base);
2887 base = new_base;
2888 ptr = basic_plus_adr(base, ConvL2X(offset));
2889 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2890 if (ptr_type->field_offset().get() != 0) {
2891 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::StrongDependency));
2892 }
2893 } else {
2894 uncommon_trap(Deoptimization::Reason_intrinsic,
2895 Deoptimization::Action_none);
2896 return true;
2897 }
2898 }
2899 } else {
2900 decorators |= C2_MISMATCHED;
2901 ptr = basic_plus_adr(base, ConvL2X(offset));
2902 }
2903
2904 if (is_store) {
2905 Node* value = argument(6);
2906 const Type* value_type = _gvn.type(value);
2907 if (!value_type->is_inlinetypeptr()) {
2908 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2909 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::StrongDependency));
2910 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2911 replace_in_map(value, new_value);
2912 value = new_value;
2913 }
2914
2915 assert(value_type->inline_klass() == value_klass, "value is of type %s while valueType is %s", value_type->inline_klass()->name()->as_utf8(), value_klass->name()->as_utf8());
2916 if (layout == LayoutKind::REFERENCE) {
2917 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2918 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2919 } else {
2920 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2921 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2922 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2923 }
2924
2925 return true;
2926 } else {
2927 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2928 InlineTypeNode* result;
2929 if (layout == LayoutKind::REFERENCE) {
2930 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2931 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2932 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2933 } else {
2934 bool atomic = layout != LayoutKind::NON_ATOMIC_FLAT;
2935 bool null_free = layout != LayoutKind::NULLABLE_ATOMIC_FLAT;
2936 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2937 }
2938
2939 set_result(result);
2940 return true;
2941 }
2942 }
2943
2944 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2945 Node* receiver = argument(0);
2946 Node* value = argument(1);
2947
2948 const Type* type = gvn().type(value);
2949 if (!type->is_inlinetypeptr()) {
2950 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2951 return false;
2952 }
2953
2954 null_check(receiver);
2955 if (stopped()) {
2956 return true;
2957 }
2958
2959 value = null_check(value);
2960 if (stopped()) {
2961 return true;
2962 }
2963
2964 ciInlineKlass* vk = type->inline_klass();
2965 Node* klass = makecon(TypeKlassPtr::make(vk));
2966 Node* obj = new_instance(klass);
2967 AllocateNode::Ideal_allocation(obj)->_larval = true;
2968
2969 assert(value->is_InlineType(), "must be an InlineTypeNode");
2970 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2971 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2972
2973 set_result(obj);
2974 return true;
2975 }
2976
2977 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2978 Node* receiver = argument(0);
2979 Node* buffer = argument(1);
2980
2981 const Type* type = gvn().type(buffer);
2982 if (!type->is_inlinetypeptr()) {
2983 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2984 return false;
2985 }
2986
2987 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2988 if (alloc == nullptr) {
2989 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2990 return false;
2991 }
2992
2993 null_check(receiver);
2994 if (stopped()) {
2995 return true;
2996 }
2997
2998 // Unset the larval bit in the object header
2999 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
3000 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
3001 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
3002
3003 // We must ensure that the buffer is properly published
3004 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
3005 assert(!type->maybe_null(), "result of an allocation should not be null");
3006 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
3007 return true;
3008 }
3009
3010 //----------------------------inline_unsafe_load_store----------------------------
3011 // This method serves a couple of different customers (depending on LoadStoreKind):
3012 //
3013 // LS_cmp_swap:
3014 //
3015 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
3016 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
3017 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
3018 //
3019 // LS_cmp_swap_weak:
3020 //
3021 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
3022 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
3023 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
3024 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
3025 //
3026 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
3027 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
3028 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
3029 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
3030 //
3031 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
3032 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
3033 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
3034 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
3035 //
3036 // LS_cmp_exchange:
3037 //
3038 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
3039 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
3040 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
3041 //
3042 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
3043 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
3044 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
3045 //
3046 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
3047 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
3048 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
3049 //
3050 // LS_get_add:
3051 //
3052 // int getAndAddInt( Object o, long offset, int delta)
3053 // long getAndAddLong(Object o, long offset, long delta)
3054 //
3055 // LS_get_set:
3056 //
3057 // int getAndSet(Object o, long offset, int newValue)
3058 // long getAndSet(Object o, long offset, long newValue)
3059 // Object getAndSet(Object o, long offset, Object newValue)
3060 //
3061 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
3062 // This basic scheme here is the same as inline_unsafe_access, but
3063 // differs in enough details that combining them would make the code
3064 // overly confusing. (This is a true fact! I originally combined
3065 // them, but even I was confused by it!) As much code/comments as
3066 // possible are retained from inline_unsafe_access though to make
3067 // the correspondences clearer. - dl
3068
3069 if (callee()->is_static()) return false; // caller must have the capability!
3070
3071 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3072 decorators |= mo_decorator_for_access_kind(access_kind);
3073
3074 #ifndef PRODUCT
3075 BasicType rtype;
3076 {
3077 ResourceMark rm;
3078 // Check the signatures.
3079 ciSignature* sig = callee()->signature();
3080 rtype = sig->return_type()->basic_type();
3081 switch(kind) {
3082 case LS_get_add:
3083 case LS_get_set: {
3084 // Check the signatures.
3085 #ifdef ASSERT
3086 assert(rtype == type, "get and set must return the expected type");
3087 assert(sig->count() == 3, "get and set has 3 arguments");
3088 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3089 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3090 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3091 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3092 #endif // ASSERT
3093 break;
3094 }
3095 case LS_cmp_swap:
3096 case LS_cmp_swap_weak: {
3097 // Check the signatures.
3098 #ifdef ASSERT
3099 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3100 assert(sig->count() == 4, "CAS has 4 arguments");
3101 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3102 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3103 #endif // ASSERT
3104 break;
3105 }
3106 case LS_cmp_exchange: {
3107 // Check the signatures.
3108 #ifdef ASSERT
3109 assert(rtype == type, "CAS must return the expected type");
3110 assert(sig->count() == 4, "CAS has 4 arguments");
3111 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3112 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3113 #endif // ASSERT
3114 break;
3115 }
3116 default:
3117 ShouldNotReachHere();
3118 }
3119 }
3120 #endif //PRODUCT
3121
3122 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3123
3124 // Get arguments:
3125 Node* receiver = nullptr;
3126 Node* base = nullptr;
3127 Node* offset = nullptr;
3128 Node* oldval = nullptr;
3129 Node* newval = nullptr;
3130 switch(kind) {
3131 case LS_cmp_swap:
3132 case LS_cmp_swap_weak:
3133 case LS_cmp_exchange: {
3134 const bool two_slot_type = type2size[type] == 2;
3135 receiver = argument(0); // type: oop
3136 base = argument(1); // type: oop
3137 offset = argument(2); // type: long
3138 oldval = argument(4); // type: oop, int, or long
3139 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3140 break;
3141 }
3142 case LS_get_add:
3143 case LS_get_set: {
3144 receiver = argument(0); // type: oop
3145 base = argument(1); // type: oop
3146 offset = argument(2); // type: long
3147 oldval = nullptr;
3148 newval = argument(4); // type: oop, int, or long
3149 break;
3150 }
3151 default:
3152 ShouldNotReachHere();
3153 }
3154
3155 // Build field offset expression.
3156 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3157 // to be plain byte offsets, which are also the same as those accepted
3158 // by oopDesc::field_addr.
3159 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3160 // 32-bit machines ignore the high half of long offsets
3161 offset = ConvL2X(offset);
3162 // Save state and restore on bailout
3163 SavedState old_state(this);
3164 Node* adr = make_unsafe_address(base, offset,type, false);
3165 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3166
3167 Compile::AliasType* alias_type = C->alias_type(adr_type);
3168 BasicType bt = alias_type->basic_type();
3169 if (bt != T_ILLEGAL &&
3170 (is_reference_type(bt) != (type == T_OBJECT))) {
3171 // Don't intrinsify mismatched object accesses.
3172 return false;
3173 }
3174
3175 old_state.discard();
3176
3177 // For CAS, unlike inline_unsafe_access, there seems no point in
3178 // trying to refine types. Just use the coarse types here.
3179 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3180 const Type *value_type = Type::get_const_basic_type(type);
3181
3182 switch (kind) {
3183 case LS_get_set:
3184 case LS_cmp_exchange: {
3185 if (type == T_OBJECT) {
3186 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3187 if (tjp != nullptr) {
3188 value_type = tjp;
3189 }
3190 }
3191 break;
3192 }
3193 case LS_cmp_swap:
3194 case LS_cmp_swap_weak:
3195 case LS_get_add:
3196 break;
3197 default:
3198 ShouldNotReachHere();
3199 }
3200
3201 // Null check receiver.
3202 receiver = null_check(receiver);
3203 if (stopped()) {
3204 return true;
3205 }
3206
3207 int alias_idx = C->get_alias_index(adr_type);
3208
3209 if (is_reference_type(type)) {
3210 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3211
3212 if (oldval != nullptr && oldval->is_InlineType()) {
3213 // Re-execute the unsafe access if allocation triggers deoptimization.
3214 PreserveReexecuteState preexecs(this);
3215 jvms()->set_should_reexecute(true);
3216 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3217 }
3218 if (newval != nullptr && newval->is_InlineType()) {
3219 // Re-execute the unsafe access if allocation triggers deoptimization.
3220 PreserveReexecuteState preexecs(this);
3221 jvms()->set_should_reexecute(true);
3222 newval = newval->as_InlineType()->buffer(this)->get_oop();
3223 }
3224
3225 // Transformation of a value which could be null pointer (CastPP #null)
3226 // could be delayed during Parse (for example, in adjust_map_after_if()).
3227 // Execute transformation here to avoid barrier generation in such case.
3228 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3229 newval = _gvn.makecon(TypePtr::NULL_PTR);
3230
3231 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3232 // Refine the value to a null constant, when it is known to be null
3233 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3234 }
3235 }
3236
3237 Node* result = nullptr;
3238 switch (kind) {
3239 case LS_cmp_exchange: {
3240 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3241 oldval, newval, value_type, type, decorators);
3242 break;
3243 }
3244 case LS_cmp_swap_weak:
3245 decorators |= C2_WEAK_CMPXCHG;
3246 case LS_cmp_swap: {
3247 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3248 oldval, newval, value_type, type, decorators);
3249 break;
3250 }
3251 case LS_get_set: {
3252 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3253 newval, value_type, type, decorators);
3254 break;
3255 }
3256 case LS_get_add: {
3257 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3258 newval, value_type, type, decorators);
3259 break;
3260 }
3261 default:
3262 ShouldNotReachHere();
3263 }
3264
3265 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3266 set_result(result);
3267 return true;
3268 }
3269
3270 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3271 // Regardless of form, don't allow previous ld/st to move down,
3272 // then issue acquire, release, or volatile mem_bar.
3273 insert_mem_bar(Op_MemBarCPUOrder);
3274 switch(id) {
3275 case vmIntrinsics::_loadFence:
3276 insert_mem_bar(Op_LoadFence);
3277 return true;
3278 case vmIntrinsics::_storeFence:
3279 insert_mem_bar(Op_StoreFence);
3280 return true;
3281 case vmIntrinsics::_storeStoreFence:
3282 insert_mem_bar(Op_StoreStoreFence);
3283 return true;
3284 case vmIntrinsics::_fullFence:
3285 insert_mem_bar(Op_MemBarVolatile);
3286 return true;
3287 default:
3288 fatal_unexpected_iid(id);
3289 return false;
3290 }
3291 }
3292
3293 bool LibraryCallKit::inline_onspinwait() {
3294 insert_mem_bar(Op_OnSpinWait);
3295 return true;
3296 }
3297
3298 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3299 if (!kls->is_Con()) {
3300 return true;
3301 }
3302 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3303 if (klsptr == nullptr) {
3304 return true;
3305 }
3306 ciInstanceKlass* ik = klsptr->instance_klass();
3307 // don't need a guard for a klass that is already initialized
3308 return !ik->is_initialized();
3309 }
3310
3311 //----------------------------inline_unsafe_writeback0-------------------------
3312 // public native void Unsafe.writeback0(long address)
3313 bool LibraryCallKit::inline_unsafe_writeback0() {
3314 if (!Matcher::has_match_rule(Op_CacheWB)) {
3315 return false;
3316 }
3317 #ifndef PRODUCT
3318 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3319 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3320 ciSignature* sig = callee()->signature();
3321 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3322 #endif
3323 null_check_receiver(); // null-check, then ignore
3324 Node *addr = argument(1);
3325 addr = new CastX2PNode(addr);
3326 addr = _gvn.transform(addr);
3327 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3328 flush = _gvn.transform(flush);
3329 set_memory(flush, TypeRawPtr::BOTTOM);
3330 return true;
3331 }
3332
3333 //----------------------------inline_unsafe_writeback0-------------------------
3334 // public native void Unsafe.writeback0(long address)
3335 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3336 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3337 return false;
3338 }
3339 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3340 return false;
3341 }
3342 #ifndef PRODUCT
3343 assert(Matcher::has_match_rule(Op_CacheWB),
3344 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3345 : "found match rule for CacheWBPostSync but not CacheWB"));
3346
3347 #endif
3348 null_check_receiver(); // null-check, then ignore
3349 Node *sync;
3350 if (is_pre) {
3351 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3352 } else {
3353 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3354 }
3355 sync = _gvn.transform(sync);
3356 set_memory(sync, TypeRawPtr::BOTTOM);
3357 return true;
3358 }
3359
3360 //----------------------------inline_unsafe_allocate---------------------------
3361 // public native Object Unsafe.allocateInstance(Class<?> cls);
3362 bool LibraryCallKit::inline_unsafe_allocate() {
3363
3364 #if INCLUDE_JVMTI
3365 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3366 return false;
3367 }
3368 #endif //INCLUDE_JVMTI
3369
3370 if (callee()->is_static()) return false; // caller must have the capability!
3371
3372 null_check_receiver(); // null-check, then ignore
3373 Node* cls = null_check(argument(1));
3374 if (stopped()) return true;
3375
3376 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3377 kls = null_check(kls);
3378 if (stopped()) return true; // argument was like int.class
3379
3380 #if INCLUDE_JVMTI
3381 // Don't try to access new allocated obj in the intrinsic.
3382 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3383 // Deoptimize and allocate in interpreter instead.
3384 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3385 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3386 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3387 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3388 {
3389 BuildCutout unless(this, tst, PROB_MAX);
3390 uncommon_trap(Deoptimization::Reason_intrinsic,
3391 Deoptimization::Action_make_not_entrant);
3392 }
3393 if (stopped()) {
3394 return true;
3395 }
3396 #endif //INCLUDE_JVMTI
3397
3398 Node* test = nullptr;
3399 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3400 // Note: The argument might still be an illegal value like
3401 // Serializable.class or Object[].class. The runtime will handle it.
3402 // But we must make an explicit check for initialization.
3403 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3404 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3405 // can generate code to load it as unsigned byte.
3406 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3407 Node* bits = intcon(InstanceKlass::fully_initialized);
3408 test = _gvn.transform(new SubINode(inst, bits));
3409 // The 'test' is non-zero if we need to take a slow path.
3410 }
3411 Node* obj = nullptr;
3412 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3413 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3414 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3415 } else {
3416 obj = new_instance(kls, test);
3417 }
3418 set_result(obj);
3419 return true;
3420 }
3421
3422 //------------------------inline_native_time_funcs--------------
3423 // inline code for System.currentTimeMillis() and System.nanoTime()
3424 // these have the same type and signature
3425 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3426 const TypeFunc* tf = OptoRuntime::void_long_Type();
3427 const TypePtr* no_memory_effects = nullptr;
3428 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3429 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3430 #ifdef ASSERT
3431 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3432 assert(value_top == top(), "second value must be top");
3433 #endif
3434 set_result(value);
3435 return true;
3436 }
3437
3438
3439 #if INCLUDE_JVMTI
3440
3441 // When notifications are disabled then just update the VTMS transition bit and return.
3442 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3443 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3444 if (!DoJVMTIVirtualThreadTransitions) {
3445 return true;
3446 }
3447 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3448 IdealKit ideal(this);
3449
3450 Node* ONE = ideal.ConI(1);
3451 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3452 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3453 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3454
3455 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3456 sync_kit(ideal);
3457 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3458 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3459 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3460 ideal.sync_kit(this);
3461 } ideal.else_(); {
3462 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3463 Node* thread = ideal.thread();
3464 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3465 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3466
3467 sync_kit(ideal);
3468 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3469 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3470
3471 ideal.sync_kit(this);
3472 } ideal.end_if();
3473 final_sync(ideal);
3474
3475 return true;
3476 }
3477
3478 // Always update the is_disable_suspend bit.
3479 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3480 if (!DoJVMTIVirtualThreadTransitions) {
3481 return true;
3482 }
3483 IdealKit ideal(this);
3484
3485 {
3486 // unconditionally update the is_disable_suspend bit in current JavaThread
3487 Node* thread = ideal.thread();
3488 Node* arg = _gvn.transform(argument(0)); // argument for notification
3489 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3490 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3491
3492 sync_kit(ideal);
3493 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3494 ideal.sync_kit(this);
3495 }
3496 final_sync(ideal);
3497
3498 return true;
3499 }
3500
3501 #endif // INCLUDE_JVMTI
3502
3503 #ifdef JFR_HAVE_INTRINSICS
3504
3505 /**
3506 * if oop->klass != null
3507 * // normal class
3508 * epoch = _epoch_state ? 2 : 1
3509 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3510 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3511 * }
3512 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3513 * else
3514 * // primitive class
3515 * if oop->array_klass != null
3516 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3517 * else
3518 * id = LAST_TYPE_ID + 1 // void class path
3519 * if (!signaled)
3520 * signaled = true
3521 */
3522 bool LibraryCallKit::inline_native_classID() {
3523 Node* cls = argument(0);
3524
3525 IdealKit ideal(this);
3526 #define __ ideal.
3527 IdealVariable result(ideal); __ declarations_done();
3528 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3529 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3530 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3531
3532
3533 __ if_then(kls, BoolTest::ne, null()); {
3534 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3535 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3536
3537 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3538 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3539 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3540 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3541 mask = _gvn.transform(new OrLNode(mask, epoch));
3542 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3543
3544 float unlikely = PROB_UNLIKELY(0.999);
3545 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3546 sync_kit(ideal);
3547 make_runtime_call(RC_LEAF,
3548 OptoRuntime::class_id_load_barrier_Type(),
3549 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3550 "class id load barrier",
3551 TypePtr::BOTTOM,
3552 kls);
3553 ideal.sync_kit(this);
3554 } __ end_if();
3555
3556 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3557 } __ else_(); {
3558 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3559 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3560 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3561 __ if_then(array_kls, BoolTest::ne, null()); {
3562 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3563 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3564 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3565 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3566 } __ else_(); {
3567 // void class case
3568 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3569 } __ end_if();
3570
3571 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3572 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3573 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3574 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3575 } __ end_if();
3576 } __ end_if();
3577
3578 final_sync(ideal);
3579 set_result(ideal.value(result));
3580 #undef __
3581 return true;
3582 }
3583
3584 //------------------------inline_native_jvm_commit------------------
3585 bool LibraryCallKit::inline_native_jvm_commit() {
3586 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3587
3588 // Save input memory and i_o state.
3589 Node* input_memory_state = reset_memory();
3590 set_all_memory(input_memory_state);
3591 Node* input_io_state = i_o();
3592
3593 // TLS.
3594 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3595 // Jfr java buffer.
3596 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3597 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3598 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3599
3600 // Load the current value of the notified field in the JfrThreadLocal.
3601 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3602 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3603
3604 // Test for notification.
3605 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3606 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3607 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3608
3609 // True branch, is notified.
3610 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3611 set_control(is_notified);
3612
3613 // Reset notified state.
3614 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3615 Node* notified_reset_memory = reset_memory();
3616
3617 // 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.
3618 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3619 // Convert the machine-word to a long.
3620 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3621
3622 // False branch, not notified.
3623 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3624 set_control(not_notified);
3625 set_all_memory(input_memory_state);
3626
3627 // Arg is the next position as a long.
3628 Node* arg = argument(0);
3629 // Convert long to machine-word.
3630 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3631
3632 // Store the next_position to the underlying jfr java buffer.
3633 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3634
3635 Node* commit_memory = reset_memory();
3636 set_all_memory(commit_memory);
3637
3638 // 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.
3639 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3640 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3641 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3642
3643 // And flags with lease constant.
3644 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3645
3646 // Branch on lease to conditionalize returning the leased java buffer.
3647 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3648 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3649 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3650
3651 // False branch, not a lease.
3652 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3653
3654 // True branch, is lease.
3655 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3656 set_control(is_lease);
3657
3658 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3659 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3660 OptoRuntime::void_void_Type(),
3661 SharedRuntime::jfr_return_lease(),
3662 "return_lease", TypePtr::BOTTOM);
3663 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3664
3665 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3666 record_for_igvn(lease_compare_rgn);
3667 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3668 record_for_igvn(lease_compare_mem);
3669 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3670 record_for_igvn(lease_compare_io);
3671 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3672 record_for_igvn(lease_result_value);
3673
3674 // Update control and phi nodes.
3675 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3676 lease_compare_rgn->init_req(_false_path, not_lease);
3677
3678 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3679 lease_compare_mem->init_req(_false_path, commit_memory);
3680
3681 lease_compare_io->init_req(_true_path, i_o());
3682 lease_compare_io->init_req(_false_path, input_io_state);
3683
3684 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3685 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3686
3687 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3688 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3689 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3690 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3691
3692 // Update control and phi nodes.
3693 result_rgn->init_req(_true_path, is_notified);
3694 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3695
3696 result_mem->init_req(_true_path, notified_reset_memory);
3697 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3698
3699 result_io->init_req(_true_path, input_io_state);
3700 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3701
3702 result_value->init_req(_true_path, current_pos);
3703 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3704
3705 // Set output state.
3706 set_control(_gvn.transform(result_rgn));
3707 set_all_memory(_gvn.transform(result_mem));
3708 set_i_o(_gvn.transform(result_io));
3709 set_result(result_rgn, result_value);
3710 return true;
3711 }
3712
3713 /*
3714 * The intrinsic is a model of this pseudo-code:
3715 *
3716 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3717 * jobject h_event_writer = tl->java_event_writer();
3718 * if (h_event_writer == nullptr) {
3719 * return nullptr;
3720 * }
3721 * oop threadObj = Thread::threadObj();
3722 * oop vthread = java_lang_Thread::vthread(threadObj);
3723 * traceid tid;
3724 * bool pinVirtualThread;
3725 * bool excluded;
3726 * if (vthread != threadObj) { // i.e. current thread is virtual
3727 * tid = java_lang_Thread::tid(vthread);
3728 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3729 * pinVirtualThread = VMContinuations;
3730 * excluded = vthread_epoch_raw & excluded_mask;
3731 * if (!excluded) {
3732 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3733 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3734 * if (vthread_epoch != current_epoch) {
3735 * write_checkpoint();
3736 * }
3737 * }
3738 * } else {
3739 * tid = java_lang_Thread::tid(threadObj);
3740 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3741 * pinVirtualThread = false;
3742 * excluded = thread_epoch_raw & excluded_mask;
3743 * }
3744 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3745 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3746 * if (tid_in_event_writer != tid) {
3747 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3748 * setField(event_writer, "excluded", excluded);
3749 * setField(event_writer, "threadID", tid);
3750 * }
3751 * return event_writer
3752 */
3753 bool LibraryCallKit::inline_native_getEventWriter() {
3754 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3755
3756 // Save input memory and i_o state.
3757 Node* input_memory_state = reset_memory();
3758 set_all_memory(input_memory_state);
3759 Node* input_io_state = i_o();
3760
3761 // The most significant bit of the u2 is used to denote thread exclusion
3762 Node* excluded_shift = _gvn.intcon(15);
3763 Node* excluded_mask = _gvn.intcon(1 << 15);
3764 // The epoch generation is the range [1-32767]
3765 Node* epoch_mask = _gvn.intcon(32767);
3766
3767 // TLS
3768 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3769
3770 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3771 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3772
3773 // Load the eventwriter jobject handle.
3774 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3775
3776 // Null check the jobject handle.
3777 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3778 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3779 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3780
3781 // False path, jobj is null.
3782 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3783
3784 // True path, jobj is not null.
3785 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3786
3787 set_control(jobj_is_not_null);
3788
3789 // Load the threadObj for the CarrierThread.
3790 Node* threadObj = generate_current_thread(tls_ptr);
3791
3792 // Load the vthread.
3793 Node* vthread = generate_virtual_thread(tls_ptr);
3794
3795 // If vthread != threadObj, this is a virtual thread.
3796 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3797 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3798 IfNode* iff_vthread_not_equal_threadObj =
3799 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3800
3801 // False branch, fallback to threadObj.
3802 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3803 set_control(vthread_equal_threadObj);
3804
3805 // Load the tid field from the vthread object.
3806 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3807
3808 // Load the raw epoch value from the threadObj.
3809 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3810 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3811 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3812 TypeInt::CHAR, T_CHAR,
3813 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3814
3815 // Mask off the excluded information from the epoch.
3816 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3817
3818 // True branch, this is a virtual thread.
3819 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3820 set_control(vthread_not_equal_threadObj);
3821
3822 // Load the tid field from the vthread object.
3823 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3824
3825 // Continuation support determines if a virtual thread should be pinned.
3826 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3827 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3828
3829 // Load the raw epoch value from the vthread.
3830 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3831 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3832 TypeInt::CHAR, T_CHAR,
3833 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3834
3835 // Mask off the excluded information from the epoch.
3836 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3837
3838 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3839 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3840 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3841 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3842
3843 // False branch, vthread is excluded, no need to write epoch info.
3844 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3845
3846 // True branch, vthread is included, update epoch info.
3847 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3848 set_control(included);
3849
3850 // Get epoch value.
3851 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3852
3853 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3854 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3855 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3856
3857 // Compare the epoch in the vthread to the current epoch generation.
3858 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3859 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3860 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3861
3862 // False path, epoch is equal, checkpoint information is valid.
3863 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3864
3865 // True path, epoch is not equal, write a checkpoint for the vthread.
3866 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3867
3868 set_control(epoch_is_not_equal);
3869
3870 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3871 // The call also updates the native thread local thread id and the vthread with the current epoch.
3872 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3873 OptoRuntime::jfr_write_checkpoint_Type(),
3874 SharedRuntime::jfr_write_checkpoint(),
3875 "write_checkpoint", TypePtr::BOTTOM);
3876 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3877
3878 // vthread epoch != current epoch
3879 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3880 record_for_igvn(epoch_compare_rgn);
3881 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3882 record_for_igvn(epoch_compare_mem);
3883 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3884 record_for_igvn(epoch_compare_io);
3885
3886 // Update control and phi nodes.
3887 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3888 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3889 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3890 epoch_compare_mem->init_req(_false_path, input_memory_state);
3891 epoch_compare_io->init_req(_true_path, i_o());
3892 epoch_compare_io->init_req(_false_path, input_io_state);
3893
3894 // excluded != true
3895 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3896 record_for_igvn(exclude_compare_rgn);
3897 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3898 record_for_igvn(exclude_compare_mem);
3899 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3900 record_for_igvn(exclude_compare_io);
3901
3902 // Update control and phi nodes.
3903 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3904 exclude_compare_rgn->init_req(_false_path, excluded);
3905 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3906 exclude_compare_mem->init_req(_false_path, input_memory_state);
3907 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3908 exclude_compare_io->init_req(_false_path, input_io_state);
3909
3910 // vthread != threadObj
3911 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3912 record_for_igvn(vthread_compare_rgn);
3913 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3914 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3915 record_for_igvn(vthread_compare_io);
3916 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3917 record_for_igvn(tid);
3918 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3919 record_for_igvn(exclusion);
3920 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3921 record_for_igvn(pinVirtualThread);
3922
3923 // Update control and phi nodes.
3924 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3925 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3926 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3927 vthread_compare_mem->init_req(_false_path, input_memory_state);
3928 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3929 vthread_compare_io->init_req(_false_path, input_io_state);
3930 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3931 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3932 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3933 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3934 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3935 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3936
3937 // Update branch state.
3938 set_control(_gvn.transform(vthread_compare_rgn));
3939 set_all_memory(_gvn.transform(vthread_compare_mem));
3940 set_i_o(_gvn.transform(vthread_compare_io));
3941
3942 // Load the event writer oop by dereferencing the jobject handle.
3943 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3944 assert(klass_EventWriter->is_loaded(), "invariant");
3945 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3946 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3947 const TypeOopPtr* const xtype = aklass->as_instance_type();
3948 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3949 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3950
3951 // Load the current thread id from the event writer object.
3952 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3953 // Get the field offset to, conditionally, store an updated tid value later.
3954 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3955 // Get the field offset to, conditionally, store an updated exclusion value later.
3956 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3957 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3958 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3959
3960 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3961 record_for_igvn(event_writer_tid_compare_rgn);
3962 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3963 record_for_igvn(event_writer_tid_compare_mem);
3964 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3965 record_for_igvn(event_writer_tid_compare_io);
3966
3967 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3968 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3969 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3970 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3971
3972 // False path, tids are the same.
3973 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3974
3975 // True path, tid is not equal, need to update the tid in the event writer.
3976 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3977 record_for_igvn(tid_is_not_equal);
3978
3979 // Store the pin state to the event writer.
3980 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3981
3982 // Store the exclusion state to the event writer.
3983 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3984 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3985
3986 // Store the tid to the event writer.
3987 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3988
3989 // Update control and phi nodes.
3990 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3991 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3992 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3993 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3994 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3995 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3996
3997 // Result of top level CFG, Memory, IO and Value.
3998 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3999 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4000 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4001 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4002
4003 // Result control.
4004 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4005 result_rgn->init_req(_false_path, jobj_is_null);
4006
4007 // Result memory.
4008 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4009 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4010
4011 // Result IO.
4012 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4013 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4014
4015 // Result value.
4016 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4017 result_value->init_req(_false_path, null()); // return null
4018
4019 // Set output state.
4020 set_control(_gvn.transform(result_rgn));
4021 set_all_memory(_gvn.transform(result_mem));
4022 set_i_o(_gvn.transform(result_io));
4023 set_result(result_rgn, result_value);
4024 return true;
4025 }
4026
4027 /*
4028 * The intrinsic is a model of this pseudo-code:
4029 *
4030 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4031 * if (carrierThread != thread) { // is virtual thread
4032 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4033 * bool excluded = vthread_epoch_raw & excluded_mask;
4034 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4035 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
4036 * if (!excluded) {
4037 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4038 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
4039 * }
4040 * Atomic::release_store(&tl->_vthread, true);
4041 * return;
4042 * }
4043 * Atomic::release_store(&tl->_vthread, false);
4044 */
4045 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4046 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4047
4048 Node* input_memory_state = reset_memory();
4049 set_all_memory(input_memory_state);
4050
4051 // The most significant bit of the u2 is used to denote thread exclusion
4052 Node* excluded_mask = _gvn.intcon(1 << 15);
4053 // The epoch generation is the range [1-32767]
4054 Node* epoch_mask = _gvn.intcon(32767);
4055
4056 Node* const carrierThread = generate_current_thread(jt);
4057 // If thread != carrierThread, this is a virtual thread.
4058 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4059 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4060 IfNode* iff_thread_not_equal_carrierThread =
4061 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4062
4063 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4064
4065 // False branch, is carrierThread.
4066 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4067 // Store release
4068 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4069
4070 set_all_memory(input_memory_state);
4071
4072 // True branch, is virtual thread.
4073 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4074 set_control(thread_not_equal_carrierThread);
4075
4076 // Load the raw epoch value from the vthread.
4077 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4078 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4079 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4080
4081 // Mask off the excluded information from the epoch.
4082 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4083
4084 // Load the tid field from the thread.
4085 Node* tid = load_field_from_object(thread, "tid", "J");
4086
4087 // Store the vthread tid to the jfr thread local.
4088 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4089 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4090
4091 // Branch is_excluded to conditionalize updating the epoch .
4092 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4093 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4094 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4095
4096 // True branch, vthread is excluded, no need to write epoch info.
4097 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4098 set_control(excluded);
4099 Node* vthread_is_excluded = _gvn.intcon(1);
4100
4101 // False branch, vthread is included, update epoch info.
4102 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4103 set_control(included);
4104 Node* vthread_is_included = _gvn.intcon(0);
4105
4106 // Get epoch value.
4107 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4108
4109 // Store the vthread epoch to the jfr thread local.
4110 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4111 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4112
4113 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4114 record_for_igvn(excluded_rgn);
4115 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4116 record_for_igvn(excluded_mem);
4117 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4118 record_for_igvn(exclusion);
4119
4120 // Merge the excluded control and memory.
4121 excluded_rgn->init_req(_true_path, excluded);
4122 excluded_rgn->init_req(_false_path, included);
4123 excluded_mem->init_req(_true_path, tid_memory);
4124 excluded_mem->init_req(_false_path, included_memory);
4125 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4126 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4127
4128 // Set intermediate state.
4129 set_control(_gvn.transform(excluded_rgn));
4130 set_all_memory(excluded_mem);
4131
4132 // Store the vthread exclusion state to the jfr thread local.
4133 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4134 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4135
4136 // Store release
4137 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4138
4139 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4140 record_for_igvn(thread_compare_rgn);
4141 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4142 record_for_igvn(thread_compare_mem);
4143 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4144 record_for_igvn(vthread);
4145
4146 // Merge the thread_compare control and memory.
4147 thread_compare_rgn->init_req(_true_path, control());
4148 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4149 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4150 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4151
4152 // Set output state.
4153 set_control(_gvn.transform(thread_compare_rgn));
4154 set_all_memory(_gvn.transform(thread_compare_mem));
4155 }
4156
4157 #endif // JFR_HAVE_INTRINSICS
4158
4159 //------------------------inline_native_currentCarrierThread------------------
4160 bool LibraryCallKit::inline_native_currentCarrierThread() {
4161 Node* junk = nullptr;
4162 set_result(generate_current_thread(junk));
4163 return true;
4164 }
4165
4166 //------------------------inline_native_currentThread------------------
4167 bool LibraryCallKit::inline_native_currentThread() {
4168 Node* junk = nullptr;
4169 set_result(generate_virtual_thread(junk));
4170 return true;
4171 }
4172
4173 //------------------------inline_native_setVthread------------------
4174 bool LibraryCallKit::inline_native_setCurrentThread() {
4175 assert(C->method()->changes_current_thread(),
4176 "method changes current Thread but is not annotated ChangesCurrentThread");
4177 Node* arr = argument(1);
4178 Node* thread = _gvn.transform(new ThreadLocalNode());
4179 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4180 Node* thread_obj_handle
4181 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4182 thread_obj_handle = _gvn.transform(thread_obj_handle);
4183 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4184 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4185
4186 // Change the _monitor_owner_id of the JavaThread
4187 Node* tid = load_field_from_object(arr, "tid", "J");
4188 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4189 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4190
4191 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4192 return true;
4193 }
4194
4195 const Type* LibraryCallKit::scopedValueCache_type() {
4196 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4197 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4198 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
4199
4200 // Because we create the scopedValue cache lazily we have to make the
4201 // type of the result BotPTR.
4202 bool xk = etype->klass_is_exact();
4203 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4204 return objects_type;
4205 }
4206
4207 Node* LibraryCallKit::scopedValueCache_helper() {
4208 Node* thread = _gvn.transform(new ThreadLocalNode());
4209 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4210 // We cannot use immutable_memory() because we might flip onto a
4211 // different carrier thread, at which point we'll need to use that
4212 // carrier thread's cache.
4213 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4214 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4215 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4216 }
4217
4218 //------------------------inline_native_scopedValueCache------------------
4219 bool LibraryCallKit::inline_native_scopedValueCache() {
4220 Node* cache_obj_handle = scopedValueCache_helper();
4221 const Type* objects_type = scopedValueCache_type();
4222 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4223
4224 return true;
4225 }
4226
4227 //------------------------inline_native_setScopedValueCache------------------
4228 bool LibraryCallKit::inline_native_setScopedValueCache() {
4229 Node* arr = argument(0);
4230 Node* cache_obj_handle = scopedValueCache_helper();
4231 const Type* objects_type = scopedValueCache_type();
4232
4233 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4234 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4235
4236 return true;
4237 }
4238
4239 //------------------------inline_native_Continuation_pin and unpin-----------
4240
4241 // Shared implementation routine for both pin and unpin.
4242 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4243 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4244
4245 // Save input memory.
4246 Node* input_memory_state = reset_memory();
4247 set_all_memory(input_memory_state);
4248
4249 // TLS
4250 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4251 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4252 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4253
4254 // Null check the last continuation object.
4255 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4256 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4257 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4258
4259 // False path, last continuation is null.
4260 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4261
4262 // True path, last continuation is not null.
4263 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4264
4265 set_control(continuation_is_not_null);
4266
4267 // Load the pin count from the last continuation.
4268 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4269 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4270
4271 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4272 Node* pin_count_rhs;
4273 if (unpin) {
4274 pin_count_rhs = _gvn.intcon(0);
4275 } else {
4276 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4277 }
4278 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4279 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4280 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4281
4282 // True branch, pin count over/underflow.
4283 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4284 {
4285 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4286 // which will throw IllegalStateException for pin count over/underflow.
4287 // No memory changed so far - we can use memory create by reset_memory()
4288 // at the beginning of this intrinsic. No need to call reset_memory() again.
4289 PreserveJVMState pjvms(this);
4290 set_control(pin_count_over_underflow);
4291 uncommon_trap(Deoptimization::Reason_intrinsic,
4292 Deoptimization::Action_none);
4293 assert(stopped(), "invariant");
4294 }
4295
4296 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4297 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4298 set_control(valid_pin_count);
4299
4300 Node* next_pin_count;
4301 if (unpin) {
4302 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4303 } else {
4304 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4305 }
4306
4307 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4308
4309 // Result of top level CFG and Memory.
4310 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4311 record_for_igvn(result_rgn);
4312 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4313 record_for_igvn(result_mem);
4314
4315 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4316 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4317 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4318 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4319
4320 // Set output state.
4321 set_control(_gvn.transform(result_rgn));
4322 set_all_memory(_gvn.transform(result_mem));
4323
4324 return true;
4325 }
4326
4327 //-----------------------load_klass_from_mirror_common-------------------------
4328 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4329 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4330 // and branch to the given path on the region.
4331 // If never_see_null, take an uncommon trap on null, so we can optimistically
4332 // compile for the non-null case.
4333 // If the region is null, force never_see_null = true.
4334 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4335 bool never_see_null,
4336 RegionNode* region,
4337 int null_path,
4338 int offset) {
4339 if (region == nullptr) never_see_null = true;
4340 Node* p = basic_plus_adr(mirror, offset);
4341 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4342 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4343 Node* null_ctl = top();
4344 kls = null_check_oop(kls, &null_ctl, never_see_null);
4345 if (region != nullptr) {
4346 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4347 region->init_req(null_path, null_ctl);
4348 } else {
4349 assert(null_ctl == top(), "no loose ends");
4350 }
4351 return kls;
4352 }
4353
4354 //--------------------(inline_native_Class_query helpers)---------------------
4355 // Use this for JVM_ACC_INTERFACE.
4356 // Fall through if (mods & mask) == bits, take the guard otherwise.
4357 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4358 ByteSize offset, const Type* type, BasicType bt) {
4359 // Branch around if the given klass has the given modifier bit set.
4360 // Like generate_guard, adds a new path onto the region.
4361 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4362 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4363 Node* mask = intcon(modifier_mask);
4364 Node* bits = intcon(modifier_bits);
4365 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4366 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4367 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4368 return generate_fair_guard(bol, region);
4369 }
4370
4371 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4372 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4373 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4374 }
4375
4376 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4377 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4378 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4379 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4380 }
4381
4382 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4383 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4384 }
4385
4386 //-------------------------inline_native_Class_query-------------------
4387 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4388 const Type* return_type = TypeInt::BOOL;
4389 Node* prim_return_value = top(); // what happens if it's a primitive class?
4390 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4391 bool expect_prim = false; // most of these guys expect to work on refs
4392
4393 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4394
4395 Node* mirror = argument(0);
4396 Node* obj = top();
4397
4398 switch (id) {
4399 case vmIntrinsics::_isInstance:
4400 // nothing is an instance of a primitive type
4401 prim_return_value = intcon(0);
4402 obj = argument(1);
4403 break;
4404 case vmIntrinsics::_isHidden:
4405 prim_return_value = intcon(0);
4406 break;
4407 case vmIntrinsics::_getSuperclass:
4408 prim_return_value = null();
4409 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4410 break;
4411 default:
4412 fatal_unexpected_iid(id);
4413 break;
4414 }
4415
4416 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4417 if (mirror_con == nullptr) return false; // cannot happen?
4418
4419 #ifndef PRODUCT
4420 if (C->print_intrinsics() || C->print_inlining()) {
4421 ciType* k = mirror_con->java_mirror_type();
4422 if (k) {
4423 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4424 k->print_name();
4425 tty->cr();
4426 }
4427 }
4428 #endif
4429
4430 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4431 RegionNode* region = new RegionNode(PATH_LIMIT);
4432 record_for_igvn(region);
4433 PhiNode* phi = new PhiNode(region, return_type);
4434
4435 // The mirror will never be null of Reflection.getClassAccessFlags, however
4436 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4437 // if it is. See bug 4774291.
4438
4439 // For Reflection.getClassAccessFlags(), the null check occurs in
4440 // the wrong place; see inline_unsafe_access(), above, for a similar
4441 // situation.
4442 mirror = null_check(mirror);
4443 // If mirror or obj is dead, only null-path is taken.
4444 if (stopped()) return true;
4445
4446 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4447
4448 // Now load the mirror's klass metaobject, and null-check it.
4449 // Side-effects region with the control path if the klass is null.
4450 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4451 // If kls is null, we have a primitive mirror.
4452 phi->init_req(_prim_path, prim_return_value);
4453 if (stopped()) { set_result(region, phi); return true; }
4454 bool safe_for_replace = (region->in(_prim_path) == top());
4455
4456 Node* p; // handy temp
4457 Node* null_ctl;
4458
4459 // Now that we have the non-null klass, we can perform the real query.
4460 // For constant classes, the query will constant-fold in LoadNode::Value.
4461 Node* query_value = top();
4462 switch (id) {
4463 case vmIntrinsics::_isInstance:
4464 // nothing is an instance of a primitive type
4465 query_value = gen_instanceof(obj, kls, safe_for_replace);
4466 break;
4467
4468 case vmIntrinsics::_isHidden:
4469 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4470 if (generate_hidden_class_guard(kls, region) != nullptr)
4471 // A guard was added. If the guard is taken, it was an hidden class.
4472 phi->add_req(intcon(1));
4473 // If we fall through, it's a plain class.
4474 query_value = intcon(0);
4475 break;
4476
4477
4478 case vmIntrinsics::_getSuperclass:
4479 // The rules here are somewhat unfortunate, but we can still do better
4480 // with random logic than with a JNI call.
4481 // Interfaces store null or Object as _super, but must report null.
4482 // Arrays store an intermediate super as _super, but must report Object.
4483 // Other types can report the actual _super.
4484 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4485 if (generate_interface_guard(kls, region) != nullptr)
4486 // A guard was added. If the guard is taken, it was an interface.
4487 phi->add_req(null());
4488 if (generate_array_guard(kls, region) != nullptr)
4489 // A guard was added. If the guard is taken, it was an array.
4490 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4491 // If we fall through, it's a plain class. Get its _super.
4492 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4493 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4494 null_ctl = top();
4495 kls = null_check_oop(kls, &null_ctl);
4496 if (null_ctl != top()) {
4497 // If the guard is taken, Object.superClass is null (both klass and mirror).
4498 region->add_req(null_ctl);
4499 phi ->add_req(null());
4500 }
4501 if (!stopped()) {
4502 query_value = load_mirror_from_klass(kls);
4503 }
4504 break;
4505
4506 default:
4507 fatal_unexpected_iid(id);
4508 break;
4509 }
4510
4511 // Fall-through is the normal case of a query to a real class.
4512 phi->init_req(1, query_value);
4513 region->init_req(1, control());
4514
4515 C->set_has_split_ifs(true); // Has chance for split-if optimization
4516 set_result(region, phi);
4517 return true;
4518 }
4519
4520
4521 //-------------------------inline_Class_cast-------------------
4522 bool LibraryCallKit::inline_Class_cast() {
4523 Node* mirror = argument(0); // Class
4524 Node* obj = argument(1);
4525 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4526 if (mirror_con == nullptr) {
4527 return false; // dead path (mirror->is_top()).
4528 }
4529 if (obj == nullptr || obj->is_top()) {
4530 return false; // dead path
4531 }
4532 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4533
4534 // First, see if Class.cast() can be folded statically.
4535 // java_mirror_type() returns non-null for compile-time Class constants.
4536 ciType* tm = mirror_con->java_mirror_type();
4537 if (tm != nullptr && tm->is_klass() &&
4538 tp != nullptr) {
4539 if (!tp->is_loaded()) {
4540 // Don't use intrinsic when class is not loaded.
4541 return false;
4542 } else {
4543 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4544 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4545 if (static_res == Compile::SSC_always_true) {
4546 // isInstance() is true - fold the code.
4547 set_result(obj);
4548 return true;
4549 } else if (static_res == Compile::SSC_always_false) {
4550 // Don't use intrinsic, have to throw ClassCastException.
4551 // If the reference is null, the non-intrinsic bytecode will
4552 // be optimized appropriately.
4553 return false;
4554 }
4555 }
4556 }
4557
4558 // Bailout intrinsic and do normal inlining if exception path is frequent.
4559 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4560 return false;
4561 }
4562
4563 // Generate dynamic checks.
4564 // Class.cast() is java implementation of _checkcast bytecode.
4565 // Do checkcast (Parse::do_checkcast()) optimizations here.
4566
4567 mirror = null_check(mirror);
4568 // If mirror is dead, only null-path is taken.
4569 if (stopped()) {
4570 return true;
4571 }
4572
4573 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4574 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4575 RegionNode* region = new RegionNode(PATH_LIMIT);
4576 record_for_igvn(region);
4577
4578 // Now load the mirror's klass metaobject, and null-check it.
4579 // If kls is null, we have a primitive mirror and
4580 // nothing is an instance of a primitive type.
4581 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4582
4583 Node* res = top();
4584 Node* io = i_o();
4585 Node* mem = merged_memory();
4586 if (!stopped()) {
4587
4588 Node* bad_type_ctrl = top();
4589 // Do checkcast optimizations.
4590 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4591 region->init_req(_bad_type_path, bad_type_ctrl);
4592 }
4593 if (region->in(_prim_path) != top() ||
4594 region->in(_bad_type_path) != top() ||
4595 region->in(_npe_path) != top()) {
4596 // Let Interpreter throw ClassCastException.
4597 PreserveJVMState pjvms(this);
4598 set_control(_gvn.transform(region));
4599 // Set IO and memory because gen_checkcast may override them when buffering inline types
4600 set_i_o(io);
4601 set_all_memory(mem);
4602 uncommon_trap(Deoptimization::Reason_intrinsic,
4603 Deoptimization::Action_maybe_recompile);
4604 }
4605 if (!stopped()) {
4606 set_result(res);
4607 }
4608 return true;
4609 }
4610
4611
4612 //--------------------------inline_native_subtype_check------------------------
4613 // This intrinsic takes the JNI calls out of the heart of
4614 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4615 bool LibraryCallKit::inline_native_subtype_check() {
4616 // Pull both arguments off the stack.
4617 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4618 args[0] = argument(0);
4619 args[1] = argument(1);
4620 Node* klasses[2]; // corresponding Klasses: superk, subk
4621 klasses[0] = klasses[1] = top();
4622
4623 enum {
4624 // A full decision tree on {superc is prim, subc is prim}:
4625 _prim_0_path = 1, // {P,N} => false
4626 // {P,P} & superc!=subc => false
4627 _prim_same_path, // {P,P} & superc==subc => true
4628 _prim_1_path, // {N,P} => false
4629 _ref_subtype_path, // {N,N} & subtype check wins => true
4630 _both_ref_path, // {N,N} & subtype check loses => false
4631 PATH_LIMIT
4632 };
4633
4634 RegionNode* region = new RegionNode(PATH_LIMIT);
4635 RegionNode* prim_region = new RegionNode(2);
4636 Node* phi = new PhiNode(region, TypeInt::BOOL);
4637 record_for_igvn(region);
4638 record_for_igvn(prim_region);
4639
4640 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4641 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4642 int class_klass_offset = java_lang_Class::klass_offset();
4643
4644 // First null-check both mirrors and load each mirror's klass metaobject.
4645 int which_arg;
4646 for (which_arg = 0; which_arg <= 1; which_arg++) {
4647 Node* arg = args[which_arg];
4648 arg = null_check(arg);
4649 if (stopped()) break;
4650 args[which_arg] = arg;
4651
4652 Node* p = basic_plus_adr(arg, class_klass_offset);
4653 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4654 klasses[which_arg] = _gvn.transform(kls);
4655 }
4656
4657 // Having loaded both klasses, test each for null.
4658 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4659 for (which_arg = 0; which_arg <= 1; which_arg++) {
4660 Node* kls = klasses[which_arg];
4661 Node* null_ctl = top();
4662 kls = null_check_oop(kls, &null_ctl, never_see_null);
4663 if (which_arg == 0) {
4664 prim_region->init_req(1, null_ctl);
4665 } else {
4666 region->init_req(_prim_1_path, null_ctl);
4667 }
4668 if (stopped()) break;
4669 klasses[which_arg] = kls;
4670 }
4671
4672 if (!stopped()) {
4673 // now we have two reference types, in klasses[0..1]
4674 Node* subk = klasses[1]; // the argument to isAssignableFrom
4675 Node* superk = klasses[0]; // the receiver
4676 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4677 region->set_req(_ref_subtype_path, control());
4678 }
4679
4680 // If both operands are primitive (both klasses null), then
4681 // we must return true when they are identical primitives.
4682 // It is convenient to test this after the first null klass check.
4683 // This path is also used if superc is a value mirror.
4684 set_control(_gvn.transform(prim_region));
4685 if (!stopped()) {
4686 // Since superc is primitive, make a guard for the superc==subc case.
4687 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4688 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4689 generate_fair_guard(bol_eq, region);
4690 if (region->req() == PATH_LIMIT+1) {
4691 // A guard was added. If the added guard is taken, superc==subc.
4692 region->swap_edges(PATH_LIMIT, _prim_same_path);
4693 region->del_req(PATH_LIMIT);
4694 }
4695 region->set_req(_prim_0_path, control()); // Not equal after all.
4696 }
4697
4698 // these are the only paths that produce 'true':
4699 phi->set_req(_prim_same_path, intcon(1));
4700 phi->set_req(_ref_subtype_path, intcon(1));
4701
4702 // pull together the cases:
4703 assert(region->req() == PATH_LIMIT, "sane region");
4704 for (uint i = 1; i < region->req(); i++) {
4705 Node* ctl = region->in(i);
4706 if (ctl == nullptr || ctl == top()) {
4707 region->set_req(i, top());
4708 phi ->set_req(i, top());
4709 } else if (phi->in(i) == nullptr) {
4710 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4711 }
4712 }
4713
4714 set_control(_gvn.transform(region));
4715 set_result(_gvn.transform(phi));
4716 return true;
4717 }
4718
4719 //---------------------generate_array_guard_common------------------------
4720 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4721
4722 if (stopped()) {
4723 return nullptr;
4724 }
4725
4726 // Like generate_guard, adds a new path onto the region.
4727 jint layout_con = 0;
4728 Node* layout_val = get_layout_helper(kls, layout_con);
4729 if (layout_val == nullptr) {
4730 bool query = 0;
4731 switch(kind) {
4732 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4733 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4734 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4735 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4736 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4737 default:
4738 ShouldNotReachHere();
4739 }
4740 if (!query) {
4741 return nullptr; // never a branch
4742 } else { // always a branch
4743 Node* always_branch = control();
4744 if (region != nullptr)
4745 region->add_req(always_branch);
4746 set_control(top());
4747 return always_branch;
4748 }
4749 }
4750 unsigned int value = 0;
4751 BoolTest::mask btest = BoolTest::illegal;
4752 switch(kind) {
4753 case RefArray:
4754 case NonRefArray: {
4755 value = Klass::_lh_array_tag_ref_value;
4756 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4757 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4758 break;
4759 }
4760 case TypeArray: {
4761 value = Klass::_lh_array_tag_type_value;
4762 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4763 btest = BoolTest::eq;
4764 break;
4765 }
4766 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4767 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4768 default:
4769 ShouldNotReachHere();
4770 }
4771 // Now test the correct condition.
4772 jint nval = (jint)value;
4773 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4774 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4775 Node* ctrl = generate_fair_guard(bol, region);
4776 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4777 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4778 // Keep track of the fact that 'obj' is an array to prevent
4779 // array specific accesses from floating above the guard.
4780 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4781 }
4782 return ctrl;
4783 }
4784
4785 // public static native Object[] newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4786 // public static native Object[] newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4787 // public static native Object[] newNullableAtomicArray(Class<?> componentType, int length);
4788 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4789 assert(null_free || atomic, "nullable implies atomic");
4790 Node* componentType = argument(0);
4791 Node* length = argument(1);
4792 Node* init_val = null_free ? argument(2) : nullptr;
4793
4794 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4795 if (tp != nullptr) {
4796 ciInstanceKlass* ik = tp->instance_klass();
4797 if (ik == C->env()->Class_klass()) {
4798 ciType* t = tp->java_mirror_type();
4799 if (t != nullptr && t->is_inlinetype()) {
4800
4801 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4802 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4803 assert(array_klass->is_elem_atomic() == atomic, "inconsistency");
4804
4805 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4806 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4807 return false;
4808 }
4809
4810 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4811 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces, true);
4812 if (null_free) {
4813 if (init_val->is_InlineType()) {
4814 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4815 // Zeroing is enough because the init value is the all-zero value
4816 init_val = nullptr;
4817 } else {
4818 init_val = init_val->as_InlineType()->buffer(this);
4819 }
4820 }
4821 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4822 }
4823 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4824 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4825 assert(arytype->is_null_free() == null_free, "inconsistency");
4826 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4827 assert(arytype->is_atomic() == atomic, "inconsistency");
4828 set_result(obj);
4829 return true;
4830 }
4831 }
4832 }
4833 }
4834 return false;
4835 }
4836
4837 Node* LibraryCallKit::load_default_array_klass(Node* klass_node) {
4838 // TODO 8366668
4839 // - Fred suggested that we could just have the first entry in the refined list point to the array with ArrayKlass::ArrayProperties::DEFAULT property
4840 // For now, we just load from ObjArrayKlass::_next_refined_array_klass, which would always be the refKlass for non-values, and deopt if it's not
4841 // - Convert this to an IGVN optimization, so it's also folded after parsing
4842 // - The generate_typeArray_guard is not needed by all callers, double-check that it's folded
4843
4844 const Type* klass_t = _gvn.type(klass_node);
4845 const TypeAryKlassPtr* ary_klass_t = klass_t->isa_aryklassptr();
4846 if (ary_klass_t && ary_klass_t->klass_is_exact()) {
4847 if (ary_klass_t->exact_klass()->is_obj_array_klass()) {
4848 ary_klass_t = ary_klass_t->get_vm_type(false);
4849 return makecon(ary_klass_t);
4850 } else {
4851 return klass_node;
4852 }
4853 }
4854
4855 // Load next refined array klass if klass is an ObjArrayKlass
4856 RegionNode* refined_region = new RegionNode(2);
4857 Node* refined_phi = new PhiNode(refined_region, klass_t);
4858
4859 generate_typeArray_guard(klass_node, refined_region);
4860 if (refined_region->req() == 3) {
4861 refined_phi->add_req(klass_node);
4862 }
4863
4864 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4865 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4866
4867 RegionNode* refined_region2 = new RegionNode(3);
4868 Node* refined_phi2 = new PhiNode(refined_region2, klass_t);
4869
4870 Node* null_ctl = top();
4871 Node* null_free_klass = null_check_common(refined_klass, T_OBJECT, false, &null_ctl);
4872 refined_region2->init_req(1, null_ctl);
4873 refined_phi2->init_req(1, klass_node);
4874
4875 refined_region2->init_req(2, control());
4876 refined_phi2->init_req(2, null_free_klass);
4877
4878 set_control(_gvn.transform(refined_region2));
4879 refined_klass = _gvn.transform(refined_phi2);
4880
4881 Node* adr_properties = basic_plus_adr(refined_klass, in_bytes(ObjArrayKlass::properties_offset()));
4882
4883 Node* properties = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_properties, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
4884 Node* default_val = makecon(TypeInt::make(ArrayKlass::ArrayProperties::DEFAULT));
4885 Node* chk = _gvn.transform(new CmpINode(properties, default_val));
4886 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
4887
4888 { // Deoptimize if not the default property
4889 BuildCutout unless(this, tst, PROB_MAX);
4890 uncommon_trap_exact(Deoptimization::Reason_class_check, Deoptimization::Action_none);
4891 }
4892
4893 refined_region->init_req(1, control());
4894 refined_phi->init_req(1, refined_klass);
4895
4896 set_control(_gvn.transform(refined_region));
4897 klass_node = _gvn.transform(refined_phi);
4898
4899 return klass_node;
4900 }
4901
4902 //-----------------------inline_native_newArray--------------------------
4903 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
4904 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4905 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4906 Node* mirror;
4907 Node* count_val;
4908 if (uninitialized) {
4909 null_check_receiver();
4910 mirror = argument(1);
4911 count_val = argument(2);
4912 } else {
4913 mirror = argument(0);
4914 count_val = argument(1);
4915 }
4916
4917 mirror = null_check(mirror);
4918 // If mirror or obj is dead, only null-path is taken.
4919 if (stopped()) return true;
4920
4921 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4922 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4923 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4924 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4925 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4926
4927 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4928 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4929 result_reg, _slow_path);
4930 Node* normal_ctl = control();
4931 Node* no_array_ctl = result_reg->in(_slow_path);
4932
4933 // Generate code for the slow case. We make a call to newArray().
4934 set_control(no_array_ctl);
4935 if (!stopped()) {
4936 // Either the input type is void.class, or else the
4937 // array klass has not yet been cached. Either the
4938 // ensuing call will throw an exception, or else it
4939 // will cache the array klass for next time.
4940 PreserveJVMState pjvms(this);
4941 CallJavaNode* slow_call = nullptr;
4942 if (uninitialized) {
4943 // Generate optimized virtual call (holder class 'Unsafe' is final)
4944 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4945 } else {
4946 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4947 }
4948 Node* slow_result = set_results_for_java_call(slow_call);
4949 // this->control() comes from set_results_for_java_call
4950 result_reg->set_req(_slow_path, control());
4951 result_val->set_req(_slow_path, slow_result);
4952 result_io ->set_req(_slow_path, i_o());
4953 result_mem->set_req(_slow_path, reset_memory());
4954 }
4955
4956 set_control(normal_ctl);
4957 if (!stopped()) {
4958 // Normal case: The array type has been cached in the java.lang.Class.
4959 // The following call works fine even if the array type is polymorphic.
4960 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4961
4962 klass_node = load_default_array_klass(klass_node);
4963
4964 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4965 result_reg->init_req(_normal_path, control());
4966 result_val->init_req(_normal_path, obj);
4967 result_io ->init_req(_normal_path, i_o());
4968 result_mem->init_req(_normal_path, reset_memory());
4969
4970 if (uninitialized) {
4971 // Mark the allocation so that zeroing is skipped
4972 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4973 alloc->maybe_set_complete(&_gvn);
4974 }
4975 }
4976
4977 // Return the combined state.
4978 set_i_o( _gvn.transform(result_io) );
4979 set_all_memory( _gvn.transform(result_mem));
4980
4981 C->set_has_split_ifs(true); // Has chance for split-if optimization
4982 set_result(result_reg, result_val);
4983 return true;
4984 }
4985
4986 //----------------------inline_native_getLength--------------------------
4987 // public static native int java.lang.reflect.Array.getLength(Object array);
4988 bool LibraryCallKit::inline_native_getLength() {
4989 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4990
4991 Node* array = null_check(argument(0));
4992 // If array is dead, only null-path is taken.
4993 if (stopped()) return true;
4994
4995 // Deoptimize if it is a non-array.
4996 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4997
4998 if (non_array != nullptr) {
4999 PreserveJVMState pjvms(this);
5000 set_control(non_array);
5001 uncommon_trap(Deoptimization::Reason_intrinsic,
5002 Deoptimization::Action_maybe_recompile);
5003 }
5004
5005 // If control is dead, only non-array-path is taken.
5006 if (stopped()) return true;
5007
5008 // The works fine even if the array type is polymorphic.
5009 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5010 Node* result = load_array_length(array);
5011
5012 C->set_has_split_ifs(true); // Has chance for split-if optimization
5013 set_result(result);
5014 return true;
5015 }
5016
5017 //------------------------inline_array_copyOf----------------------------
5018 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5019 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5020 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5021 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5022
5023 // Get the arguments.
5024 Node* original = argument(0);
5025 Node* start = is_copyOfRange? argument(1): intcon(0);
5026 Node* end = is_copyOfRange? argument(2): argument(1);
5027 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5028
5029 Node* newcopy = nullptr;
5030
5031 // Set the original stack and the reexecute bit for the interpreter to reexecute
5032 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5033 { PreserveReexecuteState preexecs(this);
5034 jvms()->set_should_reexecute(true);
5035
5036 array_type_mirror = null_check(array_type_mirror);
5037 original = null_check(original);
5038
5039 // Check if a null path was taken unconditionally.
5040 if (stopped()) return true;
5041
5042 Node* orig_length = load_array_length(original);
5043
5044 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5045 klass_node = null_check(klass_node);
5046
5047 RegionNode* bailout = new RegionNode(1);
5048 record_for_igvn(bailout);
5049
5050 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5051 // Bail out if that is so.
5052 // Inline type array may have object field that would require a
5053 // write barrier. Conservatively, go to slow path.
5054 // TODO 8251971: Optimize for the case when flat src/dst are later found
5055 // to not contain oops (i.e., move this check to the macro expansion phase).
5056 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5057 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5058 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5059 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5060 // Can src array be flat and contain oops?
5061 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5062 // Can dest array be flat and contain oops?
5063 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5064 // TODO 8366668 generate_non_refArray_guard also passed for ref arrays??
5065 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5066
5067 klass_node = load_default_array_klass(klass_node);
5068
5069 if (not_objArray != nullptr) {
5070 // Improve the klass node's type from the new optimistic assumption:
5071 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5072 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
5073 Node* cast = new CastPPNode(control(), klass_node, akls);
5074 klass_node = _gvn.transform(cast);
5075 }
5076
5077 // Bail out if either start or end is negative.
5078 generate_negative_guard(start, bailout, &start);
5079 generate_negative_guard(end, bailout, &end);
5080
5081 Node* length = end;
5082 if (_gvn.type(start) != TypeInt::ZERO) {
5083 length = _gvn.transform(new SubINode(end, start));
5084 }
5085
5086 // Bail out if length is negative (i.e., if start > end).
5087 // Without this the new_array would throw
5088 // NegativeArraySizeException but IllegalArgumentException is what
5089 // should be thrown
5090 generate_negative_guard(length, bailout, &length);
5091
5092 // Handle inline type arrays
5093 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5094 if (!stopped()) {
5095 // TODO JDK-8329224
5096 if (!orig_t->is_null_free()) {
5097 // Not statically known to be null free, add a check
5098 generate_fair_guard(null_free_array_test(original), bailout);
5099 }
5100 orig_t = _gvn.type(original)->isa_aryptr();
5101 if (orig_t != nullptr && orig_t->is_flat()) {
5102 // Src is flat, check that dest is flat as well
5103 if (exclude_flat) {
5104 // Dest can't be flat, bail out
5105 bailout->add_req(control());
5106 set_control(top());
5107 } else {
5108 generate_fair_guard(flat_array_test(klass_node, /* flat = */ false), bailout);
5109 }
5110 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5111 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5112 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5113 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5114 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5115 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5116 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5117 if (orig_t != nullptr) {
5118 orig_t = orig_t->cast_to_not_flat();
5119 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5120 }
5121 }
5122 if (!can_validate) {
5123 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5124 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5125 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5126 generate_fair_guard(flat_array_test(klass_node), bailout);
5127 generate_fair_guard(null_free_array_test(original), bailout);
5128 }
5129 }
5130
5131 // Bail out if start is larger than the original length
5132 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5133 generate_negative_guard(orig_tail, bailout, &orig_tail);
5134
5135 if (bailout->req() > 1) {
5136 PreserveJVMState pjvms(this);
5137 set_control(_gvn.transform(bailout));
5138 uncommon_trap(Deoptimization::Reason_intrinsic,
5139 Deoptimization::Action_maybe_recompile);
5140 }
5141
5142 if (!stopped()) {
5143 // How many elements will we copy from the original?
5144 // The answer is MinI(orig_tail, length).
5145 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5146
5147 // Generate a direct call to the right arraycopy function(s).
5148 // We know the copy is disjoint but we might not know if the
5149 // oop stores need checking.
5150 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5151 // This will fail a store-check if x contains any non-nulls.
5152
5153 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5154 // loads/stores but it is legal only if we're sure the
5155 // Arrays.copyOf would succeed. So we need all input arguments
5156 // to the copyOf to be validated, including that the copy to the
5157 // new array won't trigger an ArrayStoreException. That subtype
5158 // check can be optimized if we know something on the type of
5159 // the input array from type speculation.
5160 if (_gvn.type(klass_node)->singleton()) {
5161 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5162 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5163
5164 int test = C->static_subtype_check(superk, subk);
5165 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5166 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5167 if (t_original->speculative_type() != nullptr) {
5168 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5169 }
5170 }
5171 }
5172
5173 bool validated = false;
5174 // Reason_class_check rather than Reason_intrinsic because we
5175 // want to intrinsify even if this traps.
5176 if (can_validate) {
5177 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5178
5179 if (not_subtype_ctrl != top()) {
5180 PreserveJVMState pjvms(this);
5181 set_control(not_subtype_ctrl);
5182 uncommon_trap(Deoptimization::Reason_class_check,
5183 Deoptimization::Action_make_not_entrant);
5184 assert(stopped(), "Should be stopped");
5185 }
5186 validated = true;
5187 }
5188
5189 if (!stopped()) {
5190 newcopy = new_array(klass_node, length, 0); // no arguments to push
5191
5192 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5193 load_object_klass(original), klass_node);
5194 if (!is_copyOfRange) {
5195 ac->set_copyof(validated);
5196 } else {
5197 ac->set_copyofrange(validated);
5198 }
5199 Node* n = _gvn.transform(ac);
5200 if (n == ac) {
5201 ac->connect_outputs(this);
5202 } else {
5203 assert(validated, "shouldn't transform if all arguments not validated");
5204 set_all_memory(n);
5205 }
5206 }
5207 }
5208 } // original reexecute is set back here
5209
5210 C->set_has_split_ifs(true); // Has chance for split-if optimization
5211 if (!stopped()) {
5212 set_result(newcopy);
5213 }
5214 return true;
5215 }
5216
5217
5218 //----------------------generate_virtual_guard---------------------------
5219 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5220 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5221 RegionNode* slow_region) {
5222 ciMethod* method = callee();
5223 int vtable_index = method->vtable_index();
5224 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5225 "bad index %d", vtable_index);
5226 // Get the Method* out of the appropriate vtable entry.
5227 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5228 vtable_index*vtableEntry::size_in_bytes() +
5229 in_bytes(vtableEntry::method_offset());
5230 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5231 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5232
5233 // Compare the target method with the expected method (e.g., Object.hashCode).
5234 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5235
5236 Node* native_call = makecon(native_call_addr);
5237 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5238 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5239
5240 return generate_slow_guard(test_native, slow_region);
5241 }
5242
5243 //-----------------------generate_method_call----------------------------
5244 // Use generate_method_call to make a slow-call to the real
5245 // method if the fast path fails. An alternative would be to
5246 // use a stub like OptoRuntime::slow_arraycopy_Java.
5247 // This only works for expanding the current library call,
5248 // not another intrinsic. (E.g., don't use this for making an
5249 // arraycopy call inside of the copyOf intrinsic.)
5250 CallJavaNode*
5251 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5252 // When compiling the intrinsic method itself, do not use this technique.
5253 guarantee(callee() != C->method(), "cannot make slow-call to self");
5254
5255 ciMethod* method = callee();
5256 // ensure the JVMS we have will be correct for this call
5257 guarantee(method_id == method->intrinsic_id(), "must match");
5258
5259 const TypeFunc* tf = TypeFunc::make(method);
5260 if (res_not_null) {
5261 assert(tf->return_type() == T_OBJECT, "");
5262 const TypeTuple* range = tf->range_cc();
5263 const Type** fields = TypeTuple::fields(range->cnt());
5264 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5265 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5266 tf = TypeFunc::make(tf->domain_cc(), new_range);
5267 }
5268 CallJavaNode* slow_call;
5269 if (is_static) {
5270 assert(!is_virtual, "");
5271 slow_call = new CallStaticJavaNode(C, tf,
5272 SharedRuntime::get_resolve_static_call_stub(), method);
5273 } else if (is_virtual) {
5274 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5275 int vtable_index = Method::invalid_vtable_index;
5276 if (UseInlineCaches) {
5277 // Suppress the vtable call
5278 } else {
5279 // hashCode and clone are not a miranda methods,
5280 // so the vtable index is fixed.
5281 // No need to use the linkResolver to get it.
5282 vtable_index = method->vtable_index();
5283 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5284 "bad index %d", vtable_index);
5285 }
5286 slow_call = new CallDynamicJavaNode(tf,
5287 SharedRuntime::get_resolve_virtual_call_stub(),
5288 method, vtable_index);
5289 } else { // neither virtual nor static: opt_virtual
5290 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5291 slow_call = new CallStaticJavaNode(C, tf,
5292 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5293 slow_call->set_optimized_virtual(true);
5294 }
5295 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5296 // To be able to issue a direct call (optimized virtual or virtual)
5297 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5298 // about the method being invoked should be attached to the call site to
5299 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5300 slow_call->set_override_symbolic_info(true);
5301 }
5302 set_arguments_for_java_call(slow_call);
5303 set_edges_for_java_call(slow_call);
5304 return slow_call;
5305 }
5306
5307
5308 /**
5309 * Build special case code for calls to hashCode on an object. This call may
5310 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5311 * slightly different code.
5312 */
5313 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5314 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5315 assert(!(is_virtual && is_static), "either virtual, special, or static");
5316
5317 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5318
5319 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5320 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5321 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5322 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5323 Node* obj = argument(0);
5324
5325 // Don't intrinsify hashcode on inline types for now.
5326 // The "is locked" runtime check below also serves as inline type check and goes to the slow path.
5327 if (gvn().type(obj)->is_inlinetypeptr()) {
5328 return false;
5329 }
5330
5331 if (!is_static) {
5332 // Check for hashing null object
5333 obj = null_check_receiver();
5334 if (stopped()) return true; // unconditionally null
5335 result_reg->init_req(_null_path, top());
5336 result_val->init_req(_null_path, top());
5337 } else {
5338 // Do a null check, and return zero if null.
5339 // System.identityHashCode(null) == 0
5340 Node* null_ctl = top();
5341 obj = null_check_oop(obj, &null_ctl);
5342 result_reg->init_req(_null_path, null_ctl);
5343 result_val->init_req(_null_path, _gvn.intcon(0));
5344 }
5345
5346 // Unconditionally null? Then return right away.
5347 if (stopped()) {
5348 set_control( result_reg->in(_null_path));
5349 if (!stopped())
5350 set_result(result_val->in(_null_path));
5351 return true;
5352 }
5353
5354 // We only go to the fast case code if we pass a number of guards. The
5355 // paths which do not pass are accumulated in the slow_region.
5356 RegionNode* slow_region = new RegionNode(1);
5357 record_for_igvn(slow_region);
5358
5359 // If this is a virtual call, we generate a funny guard. We pull out
5360 // the vtable entry corresponding to hashCode() from the target object.
5361 // If the target method which we are calling happens to be the native
5362 // Object hashCode() method, we pass the guard. We do not need this
5363 // guard for non-virtual calls -- the caller is known to be the native
5364 // Object hashCode().
5365 if (is_virtual) {
5366 // After null check, get the object's klass.
5367 Node* obj_klass = load_object_klass(obj);
5368 generate_virtual_guard(obj_klass, slow_region);
5369 }
5370
5371 // Get the header out of the object, use LoadMarkNode when available
5372 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5373 // The control of the load must be null. Otherwise, the load can move before
5374 // the null check after castPP removal.
5375 Node* no_ctrl = nullptr;
5376 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5377
5378 if (!UseObjectMonitorTable) {
5379 // Test the header to see if it is safe to read w.r.t. locking.
5380 // This also serves as guard against inline types
5381 Node *lock_mask = _gvn.MakeConX(markWord::inline_type_mask_in_place);
5382 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5383 if (LockingMode == LM_LIGHTWEIGHT) {
5384 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5385 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5386 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5387
5388 generate_slow_guard(test_monitor, slow_region);
5389 } else {
5390 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
5391 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
5392 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
5393
5394 generate_slow_guard(test_not_unlocked, slow_region);
5395 }
5396 }
5397
5398 // Get the hash value and check to see that it has been properly assigned.
5399 // We depend on hash_mask being at most 32 bits and avoid the use of
5400 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5401 // vm: see markWord.hpp.
5402 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5403 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5404 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5405 // This hack lets the hash bits live anywhere in the mark object now, as long
5406 // as the shift drops the relevant bits into the low 32 bits. Note that
5407 // Java spec says that HashCode is an int so there's no point in capturing
5408 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5409 hshifted_header = ConvX2I(hshifted_header);
5410 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5411
5412 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5413 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5414 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5415
5416 generate_slow_guard(test_assigned, slow_region);
5417
5418 Node* init_mem = reset_memory();
5419 // fill in the rest of the null path:
5420 result_io ->init_req(_null_path, i_o());
5421 result_mem->init_req(_null_path, init_mem);
5422
5423 result_val->init_req(_fast_path, hash_val);
5424 result_reg->init_req(_fast_path, control());
5425 result_io ->init_req(_fast_path, i_o());
5426 result_mem->init_req(_fast_path, init_mem);
5427
5428 // Generate code for the slow case. We make a call to hashCode().
5429 set_control(_gvn.transform(slow_region));
5430 if (!stopped()) {
5431 // No need for PreserveJVMState, because we're using up the present state.
5432 set_all_memory(init_mem);
5433 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5434 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5435 Node* slow_result = set_results_for_java_call(slow_call);
5436 // this->control() comes from set_results_for_java_call
5437 result_reg->init_req(_slow_path, control());
5438 result_val->init_req(_slow_path, slow_result);
5439 result_io ->set_req(_slow_path, i_o());
5440 result_mem ->set_req(_slow_path, reset_memory());
5441 }
5442
5443 // Return the combined state.
5444 set_i_o( _gvn.transform(result_io) );
5445 set_all_memory( _gvn.transform(result_mem));
5446
5447 set_result(result_reg, result_val);
5448 return true;
5449 }
5450
5451 //---------------------------inline_native_getClass----------------------------
5452 // public final native Class<?> java.lang.Object.getClass();
5453 //
5454 // Build special case code for calls to getClass on an object.
5455 bool LibraryCallKit::inline_native_getClass() {
5456 Node* obj = argument(0);
5457 if (obj->is_InlineType()) {
5458 const Type* t = _gvn.type(obj);
5459 if (t->maybe_null()) {
5460 null_check(obj);
5461 }
5462 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5463 return true;
5464 }
5465 obj = null_check_receiver();
5466 if (stopped()) return true;
5467 set_result(load_mirror_from_klass(load_object_klass(obj)));
5468 return true;
5469 }
5470
5471 //-----------------inline_native_Reflection_getCallerClass---------------------
5472 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5473 //
5474 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5475 //
5476 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5477 // in that it must skip particular security frames and checks for
5478 // caller sensitive methods.
5479 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5480 #ifndef PRODUCT
5481 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5482 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5483 }
5484 #endif
5485
5486 if (!jvms()->has_method()) {
5487 #ifndef PRODUCT
5488 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5489 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5490 }
5491 #endif
5492 return false;
5493 }
5494
5495 // Walk back up the JVM state to find the caller at the required
5496 // depth.
5497 JVMState* caller_jvms = jvms();
5498
5499 // Cf. JVM_GetCallerClass
5500 // NOTE: Start the loop at depth 1 because the current JVM state does
5501 // not include the Reflection.getCallerClass() frame.
5502 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5503 ciMethod* m = caller_jvms->method();
5504 switch (n) {
5505 case 0:
5506 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5507 break;
5508 case 1:
5509 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5510 if (!m->caller_sensitive()) {
5511 #ifndef PRODUCT
5512 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5513 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5514 }
5515 #endif
5516 return false; // bail-out; let JVM_GetCallerClass do the work
5517 }
5518 break;
5519 default:
5520 if (!m->is_ignored_by_security_stack_walk()) {
5521 // We have reached the desired frame; return the holder class.
5522 // Acquire method holder as java.lang.Class and push as constant.
5523 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5524 ciInstance* caller_mirror = caller_klass->java_mirror();
5525 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5526
5527 #ifndef PRODUCT
5528 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5529 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());
5530 tty->print_cr(" JVM state at this point:");
5531 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5532 ciMethod* m = jvms()->of_depth(i)->method();
5533 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5534 }
5535 }
5536 #endif
5537 return true;
5538 }
5539 break;
5540 }
5541 }
5542
5543 #ifndef PRODUCT
5544 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5545 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5546 tty->print_cr(" JVM state at this point:");
5547 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5548 ciMethod* m = jvms()->of_depth(i)->method();
5549 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5550 }
5551 }
5552 #endif
5553
5554 return false; // bail-out; let JVM_GetCallerClass do the work
5555 }
5556
5557 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5558 Node* arg = argument(0);
5559 Node* result = nullptr;
5560
5561 switch (id) {
5562 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5563 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5564 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5565 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5566 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5567 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5568
5569 case vmIntrinsics::_doubleToLongBits: {
5570 // two paths (plus control) merge in a wood
5571 RegionNode *r = new RegionNode(3);
5572 Node *phi = new PhiNode(r, TypeLong::LONG);
5573
5574 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5575 // Build the boolean node
5576 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5577
5578 // Branch either way.
5579 // NaN case is less traveled, which makes all the difference.
5580 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5581 Node *opt_isnan = _gvn.transform(ifisnan);
5582 assert( opt_isnan->is_If(), "Expect an IfNode");
5583 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5584 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5585
5586 set_control(iftrue);
5587
5588 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5589 Node *slow_result = longcon(nan_bits); // return NaN
5590 phi->init_req(1, _gvn.transform( slow_result ));
5591 r->init_req(1, iftrue);
5592
5593 // Else fall through
5594 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5595 set_control(iffalse);
5596
5597 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5598 r->init_req(2, iffalse);
5599
5600 // Post merge
5601 set_control(_gvn.transform(r));
5602 record_for_igvn(r);
5603
5604 C->set_has_split_ifs(true); // Has chance for split-if optimization
5605 result = phi;
5606 assert(result->bottom_type()->isa_long(), "must be");
5607 break;
5608 }
5609
5610 case vmIntrinsics::_floatToIntBits: {
5611 // two paths (plus control) merge in a wood
5612 RegionNode *r = new RegionNode(3);
5613 Node *phi = new PhiNode(r, TypeInt::INT);
5614
5615 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5616 // Build the boolean node
5617 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5618
5619 // Branch either way.
5620 // NaN case is less traveled, which makes all the difference.
5621 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5622 Node *opt_isnan = _gvn.transform(ifisnan);
5623 assert( opt_isnan->is_If(), "Expect an IfNode");
5624 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5625 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5626
5627 set_control(iftrue);
5628
5629 static const jint nan_bits = 0x7fc00000;
5630 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5631 phi->init_req(1, _gvn.transform( slow_result ));
5632 r->init_req(1, iftrue);
5633
5634 // Else fall through
5635 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5636 set_control(iffalse);
5637
5638 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5639 r->init_req(2, iffalse);
5640
5641 // Post merge
5642 set_control(_gvn.transform(r));
5643 record_for_igvn(r);
5644
5645 C->set_has_split_ifs(true); // Has chance for split-if optimization
5646 result = phi;
5647 assert(result->bottom_type()->isa_int(), "must be");
5648 break;
5649 }
5650
5651 default:
5652 fatal_unexpected_iid(id);
5653 break;
5654 }
5655 set_result(_gvn.transform(result));
5656 return true;
5657 }
5658
5659 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5660 Node* arg = argument(0);
5661 Node* result = nullptr;
5662
5663 switch (id) {
5664 case vmIntrinsics::_floatIsInfinite:
5665 result = new IsInfiniteFNode(arg);
5666 break;
5667 case vmIntrinsics::_floatIsFinite:
5668 result = new IsFiniteFNode(arg);
5669 break;
5670 case vmIntrinsics::_doubleIsInfinite:
5671 result = new IsInfiniteDNode(arg);
5672 break;
5673 case vmIntrinsics::_doubleIsFinite:
5674 result = new IsFiniteDNode(arg);
5675 break;
5676 default:
5677 fatal_unexpected_iid(id);
5678 break;
5679 }
5680 set_result(_gvn.transform(result));
5681 return true;
5682 }
5683
5684 //----------------------inline_unsafe_copyMemory-------------------------
5685 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5686
5687 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5688 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5689 const Type* base_t = gvn.type(base);
5690
5691 bool in_native = (base_t == TypePtr::NULL_PTR);
5692 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5693 bool is_mixed = !in_heap && !in_native;
5694
5695 if (is_mixed) {
5696 return true; // mixed accesses can touch both on-heap and off-heap memory
5697 }
5698 if (in_heap) {
5699 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5700 if (!is_prim_array) {
5701 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5702 // there's not enough type information available to determine proper memory slice for it.
5703 return true;
5704 }
5705 }
5706 return false;
5707 }
5708
5709 bool LibraryCallKit::inline_unsafe_copyMemory() {
5710 if (callee()->is_static()) return false; // caller must have the capability!
5711 null_check_receiver(); // null-check receiver
5712 if (stopped()) return true;
5713
5714 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5715
5716 Node* src_base = argument(1); // type: oop
5717 Node* src_off = ConvL2X(argument(2)); // type: long
5718 Node* dst_base = argument(4); // type: oop
5719 Node* dst_off = ConvL2X(argument(5)); // type: long
5720 Node* size = ConvL2X(argument(7)); // type: long
5721
5722 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5723 "fieldOffset must be byte-scaled");
5724
5725 Node* src_addr = make_unsafe_address(src_base, src_off);
5726 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5727
5728 Node* thread = _gvn.transform(new ThreadLocalNode());
5729 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5730 BasicType doing_unsafe_access_bt = T_BYTE;
5731 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5732
5733 // update volatile field
5734 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5735
5736 int flags = RC_LEAF | RC_NO_FP;
5737
5738 const TypePtr* dst_type = TypePtr::BOTTOM;
5739
5740 // Adjust memory effects of the runtime call based on input values.
5741 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5742 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5743 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5744
5745 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5746 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5747 flags |= RC_NARROW_MEM; // narrow in memory
5748 }
5749 }
5750
5751 // Call it. Note that the length argument is not scaled.
5752 make_runtime_call(flags,
5753 OptoRuntime::fast_arraycopy_Type(),
5754 StubRoutines::unsafe_arraycopy(),
5755 "unsafe_arraycopy",
5756 dst_type,
5757 src_addr, dst_addr, size XTOP);
5758
5759 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5760
5761 return true;
5762 }
5763
5764 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5765 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5766 bool LibraryCallKit::inline_unsafe_setMemory() {
5767 if (callee()->is_static()) return false; // caller must have the capability!
5768 null_check_receiver(); // null-check receiver
5769 if (stopped()) return true;
5770
5771 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5772
5773 Node* dst_base = argument(1); // type: oop
5774 Node* dst_off = ConvL2X(argument(2)); // type: long
5775 Node* size = ConvL2X(argument(4)); // type: long
5776 Node* byte = argument(6); // type: byte
5777
5778 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5779 "fieldOffset must be byte-scaled");
5780
5781 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5782
5783 Node* thread = _gvn.transform(new ThreadLocalNode());
5784 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5785 BasicType doing_unsafe_access_bt = T_BYTE;
5786 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5787
5788 // update volatile field
5789 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5790
5791 int flags = RC_LEAF | RC_NO_FP;
5792
5793 const TypePtr* dst_type = TypePtr::BOTTOM;
5794
5795 // Adjust memory effects of the runtime call based on input values.
5796 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5797 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5798
5799 flags |= RC_NARROW_MEM; // narrow in memory
5800 }
5801
5802 // Call it. Note that the length argument is not scaled.
5803 make_runtime_call(flags,
5804 OptoRuntime::unsafe_setmemory_Type(),
5805 StubRoutines::unsafe_setmemory(),
5806 "unsafe_setmemory",
5807 dst_type,
5808 dst_addr, size XTOP, byte);
5809
5810 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5811
5812 return true;
5813 }
5814
5815 #undef XTOP
5816
5817 //------------------------clone_coping-----------------------------------
5818 // Helper function for inline_native_clone.
5819 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5820 assert(obj_size != nullptr, "");
5821 Node* raw_obj = alloc_obj->in(1);
5822 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5823
5824 AllocateNode* alloc = nullptr;
5825 if (ReduceBulkZeroing &&
5826 // If we are implementing an array clone without knowing its source type
5827 // (can happen when compiling the array-guarded branch of a reflective
5828 // Object.clone() invocation), initialize the array within the allocation.
5829 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5830 // to a runtime clone call that assumes fully initialized source arrays.
5831 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5832 // We will be completely responsible for initializing this object -
5833 // mark Initialize node as complete.
5834 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5835 // The object was just allocated - there should be no any stores!
5836 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5837 // Mark as complete_with_arraycopy so that on AllocateNode
5838 // expansion, we know this AllocateNode is initialized by an array
5839 // copy and a StoreStore barrier exists after the array copy.
5840 alloc->initialization()->set_complete_with_arraycopy();
5841 }
5842
5843 Node* size = _gvn.transform(obj_size);
5844 access_clone(obj, alloc_obj, size, is_array);
5845
5846 // Do not let reads from the cloned object float above the arraycopy.
5847 if (alloc != nullptr) {
5848 // Do not let stores that initialize this object be reordered with
5849 // a subsequent store that would make this object accessible by
5850 // other threads.
5851 // Record what AllocateNode this StoreStore protects so that
5852 // escape analysis can go from the MemBarStoreStoreNode to the
5853 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5854 // based on the escape status of the AllocateNode.
5855 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5856 } else {
5857 insert_mem_bar(Op_MemBarCPUOrder);
5858 }
5859 }
5860
5861 //------------------------inline_native_clone----------------------------
5862 // protected native Object java.lang.Object.clone();
5863 //
5864 // Here are the simple edge cases:
5865 // null receiver => normal trap
5866 // virtual and clone was overridden => slow path to out-of-line clone
5867 // not cloneable or finalizer => slow path to out-of-line Object.clone
5868 //
5869 // The general case has two steps, allocation and copying.
5870 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5871 //
5872 // Copying also has two cases, oop arrays and everything else.
5873 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5874 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5875 //
5876 // These steps fold up nicely if and when the cloned object's klass
5877 // can be sharply typed as an object array, a type array, or an instance.
5878 //
5879 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5880 PhiNode* result_val;
5881
5882 // Set the reexecute bit for the interpreter to reexecute
5883 // the bytecode that invokes Object.clone if deoptimization happens.
5884 { PreserveReexecuteState preexecs(this);
5885 jvms()->set_should_reexecute(true);
5886
5887 Node* obj = argument(0);
5888 obj = null_check_receiver();
5889 if (stopped()) return true;
5890
5891 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5892 if (obj_type->is_inlinetypeptr()) {
5893 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5894 // no identity.
5895 set_result(obj);
5896 return true;
5897 }
5898
5899 // If we are going to clone an instance, we need its exact type to
5900 // know the number and types of fields to convert the clone to
5901 // loads/stores. Maybe a speculative type can help us.
5902 if (!obj_type->klass_is_exact() &&
5903 obj_type->speculative_type() != nullptr &&
5904 obj_type->speculative_type()->is_instance_klass() &&
5905 !obj_type->speculative_type()->is_inlinetype()) {
5906 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5907 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5908 !spec_ik->has_injected_fields()) {
5909 if (!obj_type->isa_instptr() ||
5910 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5911 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5912 }
5913 }
5914 }
5915
5916 // Conservatively insert a memory barrier on all memory slices.
5917 // Do not let writes into the original float below the clone.
5918 insert_mem_bar(Op_MemBarCPUOrder);
5919
5920 // paths into result_reg:
5921 enum {
5922 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5923 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5924 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5925 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5926 PATH_LIMIT
5927 };
5928 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5929 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5930 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5931 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5932 record_for_igvn(result_reg);
5933
5934 Node* obj_klass = load_object_klass(obj);
5935 // We only go to the fast case code if we pass a number of guards.
5936 // The paths which do not pass are accumulated in the slow_region.
5937 RegionNode* slow_region = new RegionNode(1);
5938 record_for_igvn(slow_region);
5939
5940 Node* array_obj = obj;
5941 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5942 if (array_ctl != nullptr) {
5943 // It's an array.
5944 PreserveJVMState pjvms(this);
5945 set_control(array_ctl);
5946
5947 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5948 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
5949 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
5950 obj_type->can_be_inline_array() &&
5951 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
5952 // Flat inline type array may have object field that would require a
5953 // write barrier. Conservatively, go to slow path.
5954 generate_fair_guard(flat_array_test(obj_klass), slow_region);
5955 }
5956
5957 if (!stopped()) {
5958 Node* obj_length = load_array_length(array_obj);
5959 Node* array_size = nullptr; // Size of the array without object alignment padding.
5960 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5961
5962 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5963 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5964 // If it is an oop array, it requires very special treatment,
5965 // because gc barriers are required when accessing the array.
5966 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
5967 if (is_obja != nullptr) {
5968 PreserveJVMState pjvms2(this);
5969 set_control(is_obja);
5970 // Generate a direct call to the right arraycopy function(s).
5971 // Clones are always tightly coupled.
5972 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5973 ac->set_clone_oop_array();
5974 Node* n = _gvn.transform(ac);
5975 assert(n == ac, "cannot disappear");
5976 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5977
5978 result_reg->init_req(_objArray_path, control());
5979 result_val->init_req(_objArray_path, alloc_obj);
5980 result_i_o ->set_req(_objArray_path, i_o());
5981 result_mem ->set_req(_objArray_path, reset_memory());
5982 }
5983 }
5984 // Otherwise, there are no barriers to worry about.
5985 // (We can dispense with card marks if we know the allocation
5986 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5987 // causes the non-eden paths to take compensating steps to
5988 // simulate a fresh allocation, so that no further
5989 // card marks are required in compiled code to initialize
5990 // the object.)
5991
5992 if (!stopped()) {
5993 copy_to_clone(obj, alloc_obj, array_size, true);
5994
5995 // Present the results of the copy.
5996 result_reg->init_req(_array_path, control());
5997 result_val->init_req(_array_path, alloc_obj);
5998 result_i_o ->set_req(_array_path, i_o());
5999 result_mem ->set_req(_array_path, reset_memory());
6000 }
6001 }
6002 }
6003
6004 if (!stopped()) {
6005 // It's an instance (we did array above). Make the slow-path tests.
6006 // If this is a virtual call, we generate a funny guard. We grab
6007 // the vtable entry corresponding to clone() from the target object.
6008 // If the target method which we are calling happens to be the
6009 // Object clone() method, we pass the guard. We do not need this
6010 // guard for non-virtual calls; the caller is known to be the native
6011 // Object clone().
6012 if (is_virtual) {
6013 generate_virtual_guard(obj_klass, slow_region);
6014 }
6015
6016 // The object must be easily cloneable and must not have a finalizer.
6017 // Both of these conditions may be checked in a single test.
6018 // We could optimize the test further, but we don't care.
6019 generate_misc_flags_guard(obj_klass,
6020 // Test both conditions:
6021 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6022 // Must be cloneable but not finalizer:
6023 KlassFlags::_misc_is_cloneable_fast,
6024 slow_region);
6025 }
6026
6027 if (!stopped()) {
6028 // It's an instance, and it passed the slow-path tests.
6029 PreserveJVMState pjvms(this);
6030 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6031 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6032 // is reexecuted if deoptimization occurs and there could be problems when merging
6033 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6034 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6035
6036 copy_to_clone(obj, alloc_obj, obj_size, false);
6037
6038 // Present the results of the slow call.
6039 result_reg->init_req(_instance_path, control());
6040 result_val->init_req(_instance_path, alloc_obj);
6041 result_i_o ->set_req(_instance_path, i_o());
6042 result_mem ->set_req(_instance_path, reset_memory());
6043 }
6044
6045 // Generate code for the slow case. We make a call to clone().
6046 set_control(_gvn.transform(slow_region));
6047 if (!stopped()) {
6048 PreserveJVMState pjvms(this);
6049 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6050 // We need to deoptimize on exception (see comment above)
6051 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6052 // this->control() comes from set_results_for_java_call
6053 result_reg->init_req(_slow_path, control());
6054 result_val->init_req(_slow_path, slow_result);
6055 result_i_o ->set_req(_slow_path, i_o());
6056 result_mem ->set_req(_slow_path, reset_memory());
6057 }
6058
6059 // Return the combined state.
6060 set_control( _gvn.transform(result_reg));
6061 set_i_o( _gvn.transform(result_i_o));
6062 set_all_memory( _gvn.transform(result_mem));
6063 } // original reexecute is set back here
6064
6065 set_result(_gvn.transform(result_val));
6066 return true;
6067 }
6068
6069 // If we have a tightly coupled allocation, the arraycopy may take care
6070 // of the array initialization. If one of the guards we insert between
6071 // the allocation and the arraycopy causes a deoptimization, an
6072 // uninitialized array will escape the compiled method. To prevent that
6073 // we set the JVM state for uncommon traps between the allocation and
6074 // the arraycopy to the state before the allocation so, in case of
6075 // deoptimization, we'll reexecute the allocation and the
6076 // initialization.
6077 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6078 if (alloc != nullptr) {
6079 ciMethod* trap_method = alloc->jvms()->method();
6080 int trap_bci = alloc->jvms()->bci();
6081
6082 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6083 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6084 // Make sure there's no store between the allocation and the
6085 // arraycopy otherwise visible side effects could be rexecuted
6086 // in case of deoptimization and cause incorrect execution.
6087 bool no_interfering_store = true;
6088 Node* mem = alloc->in(TypeFunc::Memory);
6089 if (mem->is_MergeMem()) {
6090 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6091 Node* n = mms.memory();
6092 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6093 assert(n->is_Store(), "what else?");
6094 no_interfering_store = false;
6095 break;
6096 }
6097 }
6098 } else {
6099 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6100 Node* n = mms.memory();
6101 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6102 assert(n->is_Store(), "what else?");
6103 no_interfering_store = false;
6104 break;
6105 }
6106 }
6107 }
6108
6109 if (no_interfering_store) {
6110 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6111
6112 JVMState* saved_jvms = jvms();
6113 saved_reexecute_sp = _reexecute_sp;
6114
6115 set_jvms(sfpt->jvms());
6116 _reexecute_sp = jvms()->sp();
6117
6118 return saved_jvms;
6119 }
6120 }
6121 }
6122 return nullptr;
6123 }
6124
6125 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6126 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6127 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6128 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6129 uint size = alloc->req();
6130 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6131 old_jvms->set_map(sfpt);
6132 for (uint i = 0; i < size; i++) {
6133 sfpt->init_req(i, alloc->in(i));
6134 }
6135 int adjustment = 1;
6136 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6137 if (ary_klass_ptr->is_null_free()) {
6138 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6139 // also requires the componentType and initVal on stack for re-execution.
6140 // Re-create and push the componentType.
6141 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6142 ciInstance* instance = klass->component_mirror_instance();
6143 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6144 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6145 adjustment++;
6146 }
6147 // re-push array length for deoptimization
6148 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6149 if (ary_klass_ptr->is_null_free()) {
6150 // Re-create and push the initVal.
6151 Node* init_val = alloc->in(AllocateNode::InitValue);
6152 if (init_val == nullptr) {
6153 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6154 } else if (UseCompressedOops) {
6155 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6156 }
6157 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6158 adjustment++;
6159 }
6160 old_jvms->set_sp(old_jvms->sp() + adjustment);
6161 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6162 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6163 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6164 old_jvms->set_should_reexecute(true);
6165
6166 sfpt->set_i_o(map()->i_o());
6167 sfpt->set_memory(map()->memory());
6168 sfpt->set_control(map()->control());
6169 return sfpt;
6170 }
6171
6172 // In case of a deoptimization, we restart execution at the
6173 // allocation, allocating a new array. We would leave an uninitialized
6174 // array in the heap that GCs wouldn't expect. Move the allocation
6175 // after the traps so we don't allocate the array if we
6176 // deoptimize. This is possible because tightly_coupled_allocation()
6177 // guarantees there's no observer of the allocated array at this point
6178 // and the control flow is simple enough.
6179 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6180 int saved_reexecute_sp, uint new_idx) {
6181 if (saved_jvms_before_guards != nullptr && !stopped()) {
6182 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6183
6184 assert(alloc != nullptr, "only with a tightly coupled allocation");
6185 // restore JVM state to the state at the arraycopy
6186 saved_jvms_before_guards->map()->set_control(map()->control());
6187 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6188 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6189 // If we've improved the types of some nodes (null check) while
6190 // emitting the guards, propagate them to the current state
6191 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6192 set_jvms(saved_jvms_before_guards);
6193 _reexecute_sp = saved_reexecute_sp;
6194
6195 // Remove the allocation from above the guards
6196 CallProjections* callprojs = alloc->extract_projections(true);
6197 InitializeNode* init = alloc->initialization();
6198 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6199 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6200 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
6201
6202 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6203 // the allocation (i.e. is only valid if the allocation succeeds):
6204 // 1) replace CastIINode with AllocateArrayNode's length here
6205 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6206 //
6207 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6208 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6209 Node* init_control = init->proj_out(TypeFunc::Control);
6210 Node* alloc_length = alloc->Ideal_length();
6211 #ifdef ASSERT
6212 Node* prev_cast = nullptr;
6213 #endif
6214 for (uint i = 0; i < init_control->outcnt(); i++) {
6215 Node* init_out = init_control->raw_out(i);
6216 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6217 #ifdef ASSERT
6218 if (prev_cast == nullptr) {
6219 prev_cast = init_out;
6220 } else {
6221 if (prev_cast->cmp(*init_out) == false) {
6222 prev_cast->dump();
6223 init_out->dump();
6224 assert(false, "not equal CastIINode");
6225 }
6226 }
6227 #endif
6228 C->gvn_replace_by(init_out, alloc_length);
6229 }
6230 }
6231 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6232
6233 // move the allocation here (after the guards)
6234 _gvn.hash_delete(alloc);
6235 alloc->set_req(TypeFunc::Control, control());
6236 alloc->set_req(TypeFunc::I_O, i_o());
6237 Node *mem = reset_memory();
6238 set_all_memory(mem);
6239 alloc->set_req(TypeFunc::Memory, mem);
6240 set_control(init->proj_out_or_null(TypeFunc::Control));
6241 set_i_o(callprojs->fallthrough_ioproj);
6242
6243 // Update memory as done in GraphKit::set_output_for_allocation()
6244 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6245 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6246 if (ary_type->isa_aryptr() && length_type != nullptr) {
6247 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6248 }
6249 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6250 int elemidx = C->get_alias_index(telemref);
6251 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
6252 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
6253
6254 Node* allocx = _gvn.transform(alloc);
6255 assert(allocx == alloc, "where has the allocation gone?");
6256 assert(dest->is_CheckCastPP(), "not an allocation result?");
6257
6258 _gvn.hash_delete(dest);
6259 dest->set_req(0, control());
6260 Node* destx = _gvn.transform(dest);
6261 assert(destx == dest, "where has the allocation result gone?");
6262
6263 array_ideal_length(alloc, ary_type, true);
6264 }
6265 }
6266
6267 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6268 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6269 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6270 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6271 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6272 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6273 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6274 JVMState* saved_jvms_before_guards) {
6275 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6276 // There is at least one unrelated uncommon trap which needs to be replaced.
6277 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6278
6279 JVMState* saved_jvms = jvms();
6280 const int saved_reexecute_sp = _reexecute_sp;
6281 set_jvms(sfpt->jvms());
6282 _reexecute_sp = jvms()->sp();
6283
6284 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6285
6286 // Restore state
6287 set_jvms(saved_jvms);
6288 _reexecute_sp = saved_reexecute_sp;
6289 }
6290 }
6291
6292 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6293 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6294 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6295 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6296 while (if_proj->is_IfProj()) {
6297 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6298 if (uncommon_trap != nullptr) {
6299 create_new_uncommon_trap(uncommon_trap);
6300 }
6301 assert(if_proj->in(0)->is_If(), "must be If");
6302 if_proj = if_proj->in(0)->in(0);
6303 }
6304 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6305 "must have reached control projection of init node");
6306 }
6307
6308 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6309 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6310 assert(trap_request != 0, "no valid UCT trap request");
6311 PreserveJVMState pjvms(this);
6312 set_control(uncommon_trap_call->in(0));
6313 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6314 Deoptimization::trap_request_action(trap_request));
6315 assert(stopped(), "Should be stopped");
6316 _gvn.hash_delete(uncommon_trap_call);
6317 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6318 }
6319
6320 // Common checks for array sorting intrinsics arguments.
6321 // Returns `true` if checks passed.
6322 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6323 // check address of the class
6324 if (elementType == nullptr || elementType->is_top()) {
6325 return false; // dead path
6326 }
6327 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6328 if (elem_klass == nullptr) {
6329 return false; // dead path
6330 }
6331 // java_mirror_type() returns non-null for compile-time Class constants only
6332 ciType* elem_type = elem_klass->java_mirror_type();
6333 if (elem_type == nullptr) {
6334 return false;
6335 }
6336 bt = elem_type->basic_type();
6337 // Disable the intrinsic if the CPU does not support SIMD sort
6338 if (!Matcher::supports_simd_sort(bt)) {
6339 return false;
6340 }
6341 // check address of the array
6342 if (obj == nullptr || obj->is_top()) {
6343 return false; // dead path
6344 }
6345 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6346 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6347 return false; // failed input validation
6348 }
6349 return true;
6350 }
6351
6352 //------------------------------inline_array_partition-----------------------
6353 bool LibraryCallKit::inline_array_partition() {
6354 address stubAddr = StubRoutines::select_array_partition_function();
6355 if (stubAddr == nullptr) {
6356 return false; // Intrinsic's stub is not implemented on this platform
6357 }
6358 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6359
6360 // no receiver because it is a static method
6361 Node* elementType = argument(0);
6362 Node* obj = argument(1);
6363 Node* offset = argument(2); // long
6364 Node* fromIndex = argument(4);
6365 Node* toIndex = argument(5);
6366 Node* indexPivot1 = argument(6);
6367 Node* indexPivot2 = argument(7);
6368 // PartitionOperation: argument(8) is ignored
6369
6370 Node* pivotIndices = nullptr;
6371 BasicType bt = T_ILLEGAL;
6372
6373 if (!check_array_sort_arguments(elementType, obj, bt)) {
6374 return false;
6375 }
6376 null_check(obj);
6377 // If obj is dead, only null-path is taken.
6378 if (stopped()) {
6379 return true;
6380 }
6381 // Set the original stack and the reexecute bit for the interpreter to reexecute
6382 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6383 { PreserveReexecuteState preexecs(this);
6384 jvms()->set_should_reexecute(true);
6385
6386 Node* obj_adr = make_unsafe_address(obj, offset);
6387
6388 // create the pivotIndices array of type int and size = 2
6389 Node* size = intcon(2);
6390 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6391 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6392 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6393 guarantee(alloc != nullptr, "created above");
6394 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6395
6396 // pass the basic type enum to the stub
6397 Node* elemType = intcon(bt);
6398
6399 // Call the stub
6400 const char *stubName = "array_partition_stub";
6401 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6402 stubAddr, stubName, TypePtr::BOTTOM,
6403 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6404 indexPivot1, indexPivot2);
6405
6406 } // original reexecute is set back here
6407
6408 if (!stopped()) {
6409 set_result(pivotIndices);
6410 }
6411
6412 return true;
6413 }
6414
6415
6416 //------------------------------inline_array_sort-----------------------
6417 bool LibraryCallKit::inline_array_sort() {
6418 address stubAddr = StubRoutines::select_arraysort_function();
6419 if (stubAddr == nullptr) {
6420 return false; // Intrinsic's stub is not implemented on this platform
6421 }
6422 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6423
6424 // no receiver because it is a static method
6425 Node* elementType = argument(0);
6426 Node* obj = argument(1);
6427 Node* offset = argument(2); // long
6428 Node* fromIndex = argument(4);
6429 Node* toIndex = argument(5);
6430 // SortOperation: argument(6) is ignored
6431
6432 BasicType bt = T_ILLEGAL;
6433
6434 if (!check_array_sort_arguments(elementType, obj, bt)) {
6435 return false;
6436 }
6437 null_check(obj);
6438 // If obj is dead, only null-path is taken.
6439 if (stopped()) {
6440 return true;
6441 }
6442 Node* obj_adr = make_unsafe_address(obj, offset);
6443
6444 // pass the basic type enum to the stub
6445 Node* elemType = intcon(bt);
6446
6447 // Call the stub.
6448 const char *stubName = "arraysort_stub";
6449 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6450 stubAddr, stubName, TypePtr::BOTTOM,
6451 obj_adr, elemType, fromIndex, toIndex);
6452
6453 return true;
6454 }
6455
6456
6457 //------------------------------inline_arraycopy-----------------------
6458 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6459 // Object dest, int destPos,
6460 // int length);
6461 bool LibraryCallKit::inline_arraycopy() {
6462 // Get the arguments.
6463 Node* src = argument(0); // type: oop
6464 Node* src_offset = argument(1); // type: int
6465 Node* dest = argument(2); // type: oop
6466 Node* dest_offset = argument(3); // type: int
6467 Node* length = argument(4); // type: int
6468
6469 uint new_idx = C->unique();
6470
6471 // Check for allocation before we add nodes that would confuse
6472 // tightly_coupled_allocation()
6473 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6474
6475 int saved_reexecute_sp = -1;
6476 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6477 // See arraycopy_restore_alloc_state() comment
6478 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6479 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6480 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6481 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6482
6483 // The following tests must be performed
6484 // (1) src and dest are arrays.
6485 // (2) src and dest arrays must have elements of the same BasicType
6486 // (3) src and dest must not be null.
6487 // (4) src_offset must not be negative.
6488 // (5) dest_offset must not be negative.
6489 // (6) length must not be negative.
6490 // (7) src_offset + length must not exceed length of src.
6491 // (8) dest_offset + length must not exceed length of dest.
6492 // (9) each element of an oop array must be assignable
6493
6494 // (3) src and dest must not be null.
6495 // always do this here because we need the JVM state for uncommon traps
6496 Node* null_ctl = top();
6497 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6498 assert(null_ctl->is_top(), "no null control here");
6499 dest = null_check(dest, T_ARRAY);
6500
6501 if (!can_emit_guards) {
6502 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6503 // guards but the arraycopy node could still take advantage of a
6504 // tightly allocated allocation. tightly_coupled_allocation() is
6505 // called again to make sure it takes the null check above into
6506 // account: the null check is mandatory and if it caused an
6507 // uncommon trap to be emitted then the allocation can't be
6508 // considered tightly coupled in this context.
6509 alloc = tightly_coupled_allocation(dest);
6510 }
6511
6512 bool validated = false;
6513
6514 const Type* src_type = _gvn.type(src);
6515 const Type* dest_type = _gvn.type(dest);
6516 const TypeAryPtr* top_src = src_type->isa_aryptr();
6517 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6518
6519 // Do we have the type of src?
6520 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6521 // Do we have the type of dest?
6522 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6523 // Is the type for src from speculation?
6524 bool src_spec = false;
6525 // Is the type for dest from speculation?
6526 bool dest_spec = false;
6527
6528 if ((!has_src || !has_dest) && can_emit_guards) {
6529 // We don't have sufficient type information, let's see if
6530 // speculative types can help. We need to have types for both src
6531 // and dest so that it pays off.
6532
6533 // Do we already have or could we have type information for src
6534 bool could_have_src = has_src;
6535 // Do we already have or could we have type information for dest
6536 bool could_have_dest = has_dest;
6537
6538 ciKlass* src_k = nullptr;
6539 if (!has_src) {
6540 src_k = src_type->speculative_type_not_null();
6541 if (src_k != nullptr && src_k->is_array_klass()) {
6542 could_have_src = true;
6543 }
6544 }
6545
6546 ciKlass* dest_k = nullptr;
6547 if (!has_dest) {
6548 dest_k = dest_type->speculative_type_not_null();
6549 if (dest_k != nullptr && dest_k->is_array_klass()) {
6550 could_have_dest = true;
6551 }
6552 }
6553
6554 if (could_have_src && could_have_dest) {
6555 // This is going to pay off so emit the required guards
6556 if (!has_src) {
6557 src = maybe_cast_profiled_obj(src, src_k, true);
6558 src_type = _gvn.type(src);
6559 top_src = src_type->isa_aryptr();
6560 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6561 src_spec = true;
6562 }
6563 if (!has_dest) {
6564 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6565 dest_type = _gvn.type(dest);
6566 top_dest = dest_type->isa_aryptr();
6567 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6568 dest_spec = true;
6569 }
6570 }
6571 }
6572
6573 if (has_src && has_dest && can_emit_guards) {
6574 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6575 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6576 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6577 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6578
6579 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6580 // If both arrays are object arrays then having the exact types
6581 // for both will remove the need for a subtype check at runtime
6582 // before the call and may make it possible to pick a faster copy
6583 // routine (without a subtype check on every element)
6584 // Do we have the exact type of src?
6585 bool could_have_src = src_spec;
6586 // Do we have the exact type of dest?
6587 bool could_have_dest = dest_spec;
6588 ciKlass* src_k = nullptr;
6589 ciKlass* dest_k = nullptr;
6590 if (!src_spec) {
6591 src_k = src_type->speculative_type_not_null();
6592 if (src_k != nullptr && src_k->is_array_klass()) {
6593 could_have_src = true;
6594 }
6595 }
6596 if (!dest_spec) {
6597 dest_k = dest_type->speculative_type_not_null();
6598 if (dest_k != nullptr && dest_k->is_array_klass()) {
6599 could_have_dest = true;
6600 }
6601 }
6602 if (could_have_src && could_have_dest) {
6603 // If we can have both exact types, emit the missing guards
6604 if (could_have_src && !src_spec) {
6605 src = maybe_cast_profiled_obj(src, src_k, true);
6606 src_type = _gvn.type(src);
6607 top_src = src_type->isa_aryptr();
6608 }
6609 if (could_have_dest && !dest_spec) {
6610 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6611 dest_type = _gvn.type(dest);
6612 top_dest = dest_type->isa_aryptr();
6613 }
6614 }
6615 }
6616 }
6617
6618 ciMethod* trap_method = method();
6619 int trap_bci = bci();
6620 if (saved_jvms_before_guards != nullptr) {
6621 trap_method = alloc->jvms()->method();
6622 trap_bci = alloc->jvms()->bci();
6623 }
6624
6625 bool negative_length_guard_generated = false;
6626
6627 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6628 can_emit_guards && !src->is_top() && !dest->is_top()) {
6629 // validate arguments: enables transformation the ArrayCopyNode
6630 validated = true;
6631
6632 RegionNode* slow_region = new RegionNode(1);
6633 record_for_igvn(slow_region);
6634
6635 // (1) src and dest are arrays.
6636 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6637 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6638
6639 // (2) src and dest arrays must have elements of the same BasicType
6640 // done at macro expansion or at Ideal transformation time
6641
6642 // (4) src_offset must not be negative.
6643 generate_negative_guard(src_offset, slow_region);
6644
6645 // (5) dest_offset must not be negative.
6646 generate_negative_guard(dest_offset, slow_region);
6647
6648 // (7) src_offset + length must not exceed length of src.
6649 generate_limit_guard(src_offset, length,
6650 load_array_length(src),
6651 slow_region);
6652
6653 // (8) dest_offset + length must not exceed length of dest.
6654 generate_limit_guard(dest_offset, length,
6655 load_array_length(dest),
6656 slow_region);
6657
6658 // (6) length must not be negative.
6659 // This is also checked in generate_arraycopy() during macro expansion, but
6660 // we also have to check it here for the case where the ArrayCopyNode will
6661 // be eliminated by Escape Analysis.
6662 if (EliminateAllocations) {
6663 generate_negative_guard(length, slow_region);
6664 negative_length_guard_generated = true;
6665 }
6666
6667 // (9) each element of an oop array must be assignable
6668 Node* dest_klass = load_object_klass(dest);
6669 if (src != dest) {
6670 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6671 slow_region->add_req(not_subtype_ctrl);
6672 }
6673
6674 // TODO 8350865 Fix below logic. Also handle atomicity.
6675 generate_fair_guard(flat_array_test(src), slow_region);
6676 generate_fair_guard(flat_array_test(dest), slow_region);
6677
6678 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6679 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6680 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6681 src_type = _gvn.type(src);
6682 top_src = src_type->isa_aryptr();
6683
6684 // Handle flat inline type arrays (null-free arrays are handled by the subtype check above)
6685 if (!stopped() && UseArrayFlattening) {
6686 // If dest is flat, src must be flat as well (guaranteed by src <: dest check). Handle flat src here.
6687 assert(top_dest == nullptr || !top_dest->is_flat() || top_src->is_flat(), "src array must be flat");
6688 if (top_src != nullptr && top_src->is_flat()) {
6689 // Src is flat, check that dest is flat as well
6690 if (top_dest != nullptr && !top_dest->is_flat()) {
6691 generate_fair_guard(flat_array_test(dest_klass, /* flat = */ false), slow_region);
6692 // Since dest is flat and src <: dest, dest must have the same type as src.
6693 top_dest = top_src->cast_to_exactness(false);
6694 assert(top_dest->is_flat(), "dest must be flat");
6695 dest = _gvn.transform(new CheckCastPPNode(control(), dest, top_dest));
6696 }
6697 } else if (top_src == nullptr || !top_src->is_not_flat()) {
6698 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
6699 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
6700 assert(top_dest == nullptr || !top_dest->is_flat(), "dest array must not be flat");
6701 generate_fair_guard(flat_array_test(src), slow_region);
6702 if (top_src != nullptr) {
6703 top_src = top_src->cast_to_not_flat();
6704 src = _gvn.transform(new CheckCastPPNode(control(), src, top_src));
6705 }
6706 }
6707 }
6708
6709 {
6710 PreserveJVMState pjvms(this);
6711 set_control(_gvn.transform(slow_region));
6712 uncommon_trap(Deoptimization::Reason_intrinsic,
6713 Deoptimization::Action_make_not_entrant);
6714 assert(stopped(), "Should be stopped");
6715 }
6716 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6717 }
6718
6719 if (stopped()) {
6720 return true;
6721 }
6722
6723 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6724 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6725 // so the compiler has a chance to eliminate them: during macro expansion,
6726 // we have to set their control (CastPP nodes are eliminated).
6727 load_object_klass(src), load_object_klass(dest),
6728 load_array_length(src), load_array_length(dest));
6729
6730 ac->set_arraycopy(validated);
6731
6732 Node* n = _gvn.transform(ac);
6733 if (n == ac) {
6734 ac->connect_outputs(this);
6735 } else {
6736 assert(validated, "shouldn't transform if all arguments not validated");
6737 set_all_memory(n);
6738 }
6739 clear_upper_avx();
6740
6741
6742 return true;
6743 }
6744
6745
6746 // Helper function which determines if an arraycopy immediately follows
6747 // an allocation, with no intervening tests or other escapes for the object.
6748 AllocateArrayNode*
6749 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6750 if (stopped()) return nullptr; // no fast path
6751 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6752
6753 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6754 if (alloc == nullptr) return nullptr;
6755
6756 Node* rawmem = memory(Compile::AliasIdxRaw);
6757 // Is the allocation's memory state untouched?
6758 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6759 // Bail out if there have been raw-memory effects since the allocation.
6760 // (Example: There might have been a call or safepoint.)
6761 return nullptr;
6762 }
6763 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6764 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6765 return nullptr;
6766 }
6767
6768 // There must be no unexpected observers of this allocation.
6769 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6770 Node* obs = ptr->fast_out(i);
6771 if (obs != this->map()) {
6772 return nullptr;
6773 }
6774 }
6775
6776 // This arraycopy must unconditionally follow the allocation of the ptr.
6777 Node* alloc_ctl = ptr->in(0);
6778 Node* ctl = control();
6779 while (ctl != alloc_ctl) {
6780 // There may be guards which feed into the slow_region.
6781 // Any other control flow means that we might not get a chance
6782 // to finish initializing the allocated object.
6783 // Various low-level checks bottom out in uncommon traps. These
6784 // are considered safe since we've already checked above that
6785 // there is no unexpected observer of this allocation.
6786 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6787 assert(ctl->in(0)->is_If(), "must be If");
6788 ctl = ctl->in(0)->in(0);
6789 } else {
6790 return nullptr;
6791 }
6792 }
6793
6794 // If we get this far, we have an allocation which immediately
6795 // precedes the arraycopy, and we can take over zeroing the new object.
6796 // The arraycopy will finish the initialization, and provide
6797 // a new control state to which we will anchor the destination pointer.
6798
6799 return alloc;
6800 }
6801
6802 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6803 if (node->is_IfProj()) {
6804 Node* other_proj = node->as_IfProj()->other_if_proj();
6805 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6806 Node* obs = other_proj->fast_out(j);
6807 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6808 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6809 return obs->as_CallStaticJava();
6810 }
6811 }
6812 }
6813 return nullptr;
6814 }
6815
6816 //-------------inline_encodeISOArray-----------------------------------
6817 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6818 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6819 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6820 // encode char[] to byte[] in ISO_8859_1 or ASCII
6821 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6822 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6823 // no receiver since it is static method
6824 Node *src = argument(0);
6825 Node *src_offset = argument(1);
6826 Node *dst = argument(2);
6827 Node *dst_offset = argument(3);
6828 Node *length = argument(4);
6829
6830 // Cast source & target arrays to not-null
6831 if (VerifyIntrinsicChecks) {
6832 src = must_be_not_null(src, true);
6833 dst = must_be_not_null(dst, true);
6834 if (stopped()) {
6835 return true;
6836 }
6837 }
6838
6839 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6840 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6841 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6842 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6843 // failed array check
6844 return false;
6845 }
6846
6847 // Figure out the size and type of the elements we will be copying.
6848 BasicType src_elem = src_type->elem()->array_element_basic_type();
6849 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6850 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6851 return false;
6852 }
6853
6854 // Check source & target bounds
6855 if (VerifyIntrinsicChecks) {
6856 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6857 generate_string_range_check(dst, dst_offset, length, false, true);
6858 if (stopped()) {
6859 return true;
6860 }
6861 }
6862
6863 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6864 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6865 // 'src_start' points to src array + scaled offset
6866 // 'dst_start' points to dst array + scaled offset
6867
6868 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6869 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6870 enc = _gvn.transform(enc);
6871 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6872 set_memory(res_mem, mtype);
6873 set_result(enc);
6874 clear_upper_avx();
6875
6876 return true;
6877 }
6878
6879 //-------------inline_multiplyToLen-----------------------------------
6880 bool LibraryCallKit::inline_multiplyToLen() {
6881 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6882
6883 address stubAddr = StubRoutines::multiplyToLen();
6884 if (stubAddr == nullptr) {
6885 return false; // Intrinsic's stub is not implemented on this platform
6886 }
6887 const char* stubName = "multiplyToLen";
6888
6889 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6890
6891 // no receiver because it is a static method
6892 Node* x = argument(0);
6893 Node* xlen = argument(1);
6894 Node* y = argument(2);
6895 Node* ylen = argument(3);
6896 Node* z = argument(4);
6897
6898 x = must_be_not_null(x, true);
6899 y = must_be_not_null(y, true);
6900
6901 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6902 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6903 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6904 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6905 // failed array check
6906 return false;
6907 }
6908
6909 BasicType x_elem = x_type->elem()->array_element_basic_type();
6910 BasicType y_elem = y_type->elem()->array_element_basic_type();
6911 if (x_elem != T_INT || y_elem != T_INT) {
6912 return false;
6913 }
6914
6915 Node* x_start = array_element_address(x, intcon(0), x_elem);
6916 Node* y_start = array_element_address(y, intcon(0), y_elem);
6917 // 'x_start' points to x array + scaled xlen
6918 // 'y_start' points to y array + scaled ylen
6919
6920 Node* z_start = array_element_address(z, intcon(0), T_INT);
6921
6922 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6923 OptoRuntime::multiplyToLen_Type(),
6924 stubAddr, stubName, TypePtr::BOTTOM,
6925 x_start, xlen, y_start, ylen, z_start);
6926
6927 C->set_has_split_ifs(true); // Has chance for split-if optimization
6928 set_result(z);
6929 return true;
6930 }
6931
6932 //-------------inline_squareToLen------------------------------------
6933 bool LibraryCallKit::inline_squareToLen() {
6934 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6935
6936 address stubAddr = StubRoutines::squareToLen();
6937 if (stubAddr == nullptr) {
6938 return false; // Intrinsic's stub is not implemented on this platform
6939 }
6940 const char* stubName = "squareToLen";
6941
6942 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6943
6944 Node* x = argument(0);
6945 Node* len = argument(1);
6946 Node* z = argument(2);
6947 Node* zlen = argument(3);
6948
6949 x = must_be_not_null(x, true);
6950 z = must_be_not_null(z, true);
6951
6952 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6953 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6954 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6955 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6956 // failed array check
6957 return false;
6958 }
6959
6960 BasicType x_elem = x_type->elem()->array_element_basic_type();
6961 BasicType z_elem = z_type->elem()->array_element_basic_type();
6962 if (x_elem != T_INT || z_elem != T_INT) {
6963 return false;
6964 }
6965
6966
6967 Node* x_start = array_element_address(x, intcon(0), x_elem);
6968 Node* z_start = array_element_address(z, intcon(0), z_elem);
6969
6970 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6971 OptoRuntime::squareToLen_Type(),
6972 stubAddr, stubName, TypePtr::BOTTOM,
6973 x_start, len, z_start, zlen);
6974
6975 set_result(z);
6976 return true;
6977 }
6978
6979 //-------------inline_mulAdd------------------------------------------
6980 bool LibraryCallKit::inline_mulAdd() {
6981 assert(UseMulAddIntrinsic, "not implemented on this platform");
6982
6983 address stubAddr = StubRoutines::mulAdd();
6984 if (stubAddr == nullptr) {
6985 return false; // Intrinsic's stub is not implemented on this platform
6986 }
6987 const char* stubName = "mulAdd";
6988
6989 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6990
6991 Node* out = argument(0);
6992 Node* in = argument(1);
6993 Node* offset = argument(2);
6994 Node* len = argument(3);
6995 Node* k = argument(4);
6996
6997 in = must_be_not_null(in, true);
6998 out = must_be_not_null(out, true);
6999
7000 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7001 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7002 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7003 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7004 // failed array check
7005 return false;
7006 }
7007
7008 BasicType out_elem = out_type->elem()->array_element_basic_type();
7009 BasicType in_elem = in_type->elem()->array_element_basic_type();
7010 if (out_elem != T_INT || in_elem != T_INT) {
7011 return false;
7012 }
7013
7014 Node* outlen = load_array_length(out);
7015 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7016 Node* out_start = array_element_address(out, intcon(0), out_elem);
7017 Node* in_start = array_element_address(in, intcon(0), in_elem);
7018
7019 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7020 OptoRuntime::mulAdd_Type(),
7021 stubAddr, stubName, TypePtr::BOTTOM,
7022 out_start,in_start, new_offset, len, k);
7023 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7024 set_result(result);
7025 return true;
7026 }
7027
7028 //-------------inline_montgomeryMultiply-----------------------------------
7029 bool LibraryCallKit::inline_montgomeryMultiply() {
7030 address stubAddr = StubRoutines::montgomeryMultiply();
7031 if (stubAddr == nullptr) {
7032 return false; // Intrinsic's stub is not implemented on this platform
7033 }
7034
7035 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7036 const char* stubName = "montgomery_multiply";
7037
7038 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7039
7040 Node* a = argument(0);
7041 Node* b = argument(1);
7042 Node* n = argument(2);
7043 Node* len = argument(3);
7044 Node* inv = argument(4);
7045 Node* m = argument(6);
7046
7047 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7048 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7049 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7050 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7051 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7052 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7053 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7054 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7055 // failed array check
7056 return false;
7057 }
7058
7059 BasicType a_elem = a_type->elem()->array_element_basic_type();
7060 BasicType b_elem = b_type->elem()->array_element_basic_type();
7061 BasicType n_elem = n_type->elem()->array_element_basic_type();
7062 BasicType m_elem = m_type->elem()->array_element_basic_type();
7063 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7064 return false;
7065 }
7066
7067 // Make the call
7068 {
7069 Node* a_start = array_element_address(a, intcon(0), a_elem);
7070 Node* b_start = array_element_address(b, intcon(0), b_elem);
7071 Node* n_start = array_element_address(n, intcon(0), n_elem);
7072 Node* m_start = array_element_address(m, intcon(0), m_elem);
7073
7074 Node* call = make_runtime_call(RC_LEAF,
7075 OptoRuntime::montgomeryMultiply_Type(),
7076 stubAddr, stubName, TypePtr::BOTTOM,
7077 a_start, b_start, n_start, len, inv, top(),
7078 m_start);
7079 set_result(m);
7080 }
7081
7082 return true;
7083 }
7084
7085 bool LibraryCallKit::inline_montgomerySquare() {
7086 address stubAddr = StubRoutines::montgomerySquare();
7087 if (stubAddr == nullptr) {
7088 return false; // Intrinsic's stub is not implemented on this platform
7089 }
7090
7091 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7092 const char* stubName = "montgomery_square";
7093
7094 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7095
7096 Node* a = argument(0);
7097 Node* n = argument(1);
7098 Node* len = argument(2);
7099 Node* inv = argument(3);
7100 Node* m = argument(5);
7101
7102 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7103 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7104 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7105 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7106 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7107 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7108 // failed array check
7109 return false;
7110 }
7111
7112 BasicType a_elem = a_type->elem()->array_element_basic_type();
7113 BasicType n_elem = n_type->elem()->array_element_basic_type();
7114 BasicType m_elem = m_type->elem()->array_element_basic_type();
7115 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7116 return false;
7117 }
7118
7119 // Make the call
7120 {
7121 Node* a_start = array_element_address(a, intcon(0), a_elem);
7122 Node* n_start = array_element_address(n, intcon(0), n_elem);
7123 Node* m_start = array_element_address(m, intcon(0), m_elem);
7124
7125 Node* call = make_runtime_call(RC_LEAF,
7126 OptoRuntime::montgomerySquare_Type(),
7127 stubAddr, stubName, TypePtr::BOTTOM,
7128 a_start, n_start, len, inv, top(),
7129 m_start);
7130 set_result(m);
7131 }
7132
7133 return true;
7134 }
7135
7136 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7137 address stubAddr = nullptr;
7138 const char* stubName = nullptr;
7139
7140 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7141 if (stubAddr == nullptr) {
7142 return false; // Intrinsic's stub is not implemented on this platform
7143 }
7144
7145 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7146
7147 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7148
7149 Node* newArr = argument(0);
7150 Node* oldArr = argument(1);
7151 Node* newIdx = argument(2);
7152 Node* shiftCount = argument(3);
7153 Node* numIter = argument(4);
7154
7155 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7156 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7157 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7158 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7159 return false;
7160 }
7161
7162 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7163 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7164 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7165 return false;
7166 }
7167
7168 // Make the call
7169 {
7170 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7171 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7172
7173 Node* call = make_runtime_call(RC_LEAF,
7174 OptoRuntime::bigIntegerShift_Type(),
7175 stubAddr,
7176 stubName,
7177 TypePtr::BOTTOM,
7178 newArr_start,
7179 oldArr_start,
7180 newIdx,
7181 shiftCount,
7182 numIter);
7183 }
7184
7185 return true;
7186 }
7187
7188 //-------------inline_vectorizedMismatch------------------------------
7189 bool LibraryCallKit::inline_vectorizedMismatch() {
7190 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7191
7192 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7193 Node* obja = argument(0); // Object
7194 Node* aoffset = argument(1); // long
7195 Node* objb = argument(3); // Object
7196 Node* boffset = argument(4); // long
7197 Node* length = argument(6); // int
7198 Node* scale = argument(7); // int
7199
7200 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7201 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7202 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7203 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7204 scale == top()) {
7205 return false; // failed input validation
7206 }
7207
7208 Node* obja_adr = make_unsafe_address(obja, aoffset);
7209 Node* objb_adr = make_unsafe_address(objb, boffset);
7210
7211 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7212 //
7213 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7214 // if (length <= inline_limit) {
7215 // inline_path:
7216 // vmask = VectorMaskGen length
7217 // vload1 = LoadVectorMasked obja, vmask
7218 // vload2 = LoadVectorMasked objb, vmask
7219 // result1 = VectorCmpMasked vload1, vload2, vmask
7220 // } else {
7221 // call_stub_path:
7222 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7223 // }
7224 // exit_block:
7225 // return Phi(result1, result2);
7226 //
7227 enum { inline_path = 1, // input is small enough to process it all at once
7228 stub_path = 2, // input is too large; call into the VM
7229 PATH_LIMIT = 3
7230 };
7231
7232 Node* exit_block = new RegionNode(PATH_LIMIT);
7233 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7234 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7235
7236 Node* call_stub_path = control();
7237
7238 BasicType elem_bt = T_ILLEGAL;
7239
7240 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7241 if (scale_t->is_con()) {
7242 switch (scale_t->get_con()) {
7243 case 0: elem_bt = T_BYTE; break;
7244 case 1: elem_bt = T_SHORT; break;
7245 case 2: elem_bt = T_INT; break;
7246 case 3: elem_bt = T_LONG; break;
7247
7248 default: elem_bt = T_ILLEGAL; break; // not supported
7249 }
7250 }
7251
7252 int inline_limit = 0;
7253 bool do_partial_inline = false;
7254
7255 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7256 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7257 do_partial_inline = inline_limit >= 16;
7258 }
7259
7260 if (do_partial_inline) {
7261 assert(elem_bt != T_ILLEGAL, "sanity");
7262
7263 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7264 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7265 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7266
7267 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7268 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7269 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7270
7271 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7272
7273 if (!stopped()) {
7274 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7275
7276 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7277 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7278 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7279 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7280
7281 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7282 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7283 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7284 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7285
7286 exit_block->init_req(inline_path, control());
7287 memory_phi->init_req(inline_path, map()->memory());
7288 result_phi->init_req(inline_path, result);
7289
7290 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7291 clear_upper_avx();
7292 }
7293 }
7294 }
7295
7296 if (call_stub_path != nullptr) {
7297 set_control(call_stub_path);
7298
7299 Node* call = make_runtime_call(RC_LEAF,
7300 OptoRuntime::vectorizedMismatch_Type(),
7301 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7302 obja_adr, objb_adr, length, scale);
7303
7304 exit_block->init_req(stub_path, control());
7305 memory_phi->init_req(stub_path, map()->memory());
7306 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7307 }
7308
7309 exit_block = _gvn.transform(exit_block);
7310 memory_phi = _gvn.transform(memory_phi);
7311 result_phi = _gvn.transform(result_phi);
7312
7313 record_for_igvn(exit_block);
7314 record_for_igvn(memory_phi);
7315 record_for_igvn(result_phi);
7316
7317 set_control(exit_block);
7318 set_all_memory(memory_phi);
7319 set_result(result_phi);
7320
7321 return true;
7322 }
7323
7324 //------------------------------inline_vectorizedHashcode----------------------------
7325 bool LibraryCallKit::inline_vectorizedHashCode() {
7326 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7327
7328 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7329 Node* array = argument(0);
7330 Node* offset = argument(1);
7331 Node* length = argument(2);
7332 Node* initialValue = argument(3);
7333 Node* basic_type = argument(4);
7334
7335 if (basic_type == top()) {
7336 return false; // failed input validation
7337 }
7338
7339 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7340 if (!basic_type_t->is_con()) {
7341 return false; // Only intrinsify if mode argument is constant
7342 }
7343
7344 array = must_be_not_null(array, true);
7345
7346 BasicType bt = (BasicType)basic_type_t->get_con();
7347
7348 // Resolve address of first element
7349 Node* array_start = array_element_address(array, offset, bt);
7350
7351 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7352 array_start, length, initialValue, basic_type)));
7353 clear_upper_avx();
7354
7355 return true;
7356 }
7357
7358 /**
7359 * Calculate CRC32 for byte.
7360 * int java.util.zip.CRC32.update(int crc, int b)
7361 */
7362 bool LibraryCallKit::inline_updateCRC32() {
7363 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7364 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7365 // no receiver since it is static method
7366 Node* crc = argument(0); // type: int
7367 Node* b = argument(1); // type: int
7368
7369 /*
7370 * int c = ~ crc;
7371 * b = timesXtoThe32[(b ^ c) & 0xFF];
7372 * b = b ^ (c >>> 8);
7373 * crc = ~b;
7374 */
7375
7376 Node* M1 = intcon(-1);
7377 crc = _gvn.transform(new XorINode(crc, M1));
7378 Node* result = _gvn.transform(new XorINode(crc, b));
7379 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7380
7381 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7382 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7383 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7384 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7385
7386 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7387 result = _gvn.transform(new XorINode(crc, result));
7388 result = _gvn.transform(new XorINode(result, M1));
7389 set_result(result);
7390 return true;
7391 }
7392
7393 /**
7394 * Calculate CRC32 for byte[] array.
7395 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7396 */
7397 bool LibraryCallKit::inline_updateBytesCRC32() {
7398 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7399 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7400 // no receiver since it is static method
7401 Node* crc = argument(0); // type: int
7402 Node* src = argument(1); // type: oop
7403 Node* offset = argument(2); // type: int
7404 Node* length = argument(3); // type: int
7405
7406 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7407 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7408 // failed array check
7409 return false;
7410 }
7411
7412 // Figure out the size and type of the elements we will be copying.
7413 BasicType src_elem = src_type->elem()->array_element_basic_type();
7414 if (src_elem != T_BYTE) {
7415 return false;
7416 }
7417
7418 // 'src_start' points to src array + scaled offset
7419 src = must_be_not_null(src, true);
7420 Node* src_start = array_element_address(src, offset, src_elem);
7421
7422 // We assume that range check is done by caller.
7423 // TODO: generate range check (offset+length < src.length) in debug VM.
7424
7425 // Call the stub.
7426 address stubAddr = StubRoutines::updateBytesCRC32();
7427 const char *stubName = "updateBytesCRC32";
7428
7429 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7430 stubAddr, stubName, TypePtr::BOTTOM,
7431 crc, src_start, length);
7432 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7433 set_result(result);
7434 return true;
7435 }
7436
7437 /**
7438 * Calculate CRC32 for ByteBuffer.
7439 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7440 */
7441 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7442 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7443 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7444 // no receiver since it is static method
7445 Node* crc = argument(0); // type: int
7446 Node* src = argument(1); // type: long
7447 Node* offset = argument(3); // type: int
7448 Node* length = argument(4); // type: int
7449
7450 src = ConvL2X(src); // adjust Java long to machine word
7451 Node* base = _gvn.transform(new CastX2PNode(src));
7452 offset = ConvI2X(offset);
7453
7454 // 'src_start' points to src array + scaled offset
7455 Node* src_start = basic_plus_adr(top(), base, offset);
7456
7457 // Call the stub.
7458 address stubAddr = StubRoutines::updateBytesCRC32();
7459 const char *stubName = "updateBytesCRC32";
7460
7461 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7462 stubAddr, stubName, TypePtr::BOTTOM,
7463 crc, src_start, length);
7464 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7465 set_result(result);
7466 return true;
7467 }
7468
7469 //------------------------------get_table_from_crc32c_class-----------------------
7470 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7471 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7472 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7473
7474 return table;
7475 }
7476
7477 //------------------------------inline_updateBytesCRC32C-----------------------
7478 //
7479 // Calculate CRC32C for byte[] array.
7480 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7481 //
7482 bool LibraryCallKit::inline_updateBytesCRC32C() {
7483 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7484 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7485 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7486 // no receiver since it is a static method
7487 Node* crc = argument(0); // type: int
7488 Node* src = argument(1); // type: oop
7489 Node* offset = argument(2); // type: int
7490 Node* end = argument(3); // type: int
7491
7492 Node* length = _gvn.transform(new SubINode(end, offset));
7493
7494 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7495 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7496 // failed array check
7497 return false;
7498 }
7499
7500 // Figure out the size and type of the elements we will be copying.
7501 BasicType src_elem = src_type->elem()->array_element_basic_type();
7502 if (src_elem != T_BYTE) {
7503 return false;
7504 }
7505
7506 // 'src_start' points to src array + scaled offset
7507 src = must_be_not_null(src, true);
7508 Node* src_start = array_element_address(src, offset, src_elem);
7509
7510 // static final int[] byteTable in class CRC32C
7511 Node* table = get_table_from_crc32c_class(callee()->holder());
7512 table = must_be_not_null(table, true);
7513 Node* table_start = array_element_address(table, intcon(0), T_INT);
7514
7515 // We assume that range check is done by caller.
7516 // TODO: generate range check (offset+length < src.length) in debug VM.
7517
7518 // Call the stub.
7519 address stubAddr = StubRoutines::updateBytesCRC32C();
7520 const char *stubName = "updateBytesCRC32C";
7521
7522 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7523 stubAddr, stubName, TypePtr::BOTTOM,
7524 crc, src_start, length, table_start);
7525 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7526 set_result(result);
7527 return true;
7528 }
7529
7530 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7531 //
7532 // Calculate CRC32C for DirectByteBuffer.
7533 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7534 //
7535 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7536 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7537 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7538 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7539 // no receiver since it is a static method
7540 Node* crc = argument(0); // type: int
7541 Node* src = argument(1); // type: long
7542 Node* offset = argument(3); // type: int
7543 Node* end = argument(4); // type: int
7544
7545 Node* length = _gvn.transform(new SubINode(end, offset));
7546
7547 src = ConvL2X(src); // adjust Java long to machine word
7548 Node* base = _gvn.transform(new CastX2PNode(src));
7549 offset = ConvI2X(offset);
7550
7551 // 'src_start' points to src array + scaled offset
7552 Node* src_start = basic_plus_adr(top(), base, offset);
7553
7554 // static final int[] byteTable in class CRC32C
7555 Node* table = get_table_from_crc32c_class(callee()->holder());
7556 table = must_be_not_null(table, true);
7557 Node* table_start = array_element_address(table, intcon(0), T_INT);
7558
7559 // Call the stub.
7560 address stubAddr = StubRoutines::updateBytesCRC32C();
7561 const char *stubName = "updateBytesCRC32C";
7562
7563 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7564 stubAddr, stubName, TypePtr::BOTTOM,
7565 crc, src_start, length, table_start);
7566 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7567 set_result(result);
7568 return true;
7569 }
7570
7571 //------------------------------inline_updateBytesAdler32----------------------
7572 //
7573 // Calculate Adler32 checksum for byte[] array.
7574 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7575 //
7576 bool LibraryCallKit::inline_updateBytesAdler32() {
7577 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7578 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7579 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7580 // no receiver since it is static method
7581 Node* crc = argument(0); // type: int
7582 Node* src = argument(1); // type: oop
7583 Node* offset = argument(2); // type: int
7584 Node* length = argument(3); // type: int
7585
7586 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7587 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7588 // failed array check
7589 return false;
7590 }
7591
7592 // Figure out the size and type of the elements we will be copying.
7593 BasicType src_elem = src_type->elem()->array_element_basic_type();
7594 if (src_elem != T_BYTE) {
7595 return false;
7596 }
7597
7598 // 'src_start' points to src array + scaled offset
7599 Node* src_start = array_element_address(src, offset, src_elem);
7600
7601 // We assume that range check is done by caller.
7602 // TODO: generate range check (offset+length < src.length) in debug VM.
7603
7604 // Call the stub.
7605 address stubAddr = StubRoutines::updateBytesAdler32();
7606 const char *stubName = "updateBytesAdler32";
7607
7608 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7609 stubAddr, stubName, TypePtr::BOTTOM,
7610 crc, src_start, length);
7611 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7612 set_result(result);
7613 return true;
7614 }
7615
7616 //------------------------------inline_updateByteBufferAdler32---------------
7617 //
7618 // Calculate Adler32 checksum for DirectByteBuffer.
7619 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7620 //
7621 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7622 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7623 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7624 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7625 // no receiver since it is static method
7626 Node* crc = argument(0); // type: int
7627 Node* src = argument(1); // type: long
7628 Node* offset = argument(3); // type: int
7629 Node* length = argument(4); // type: int
7630
7631 src = ConvL2X(src); // adjust Java long to machine word
7632 Node* base = _gvn.transform(new CastX2PNode(src));
7633 offset = ConvI2X(offset);
7634
7635 // 'src_start' points to src array + scaled offset
7636 Node* src_start = basic_plus_adr(top(), base, offset);
7637
7638 // Call the stub.
7639 address stubAddr = StubRoutines::updateBytesAdler32();
7640 const char *stubName = "updateBytesAdler32";
7641
7642 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7643 stubAddr, stubName, TypePtr::BOTTOM,
7644 crc, src_start, length);
7645
7646 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7647 set_result(result);
7648 return true;
7649 }
7650
7651 //----------------------------inline_reference_get0----------------------------
7652 // public T java.lang.ref.Reference.get();
7653 bool LibraryCallKit::inline_reference_get0() {
7654 const int referent_offset = java_lang_ref_Reference::referent_offset();
7655
7656 // Get the argument:
7657 Node* reference_obj = null_check_receiver();
7658 if (stopped()) return true;
7659
7660 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7661 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7662 decorators, /*is_static*/ false, nullptr);
7663 if (result == nullptr) return false;
7664
7665 // Add memory barrier to prevent commoning reads from this field
7666 // across safepoint since GC can change its value.
7667 insert_mem_bar(Op_MemBarCPUOrder);
7668
7669 set_result(result);
7670 return true;
7671 }
7672
7673 //----------------------------inline_reference_refersTo0----------------------------
7674 // bool java.lang.ref.Reference.refersTo0();
7675 // bool java.lang.ref.PhantomReference.refersTo0();
7676 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7677 // Get arguments:
7678 Node* reference_obj = null_check_receiver();
7679 Node* other_obj = argument(1);
7680 if (stopped()) return true;
7681
7682 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7683 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7684 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7685 decorators, /*is_static*/ false, nullptr);
7686 if (referent == nullptr) return false;
7687
7688 // Add memory barrier to prevent commoning reads from this field
7689 // across safepoint since GC can change its value.
7690 insert_mem_bar(Op_MemBarCPUOrder);
7691
7692 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7693 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7694 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7695
7696 RegionNode* region = new RegionNode(3);
7697 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7698
7699 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7700 region->init_req(1, if_true);
7701 phi->init_req(1, intcon(1));
7702
7703 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7704 region->init_req(2, if_false);
7705 phi->init_req(2, intcon(0));
7706
7707 set_control(_gvn.transform(region));
7708 record_for_igvn(region);
7709 set_result(_gvn.transform(phi));
7710 return true;
7711 }
7712
7713 //----------------------------inline_reference_clear0----------------------------
7714 // void java.lang.ref.Reference.clear0();
7715 // void java.lang.ref.PhantomReference.clear0();
7716 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7717 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7718
7719 // Get arguments
7720 Node* reference_obj = null_check_receiver();
7721 if (stopped()) return true;
7722
7723 // Common access parameters
7724 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7725 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7726 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7727 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7728 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7729
7730 Node* referent = access_load_at(reference_obj,
7731 referent_field_addr,
7732 referent_field_addr_type,
7733 val_type,
7734 T_OBJECT,
7735 decorators);
7736
7737 IdealKit ideal(this);
7738 #define __ ideal.
7739 __ if_then(referent, BoolTest::ne, null());
7740 sync_kit(ideal);
7741 access_store_at(reference_obj,
7742 referent_field_addr,
7743 referent_field_addr_type,
7744 null(),
7745 val_type,
7746 T_OBJECT,
7747 decorators);
7748 __ sync_kit(this);
7749 __ end_if();
7750 final_sync(ideal);
7751 #undef __
7752
7753 return true;
7754 }
7755
7756 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7757 DecoratorSet decorators, bool is_static,
7758 ciInstanceKlass* fromKls) {
7759 if (fromKls == nullptr) {
7760 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7761 assert(tinst != nullptr, "obj is null");
7762 assert(tinst->is_loaded(), "obj is not loaded");
7763 fromKls = tinst->instance_klass();
7764 } else {
7765 assert(is_static, "only for static field access");
7766 }
7767 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7768 ciSymbol::make(fieldTypeString),
7769 is_static);
7770
7771 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7772 if (field == nullptr) return (Node *) nullptr;
7773
7774 if (is_static) {
7775 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7776 fromObj = makecon(tip);
7777 }
7778
7779 // Next code copied from Parse::do_get_xxx():
7780
7781 // Compute address and memory type.
7782 int offset = field->offset_in_bytes();
7783 bool is_vol = field->is_volatile();
7784 ciType* field_klass = field->type();
7785 assert(field_klass->is_loaded(), "should be loaded");
7786 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7787 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7788 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7789 "slice of address and input slice don't match");
7790 BasicType bt = field->layout_type();
7791
7792 // Build the resultant type of the load
7793 const Type *type;
7794 if (bt == T_OBJECT) {
7795 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7796 } else {
7797 type = Type::get_const_basic_type(bt);
7798 }
7799
7800 if (is_vol) {
7801 decorators |= MO_SEQ_CST;
7802 }
7803
7804 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7805 }
7806
7807 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7808 bool is_exact /* true */, bool is_static /* false */,
7809 ciInstanceKlass * fromKls /* nullptr */) {
7810 if (fromKls == nullptr) {
7811 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7812 assert(tinst != nullptr, "obj is null");
7813 assert(tinst->is_loaded(), "obj is not loaded");
7814 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7815 fromKls = tinst->instance_klass();
7816 }
7817 else {
7818 assert(is_static, "only for static field access");
7819 }
7820 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7821 ciSymbol::make(fieldTypeString),
7822 is_static);
7823
7824 assert(field != nullptr, "undefined field");
7825 assert(!field->is_volatile(), "not defined for volatile fields");
7826
7827 if (is_static) {
7828 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7829 fromObj = makecon(tip);
7830 }
7831
7832 // Next code copied from Parse::do_get_xxx():
7833
7834 // Compute address and memory type.
7835 int offset = field->offset_in_bytes();
7836 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7837
7838 return adr;
7839 }
7840
7841 //------------------------------inline_aescrypt_Block-----------------------
7842 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7843 address stubAddr = nullptr;
7844 const char *stubName;
7845 assert(UseAES, "need AES instruction support");
7846
7847 switch(id) {
7848 case vmIntrinsics::_aescrypt_encryptBlock:
7849 stubAddr = StubRoutines::aescrypt_encryptBlock();
7850 stubName = "aescrypt_encryptBlock";
7851 break;
7852 case vmIntrinsics::_aescrypt_decryptBlock:
7853 stubAddr = StubRoutines::aescrypt_decryptBlock();
7854 stubName = "aescrypt_decryptBlock";
7855 break;
7856 default:
7857 break;
7858 }
7859 if (stubAddr == nullptr) return false;
7860
7861 Node* aescrypt_object = argument(0);
7862 Node* src = argument(1);
7863 Node* src_offset = argument(2);
7864 Node* dest = argument(3);
7865 Node* dest_offset = argument(4);
7866
7867 src = must_be_not_null(src, true);
7868 dest = must_be_not_null(dest, true);
7869
7870 // (1) src and dest are arrays.
7871 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7872 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7873 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7874 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7875
7876 // for the quick and dirty code we will skip all the checks.
7877 // we are just trying to get the call to be generated.
7878 Node* src_start = src;
7879 Node* dest_start = dest;
7880 if (src_offset != nullptr || dest_offset != nullptr) {
7881 assert(src_offset != nullptr && dest_offset != nullptr, "");
7882 src_start = array_element_address(src, src_offset, T_BYTE);
7883 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7884 }
7885
7886 // now need to get the start of its expanded key array
7887 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7888 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7889 if (k_start == nullptr) return false;
7890
7891 // Call the stub.
7892 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7893 stubAddr, stubName, TypePtr::BOTTOM,
7894 src_start, dest_start, k_start);
7895
7896 return true;
7897 }
7898
7899 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7900 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7901 address stubAddr = nullptr;
7902 const char *stubName = nullptr;
7903
7904 assert(UseAES, "need AES instruction support");
7905
7906 switch(id) {
7907 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7908 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7909 stubName = "cipherBlockChaining_encryptAESCrypt";
7910 break;
7911 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7912 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7913 stubName = "cipherBlockChaining_decryptAESCrypt";
7914 break;
7915 default:
7916 break;
7917 }
7918 if (stubAddr == nullptr) return false;
7919
7920 Node* cipherBlockChaining_object = argument(0);
7921 Node* src = argument(1);
7922 Node* src_offset = argument(2);
7923 Node* len = argument(3);
7924 Node* dest = argument(4);
7925 Node* dest_offset = argument(5);
7926
7927 src = must_be_not_null(src, false);
7928 dest = must_be_not_null(dest, false);
7929
7930 // (1) src and dest are arrays.
7931 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7932 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7933 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7934 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7935
7936 // checks are the responsibility of the caller
7937 Node* src_start = src;
7938 Node* dest_start = dest;
7939 if (src_offset != nullptr || dest_offset != nullptr) {
7940 assert(src_offset != nullptr && dest_offset != nullptr, "");
7941 src_start = array_element_address(src, src_offset, T_BYTE);
7942 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7943 }
7944
7945 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7946 // (because of the predicated logic executed earlier).
7947 // so we cast it here safely.
7948 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7949
7950 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7951 if (embeddedCipherObj == nullptr) return false;
7952
7953 // cast it to what we know it will be at runtime
7954 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7955 assert(tinst != nullptr, "CBC obj is null");
7956 assert(tinst->is_loaded(), "CBC obj is not loaded");
7957 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7958 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7959
7960 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7961 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7962 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7963 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7964 aescrypt_object = _gvn.transform(aescrypt_object);
7965
7966 // we need to get the start of the aescrypt_object's expanded key array
7967 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7968 if (k_start == nullptr) return false;
7969
7970 // similarly, get the start address of the r vector
7971 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7972 if (objRvec == nullptr) return false;
7973 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7974
7975 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7976 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7977 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7978 stubAddr, stubName, TypePtr::BOTTOM,
7979 src_start, dest_start, k_start, r_start, len);
7980
7981 // return cipher length (int)
7982 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7983 set_result(retvalue);
7984 return true;
7985 }
7986
7987 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7988 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7989 address stubAddr = nullptr;
7990 const char *stubName = nullptr;
7991
7992 assert(UseAES, "need AES instruction support");
7993
7994 switch (id) {
7995 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7996 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7997 stubName = "electronicCodeBook_encryptAESCrypt";
7998 break;
7999 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8000 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8001 stubName = "electronicCodeBook_decryptAESCrypt";
8002 break;
8003 default:
8004 break;
8005 }
8006
8007 if (stubAddr == nullptr) return false;
8008
8009 Node* electronicCodeBook_object = argument(0);
8010 Node* src = argument(1);
8011 Node* src_offset = argument(2);
8012 Node* len = argument(3);
8013 Node* dest = argument(4);
8014 Node* dest_offset = argument(5);
8015
8016 // (1) src and dest are arrays.
8017 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8018 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8019 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8020 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8021
8022 // checks are the responsibility of the caller
8023 Node* src_start = src;
8024 Node* dest_start = dest;
8025 if (src_offset != nullptr || dest_offset != nullptr) {
8026 assert(src_offset != nullptr && dest_offset != nullptr, "");
8027 src_start = array_element_address(src, src_offset, T_BYTE);
8028 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8029 }
8030
8031 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8032 // (because of the predicated logic executed earlier).
8033 // so we cast it here safely.
8034 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8035
8036 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8037 if (embeddedCipherObj == nullptr) return false;
8038
8039 // cast it to what we know it will be at runtime
8040 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8041 assert(tinst != nullptr, "ECB obj is null");
8042 assert(tinst->is_loaded(), "ECB obj is not loaded");
8043 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8044 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8045
8046 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8047 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8048 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8049 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8050 aescrypt_object = _gvn.transform(aescrypt_object);
8051
8052 // we need to get the start of the aescrypt_object's expanded key array
8053 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8054 if (k_start == nullptr) return false;
8055
8056 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8057 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8058 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8059 stubAddr, stubName, TypePtr::BOTTOM,
8060 src_start, dest_start, k_start, len);
8061
8062 // return cipher length (int)
8063 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8064 set_result(retvalue);
8065 return true;
8066 }
8067
8068 //------------------------------inline_counterMode_AESCrypt-----------------------
8069 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8070 assert(UseAES, "need AES instruction support");
8071 if (!UseAESCTRIntrinsics) return false;
8072
8073 address stubAddr = nullptr;
8074 const char *stubName = nullptr;
8075 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8076 stubAddr = StubRoutines::counterMode_AESCrypt();
8077 stubName = "counterMode_AESCrypt";
8078 }
8079 if (stubAddr == nullptr) return false;
8080
8081 Node* counterMode_object = argument(0);
8082 Node* src = argument(1);
8083 Node* src_offset = argument(2);
8084 Node* len = argument(3);
8085 Node* dest = argument(4);
8086 Node* dest_offset = argument(5);
8087
8088 // (1) src and dest are arrays.
8089 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8090 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8091 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8092 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8093
8094 // checks are the responsibility of the caller
8095 Node* src_start = src;
8096 Node* dest_start = dest;
8097 if (src_offset != nullptr || dest_offset != nullptr) {
8098 assert(src_offset != nullptr && dest_offset != nullptr, "");
8099 src_start = array_element_address(src, src_offset, T_BYTE);
8100 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8101 }
8102
8103 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8104 // (because of the predicated logic executed earlier).
8105 // so we cast it here safely.
8106 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8107 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8108 if (embeddedCipherObj == nullptr) return false;
8109 // cast it to what we know it will be at runtime
8110 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8111 assert(tinst != nullptr, "CTR obj is null");
8112 assert(tinst->is_loaded(), "CTR obj is not loaded");
8113 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8114 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8115 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8116 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8117 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8118 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8119 aescrypt_object = _gvn.transform(aescrypt_object);
8120 // we need to get the start of the aescrypt_object's expanded key array
8121 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8122 if (k_start == nullptr) return false;
8123 // similarly, get the start address of the r vector
8124 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8125 if (obj_counter == nullptr) return false;
8126 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8127
8128 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8129 if (saved_encCounter == nullptr) return false;
8130 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8131 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8132
8133 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8134 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8135 OptoRuntime::counterMode_aescrypt_Type(),
8136 stubAddr, stubName, TypePtr::BOTTOM,
8137 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8138
8139 // return cipher length (int)
8140 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8141 set_result(retvalue);
8142 return true;
8143 }
8144
8145 //------------------------------get_key_start_from_aescrypt_object-----------------------
8146 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
8147 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8148 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8149 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8150 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8151 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
8152 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
8153 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8154 if (objSessionK == nullptr) {
8155 return (Node *) nullptr;
8156 }
8157 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
8158 #else
8159 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
8160 #endif // PPC64
8161 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
8162 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8163
8164 // now have the array, need to get the start address of the K array
8165 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8166 return k_start;
8167 }
8168
8169 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8170 // Return node representing slow path of predicate check.
8171 // the pseudo code we want to emulate with this predicate is:
8172 // for encryption:
8173 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8174 // for decryption:
8175 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8176 // note cipher==plain is more conservative than the original java code but that's OK
8177 //
8178 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8179 // The receiver was checked for null already.
8180 Node* objCBC = argument(0);
8181
8182 Node* src = argument(1);
8183 Node* dest = argument(4);
8184
8185 // Load embeddedCipher field of CipherBlockChaining object.
8186 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8187
8188 // get AESCrypt klass for instanceOf check
8189 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8190 // will have same classloader as CipherBlockChaining object
8191 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8192 assert(tinst != nullptr, "CBCobj is null");
8193 assert(tinst->is_loaded(), "CBCobj is not loaded");
8194
8195 // we want to do an instanceof comparison against the AESCrypt class
8196 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8197 if (!klass_AESCrypt->is_loaded()) {
8198 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8199 Node* ctrl = control();
8200 set_control(top()); // no regular fast path
8201 return ctrl;
8202 }
8203
8204 src = must_be_not_null(src, true);
8205 dest = must_be_not_null(dest, true);
8206
8207 // Resolve oops to stable for CmpP below.
8208 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8209
8210 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8211 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8212 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8213
8214 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8215
8216 // for encryption, we are done
8217 if (!decrypting)
8218 return instof_false; // even if it is null
8219
8220 // for decryption, we need to add a further check to avoid
8221 // taking the intrinsic path when cipher and plain are the same
8222 // see the original java code for why.
8223 RegionNode* region = new RegionNode(3);
8224 region->init_req(1, instof_false);
8225
8226 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8227 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8228 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8229 region->init_req(2, src_dest_conjoint);
8230
8231 record_for_igvn(region);
8232 return _gvn.transform(region);
8233 }
8234
8235 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8236 // Return node representing slow path of predicate check.
8237 // the pseudo code we want to emulate with this predicate is:
8238 // for encryption:
8239 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8240 // for decryption:
8241 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8242 // note cipher==plain is more conservative than the original java code but that's OK
8243 //
8244 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8245 // The receiver was checked for null already.
8246 Node* objECB = argument(0);
8247
8248 // Load embeddedCipher field of ElectronicCodeBook object.
8249 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8250
8251 // get AESCrypt klass for instanceOf check
8252 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8253 // will have same classloader as ElectronicCodeBook object
8254 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8255 assert(tinst != nullptr, "ECBobj is null");
8256 assert(tinst->is_loaded(), "ECBobj is not loaded");
8257
8258 // we want to do an instanceof comparison against the AESCrypt class
8259 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8260 if (!klass_AESCrypt->is_loaded()) {
8261 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8262 Node* ctrl = control();
8263 set_control(top()); // no regular fast path
8264 return ctrl;
8265 }
8266 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8267
8268 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8269 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8270 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8271
8272 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8273
8274 // for encryption, we are done
8275 if (!decrypting)
8276 return instof_false; // even if it is null
8277
8278 // for decryption, we need to add a further check to avoid
8279 // taking the intrinsic path when cipher and plain are the same
8280 // see the original java code for why.
8281 RegionNode* region = new RegionNode(3);
8282 region->init_req(1, instof_false);
8283 Node* src = argument(1);
8284 Node* dest = argument(4);
8285 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8286 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8287 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8288 region->init_req(2, src_dest_conjoint);
8289
8290 record_for_igvn(region);
8291 return _gvn.transform(region);
8292 }
8293
8294 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8295 // Return node representing slow path of predicate check.
8296 // the pseudo code we want to emulate with this predicate is:
8297 // for encryption:
8298 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8299 // for decryption:
8300 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8301 // note cipher==plain is more conservative than the original java code but that's OK
8302 //
8303
8304 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8305 // The receiver was checked for null already.
8306 Node* objCTR = argument(0);
8307
8308 // Load embeddedCipher field of CipherBlockChaining object.
8309 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8310
8311 // get AESCrypt klass for instanceOf check
8312 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8313 // will have same classloader as CipherBlockChaining object
8314 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8315 assert(tinst != nullptr, "CTRobj is null");
8316 assert(tinst->is_loaded(), "CTRobj is not loaded");
8317
8318 // we want to do an instanceof comparison against the AESCrypt class
8319 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8320 if (!klass_AESCrypt->is_loaded()) {
8321 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8322 Node* ctrl = control();
8323 set_control(top()); // no regular fast path
8324 return ctrl;
8325 }
8326
8327 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8328 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8329 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8330 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8331 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8332
8333 return instof_false; // even if it is null
8334 }
8335
8336 //------------------------------inline_ghash_processBlocks
8337 bool LibraryCallKit::inline_ghash_processBlocks() {
8338 address stubAddr;
8339 const char *stubName;
8340 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8341
8342 stubAddr = StubRoutines::ghash_processBlocks();
8343 stubName = "ghash_processBlocks";
8344
8345 Node* data = argument(0);
8346 Node* offset = argument(1);
8347 Node* len = argument(2);
8348 Node* state = argument(3);
8349 Node* subkeyH = argument(4);
8350
8351 state = must_be_not_null(state, true);
8352 subkeyH = must_be_not_null(subkeyH, true);
8353 data = must_be_not_null(data, true);
8354
8355 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8356 assert(state_start, "state is null");
8357 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8358 assert(subkeyH_start, "subkeyH is null");
8359 Node* data_start = array_element_address(data, offset, T_BYTE);
8360 assert(data_start, "data is null");
8361
8362 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8363 OptoRuntime::ghash_processBlocks_Type(),
8364 stubAddr, stubName, TypePtr::BOTTOM,
8365 state_start, subkeyH_start, data_start, len);
8366 return true;
8367 }
8368
8369 //------------------------------inline_chacha20Block
8370 bool LibraryCallKit::inline_chacha20Block() {
8371 address stubAddr;
8372 const char *stubName;
8373 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8374
8375 stubAddr = StubRoutines::chacha20Block();
8376 stubName = "chacha20Block";
8377
8378 Node* state = argument(0);
8379 Node* result = argument(1);
8380
8381 state = must_be_not_null(state, true);
8382 result = must_be_not_null(result, true);
8383
8384 Node* state_start = array_element_address(state, intcon(0), T_INT);
8385 assert(state_start, "state is null");
8386 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8387 assert(result_start, "result is null");
8388
8389 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8390 OptoRuntime::chacha20Block_Type(),
8391 stubAddr, stubName, TypePtr::BOTTOM,
8392 state_start, result_start);
8393 // return key stream length (int)
8394 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8395 set_result(retvalue);
8396 return true;
8397 }
8398
8399 //------------------------------inline_kyberNtt
8400 bool LibraryCallKit::inline_kyberNtt() {
8401 address stubAddr;
8402 const char *stubName;
8403 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8404 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8405
8406 stubAddr = StubRoutines::kyberNtt();
8407 stubName = "kyberNtt";
8408 if (!stubAddr) return false;
8409
8410 Node* coeffs = argument(0);
8411 Node* ntt_zetas = argument(1);
8412
8413 coeffs = must_be_not_null(coeffs, true);
8414 ntt_zetas = must_be_not_null(ntt_zetas, true);
8415
8416 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8417 assert(coeffs_start, "coeffs is null");
8418 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8419 assert(ntt_zetas_start, "ntt_zetas is null");
8420 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8421 OptoRuntime::kyberNtt_Type(),
8422 stubAddr, stubName, TypePtr::BOTTOM,
8423 coeffs_start, ntt_zetas_start);
8424 // return an int
8425 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8426 set_result(retvalue);
8427 return true;
8428 }
8429
8430 //------------------------------inline_kyberInverseNtt
8431 bool LibraryCallKit::inline_kyberInverseNtt() {
8432 address stubAddr;
8433 const char *stubName;
8434 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8435 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8436
8437 stubAddr = StubRoutines::kyberInverseNtt();
8438 stubName = "kyberInverseNtt";
8439 if (!stubAddr) return false;
8440
8441 Node* coeffs = argument(0);
8442 Node* zetas = argument(1);
8443
8444 coeffs = must_be_not_null(coeffs, true);
8445 zetas = must_be_not_null(zetas, true);
8446
8447 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8448 assert(coeffs_start, "coeffs is null");
8449 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8450 assert(zetas_start, "inverseNtt_zetas is null");
8451 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8452 OptoRuntime::kyberInverseNtt_Type(),
8453 stubAddr, stubName, TypePtr::BOTTOM,
8454 coeffs_start, zetas_start);
8455
8456 // return an int
8457 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8458 set_result(retvalue);
8459 return true;
8460 }
8461
8462 //------------------------------inline_kyberNttMult
8463 bool LibraryCallKit::inline_kyberNttMult() {
8464 address stubAddr;
8465 const char *stubName;
8466 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8467 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8468
8469 stubAddr = StubRoutines::kyberNttMult();
8470 stubName = "kyberNttMult";
8471 if (!stubAddr) return false;
8472
8473 Node* result = argument(0);
8474 Node* ntta = argument(1);
8475 Node* nttb = argument(2);
8476 Node* zetas = argument(3);
8477
8478 result = must_be_not_null(result, true);
8479 ntta = must_be_not_null(ntta, true);
8480 nttb = must_be_not_null(nttb, true);
8481 zetas = must_be_not_null(zetas, true);
8482
8483 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8484 assert(result_start, "result is null");
8485 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8486 assert(ntta_start, "ntta is null");
8487 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8488 assert(nttb_start, "nttb is null");
8489 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8490 assert(zetas_start, "nttMult_zetas is null");
8491 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8492 OptoRuntime::kyberNttMult_Type(),
8493 stubAddr, stubName, TypePtr::BOTTOM,
8494 result_start, ntta_start, nttb_start,
8495 zetas_start);
8496
8497 // return an int
8498 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8499 set_result(retvalue);
8500
8501 return true;
8502 }
8503
8504 //------------------------------inline_kyberAddPoly_2
8505 bool LibraryCallKit::inline_kyberAddPoly_2() {
8506 address stubAddr;
8507 const char *stubName;
8508 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8509 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8510
8511 stubAddr = StubRoutines::kyberAddPoly_2();
8512 stubName = "kyberAddPoly_2";
8513 if (!stubAddr) return false;
8514
8515 Node* result = argument(0);
8516 Node* a = argument(1);
8517 Node* b = argument(2);
8518
8519 result = must_be_not_null(result, true);
8520 a = must_be_not_null(a, true);
8521 b = must_be_not_null(b, true);
8522
8523 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8524 assert(result_start, "result is null");
8525 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8526 assert(a_start, "a is null");
8527 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8528 assert(b_start, "b is null");
8529 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8530 OptoRuntime::kyberAddPoly_2_Type(),
8531 stubAddr, stubName, TypePtr::BOTTOM,
8532 result_start, a_start, b_start);
8533 // return an int
8534 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8535 set_result(retvalue);
8536 return true;
8537 }
8538
8539 //------------------------------inline_kyberAddPoly_3
8540 bool LibraryCallKit::inline_kyberAddPoly_3() {
8541 address stubAddr;
8542 const char *stubName;
8543 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8544 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8545
8546 stubAddr = StubRoutines::kyberAddPoly_3();
8547 stubName = "kyberAddPoly_3";
8548 if (!stubAddr) return false;
8549
8550 Node* result = argument(0);
8551 Node* a = argument(1);
8552 Node* b = argument(2);
8553 Node* c = argument(3);
8554
8555 result = must_be_not_null(result, true);
8556 a = must_be_not_null(a, true);
8557 b = must_be_not_null(b, true);
8558 c = must_be_not_null(c, true);
8559
8560 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8561 assert(result_start, "result is null");
8562 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8563 assert(a_start, "a is null");
8564 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8565 assert(b_start, "b is null");
8566 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8567 assert(c_start, "c is null");
8568 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8569 OptoRuntime::kyberAddPoly_3_Type(),
8570 stubAddr, stubName, TypePtr::BOTTOM,
8571 result_start, a_start, b_start, c_start);
8572 // return an int
8573 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8574 set_result(retvalue);
8575 return true;
8576 }
8577
8578 //------------------------------inline_kyber12To16
8579 bool LibraryCallKit::inline_kyber12To16() {
8580 address stubAddr;
8581 const char *stubName;
8582 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8583 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8584
8585 stubAddr = StubRoutines::kyber12To16();
8586 stubName = "kyber12To16";
8587 if (!stubAddr) return false;
8588
8589 Node* condensed = argument(0);
8590 Node* condensedOffs = argument(1);
8591 Node* parsed = argument(2);
8592 Node* parsedLength = argument(3);
8593
8594 condensed = must_be_not_null(condensed, true);
8595 parsed = must_be_not_null(parsed, true);
8596
8597 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8598 assert(condensed_start, "condensed is null");
8599 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8600 assert(parsed_start, "parsed is null");
8601 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8602 OptoRuntime::kyber12To16_Type(),
8603 stubAddr, stubName, TypePtr::BOTTOM,
8604 condensed_start, condensedOffs, parsed_start, parsedLength);
8605 // return an int
8606 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8607 set_result(retvalue);
8608 return true;
8609
8610 }
8611
8612 //------------------------------inline_kyberBarrettReduce
8613 bool LibraryCallKit::inline_kyberBarrettReduce() {
8614 address stubAddr;
8615 const char *stubName;
8616 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8617 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8618
8619 stubAddr = StubRoutines::kyberBarrettReduce();
8620 stubName = "kyberBarrettReduce";
8621 if (!stubAddr) return false;
8622
8623 Node* coeffs = argument(0);
8624
8625 coeffs = must_be_not_null(coeffs, true);
8626
8627 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8628 assert(coeffs_start, "coeffs is null");
8629 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8630 OptoRuntime::kyberBarrettReduce_Type(),
8631 stubAddr, stubName, TypePtr::BOTTOM,
8632 coeffs_start);
8633 // return an int
8634 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8635 set_result(retvalue);
8636 return true;
8637 }
8638
8639 //------------------------------inline_dilithiumAlmostNtt
8640 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8641 address stubAddr;
8642 const char *stubName;
8643 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8644 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8645
8646 stubAddr = StubRoutines::dilithiumAlmostNtt();
8647 stubName = "dilithiumAlmostNtt";
8648 if (!stubAddr) return false;
8649
8650 Node* coeffs = argument(0);
8651 Node* ntt_zetas = argument(1);
8652
8653 coeffs = must_be_not_null(coeffs, true);
8654 ntt_zetas = must_be_not_null(ntt_zetas, true);
8655
8656 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8657 assert(coeffs_start, "coeffs is null");
8658 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8659 assert(ntt_zetas_start, "ntt_zetas is null");
8660 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8661 OptoRuntime::dilithiumAlmostNtt_Type(),
8662 stubAddr, stubName, TypePtr::BOTTOM,
8663 coeffs_start, ntt_zetas_start);
8664 // return an int
8665 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8666 set_result(retvalue);
8667 return true;
8668 }
8669
8670 //------------------------------inline_dilithiumAlmostInverseNtt
8671 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8672 address stubAddr;
8673 const char *stubName;
8674 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8675 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8676
8677 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8678 stubName = "dilithiumAlmostInverseNtt";
8679 if (!stubAddr) return false;
8680
8681 Node* coeffs = argument(0);
8682 Node* zetas = argument(1);
8683
8684 coeffs = must_be_not_null(coeffs, true);
8685 zetas = must_be_not_null(zetas, true);
8686
8687 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8688 assert(coeffs_start, "coeffs is null");
8689 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8690 assert(zetas_start, "inverseNtt_zetas is null");
8691 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8692 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8693 stubAddr, stubName, TypePtr::BOTTOM,
8694 coeffs_start, zetas_start);
8695 // return an int
8696 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8697 set_result(retvalue);
8698 return true;
8699 }
8700
8701 //------------------------------inline_dilithiumNttMult
8702 bool LibraryCallKit::inline_dilithiumNttMult() {
8703 address stubAddr;
8704 const char *stubName;
8705 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8706 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8707
8708 stubAddr = StubRoutines::dilithiumNttMult();
8709 stubName = "dilithiumNttMult";
8710 if (!stubAddr) return false;
8711
8712 Node* result = argument(0);
8713 Node* ntta = argument(1);
8714 Node* nttb = argument(2);
8715 Node* zetas = argument(3);
8716
8717 result = must_be_not_null(result, true);
8718 ntta = must_be_not_null(ntta, true);
8719 nttb = must_be_not_null(nttb, true);
8720 zetas = must_be_not_null(zetas, true);
8721
8722 Node* result_start = array_element_address(result, intcon(0), T_INT);
8723 assert(result_start, "result is null");
8724 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8725 assert(ntta_start, "ntta is null");
8726 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8727 assert(nttb_start, "nttb is null");
8728 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8729 OptoRuntime::dilithiumNttMult_Type(),
8730 stubAddr, stubName, TypePtr::BOTTOM,
8731 result_start, ntta_start, nttb_start);
8732
8733 // return an int
8734 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8735 set_result(retvalue);
8736
8737 return true;
8738 }
8739
8740 //------------------------------inline_dilithiumMontMulByConstant
8741 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8742 address stubAddr;
8743 const char *stubName;
8744 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8745 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8746
8747 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8748 stubName = "dilithiumMontMulByConstant";
8749 if (!stubAddr) return false;
8750
8751 Node* coeffs = argument(0);
8752 Node* constant = argument(1);
8753
8754 coeffs = must_be_not_null(coeffs, true);
8755
8756 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8757 assert(coeffs_start, "coeffs is null");
8758 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8759 OptoRuntime::dilithiumMontMulByConstant_Type(),
8760 stubAddr, stubName, TypePtr::BOTTOM,
8761 coeffs_start, constant);
8762
8763 // return an int
8764 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8765 set_result(retvalue);
8766 return true;
8767 }
8768
8769
8770 //------------------------------inline_dilithiumDecomposePoly
8771 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8772 address stubAddr;
8773 const char *stubName;
8774 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8775 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8776
8777 stubAddr = StubRoutines::dilithiumDecomposePoly();
8778 stubName = "dilithiumDecomposePoly";
8779 if (!stubAddr) return false;
8780
8781 Node* input = argument(0);
8782 Node* lowPart = argument(1);
8783 Node* highPart = argument(2);
8784 Node* twoGamma2 = argument(3);
8785 Node* multiplier = argument(4);
8786
8787 input = must_be_not_null(input, true);
8788 lowPart = must_be_not_null(lowPart, true);
8789 highPart = must_be_not_null(highPart, true);
8790
8791 Node* input_start = array_element_address(input, intcon(0), T_INT);
8792 assert(input_start, "input is null");
8793 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8794 assert(lowPart_start, "lowPart is null");
8795 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8796 assert(highPart_start, "highPart is null");
8797
8798 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8799 OptoRuntime::dilithiumDecomposePoly_Type(),
8800 stubAddr, stubName, TypePtr::BOTTOM,
8801 input_start, lowPart_start, highPart_start,
8802 twoGamma2, multiplier);
8803
8804 // return an int
8805 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8806 set_result(retvalue);
8807 return true;
8808 }
8809
8810 bool LibraryCallKit::inline_base64_encodeBlock() {
8811 address stubAddr;
8812 const char *stubName;
8813 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8814 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8815 stubAddr = StubRoutines::base64_encodeBlock();
8816 stubName = "encodeBlock";
8817
8818 if (!stubAddr) return false;
8819 Node* base64obj = argument(0);
8820 Node* src = argument(1);
8821 Node* offset = argument(2);
8822 Node* len = argument(3);
8823 Node* dest = argument(4);
8824 Node* dp = argument(5);
8825 Node* isURL = argument(6);
8826
8827 src = must_be_not_null(src, true);
8828 dest = must_be_not_null(dest, true);
8829
8830 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8831 assert(src_start, "source array is null");
8832 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8833 assert(dest_start, "destination array is null");
8834
8835 Node* base64 = make_runtime_call(RC_LEAF,
8836 OptoRuntime::base64_encodeBlock_Type(),
8837 stubAddr, stubName, TypePtr::BOTTOM,
8838 src_start, offset, len, dest_start, dp, isURL);
8839 return true;
8840 }
8841
8842 bool LibraryCallKit::inline_base64_decodeBlock() {
8843 address stubAddr;
8844 const char *stubName;
8845 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8846 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8847 stubAddr = StubRoutines::base64_decodeBlock();
8848 stubName = "decodeBlock";
8849
8850 if (!stubAddr) return false;
8851 Node* base64obj = argument(0);
8852 Node* src = argument(1);
8853 Node* src_offset = argument(2);
8854 Node* len = argument(3);
8855 Node* dest = argument(4);
8856 Node* dest_offset = argument(5);
8857 Node* isURL = argument(6);
8858 Node* isMIME = argument(7);
8859
8860 src = must_be_not_null(src, true);
8861 dest = must_be_not_null(dest, true);
8862
8863 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8864 assert(src_start, "source array is null");
8865 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8866 assert(dest_start, "destination array is null");
8867
8868 Node* call = make_runtime_call(RC_LEAF,
8869 OptoRuntime::base64_decodeBlock_Type(),
8870 stubAddr, stubName, TypePtr::BOTTOM,
8871 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8872 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8873 set_result(result);
8874 return true;
8875 }
8876
8877 bool LibraryCallKit::inline_poly1305_processBlocks() {
8878 address stubAddr;
8879 const char *stubName;
8880 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8881 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8882 stubAddr = StubRoutines::poly1305_processBlocks();
8883 stubName = "poly1305_processBlocks";
8884
8885 if (!stubAddr) return false;
8886 null_check_receiver(); // null-check receiver
8887 if (stopped()) return true;
8888
8889 Node* input = argument(1);
8890 Node* input_offset = argument(2);
8891 Node* len = argument(3);
8892 Node* alimbs = argument(4);
8893 Node* rlimbs = argument(5);
8894
8895 input = must_be_not_null(input, true);
8896 alimbs = must_be_not_null(alimbs, true);
8897 rlimbs = must_be_not_null(rlimbs, true);
8898
8899 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8900 assert(input_start, "input array is null");
8901 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8902 assert(acc_start, "acc array is null");
8903 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8904 assert(r_start, "r array is null");
8905
8906 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8907 OptoRuntime::poly1305_processBlocks_Type(),
8908 stubAddr, stubName, TypePtr::BOTTOM,
8909 input_start, len, acc_start, r_start);
8910 return true;
8911 }
8912
8913 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8914 address stubAddr;
8915 const char *stubName;
8916 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8917 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8918 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8919 stubName = "intpoly_montgomeryMult_P256";
8920
8921 if (!stubAddr) return false;
8922 null_check_receiver(); // null-check receiver
8923 if (stopped()) return true;
8924
8925 Node* a = argument(1);
8926 Node* b = argument(2);
8927 Node* r = argument(3);
8928
8929 a = must_be_not_null(a, true);
8930 b = must_be_not_null(b, true);
8931 r = must_be_not_null(r, true);
8932
8933 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8934 assert(a_start, "a array is null");
8935 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8936 assert(b_start, "b array is null");
8937 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8938 assert(r_start, "r array is null");
8939
8940 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8941 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8942 stubAddr, stubName, TypePtr::BOTTOM,
8943 a_start, b_start, r_start);
8944 return true;
8945 }
8946
8947 bool LibraryCallKit::inline_intpoly_assign() {
8948 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8949 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8950 const char *stubName = "intpoly_assign";
8951 address stubAddr = StubRoutines::intpoly_assign();
8952 if (!stubAddr) return false;
8953
8954 Node* set = argument(0);
8955 Node* a = argument(1);
8956 Node* b = argument(2);
8957 Node* arr_length = load_array_length(a);
8958
8959 a = must_be_not_null(a, true);
8960 b = must_be_not_null(b, true);
8961
8962 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8963 assert(a_start, "a array is null");
8964 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8965 assert(b_start, "b array is null");
8966
8967 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8968 OptoRuntime::intpoly_assign_Type(),
8969 stubAddr, stubName, TypePtr::BOTTOM,
8970 set, a_start, b_start, arr_length);
8971 return true;
8972 }
8973
8974 //------------------------------inline_digestBase_implCompress-----------------------
8975 //
8976 // Calculate MD5 for single-block byte[] array.
8977 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8978 //
8979 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8980 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8981 //
8982 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8983 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8984 //
8985 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8986 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8987 //
8988 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8989 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8990 //
8991 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8992 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8993
8994 Node* digestBase_obj = argument(0);
8995 Node* src = argument(1); // type oop
8996 Node* ofs = argument(2); // type int
8997
8998 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8999 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9000 // failed array check
9001 return false;
9002 }
9003 // Figure out the size and type of the elements we will be copying.
9004 BasicType src_elem = src_type->elem()->array_element_basic_type();
9005 if (src_elem != T_BYTE) {
9006 return false;
9007 }
9008 // 'src_start' points to src array + offset
9009 src = must_be_not_null(src, true);
9010 Node* src_start = array_element_address(src, ofs, src_elem);
9011 Node* state = nullptr;
9012 Node* block_size = nullptr;
9013 address stubAddr;
9014 const char *stubName;
9015
9016 switch(id) {
9017 case vmIntrinsics::_md5_implCompress:
9018 assert(UseMD5Intrinsics, "need MD5 instruction support");
9019 state = get_state_from_digest_object(digestBase_obj, T_INT);
9020 stubAddr = StubRoutines::md5_implCompress();
9021 stubName = "md5_implCompress";
9022 break;
9023 case vmIntrinsics::_sha_implCompress:
9024 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9025 state = get_state_from_digest_object(digestBase_obj, T_INT);
9026 stubAddr = StubRoutines::sha1_implCompress();
9027 stubName = "sha1_implCompress";
9028 break;
9029 case vmIntrinsics::_sha2_implCompress:
9030 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9031 state = get_state_from_digest_object(digestBase_obj, T_INT);
9032 stubAddr = StubRoutines::sha256_implCompress();
9033 stubName = "sha256_implCompress";
9034 break;
9035 case vmIntrinsics::_sha5_implCompress:
9036 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9037 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9038 stubAddr = StubRoutines::sha512_implCompress();
9039 stubName = "sha512_implCompress";
9040 break;
9041 case vmIntrinsics::_sha3_implCompress:
9042 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9043 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9044 stubAddr = StubRoutines::sha3_implCompress();
9045 stubName = "sha3_implCompress";
9046 block_size = get_block_size_from_digest_object(digestBase_obj);
9047 if (block_size == nullptr) return false;
9048 break;
9049 default:
9050 fatal_unexpected_iid(id);
9051 return false;
9052 }
9053 if (state == nullptr) return false;
9054
9055 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9056 if (stubAddr == nullptr) return false;
9057
9058 // Call the stub.
9059 Node* call;
9060 if (block_size == nullptr) {
9061 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9062 stubAddr, stubName, TypePtr::BOTTOM,
9063 src_start, state);
9064 } else {
9065 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9066 stubAddr, stubName, TypePtr::BOTTOM,
9067 src_start, state, block_size);
9068 }
9069
9070 return true;
9071 }
9072
9073 //------------------------------inline_double_keccak
9074 bool LibraryCallKit::inline_double_keccak() {
9075 address stubAddr;
9076 const char *stubName;
9077 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9078 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9079
9080 stubAddr = StubRoutines::double_keccak();
9081 stubName = "double_keccak";
9082 if (!stubAddr) return false;
9083
9084 Node* status0 = argument(0);
9085 Node* status1 = argument(1);
9086
9087 status0 = must_be_not_null(status0, true);
9088 status1 = must_be_not_null(status1, true);
9089
9090 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9091 assert(status0_start, "status0 is null");
9092 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9093 assert(status1_start, "status1 is null");
9094 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9095 OptoRuntime::double_keccak_Type(),
9096 stubAddr, stubName, TypePtr::BOTTOM,
9097 status0_start, status1_start);
9098 // return an int
9099 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9100 set_result(retvalue);
9101 return true;
9102 }
9103
9104
9105 //------------------------------inline_digestBase_implCompressMB-----------------------
9106 //
9107 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9108 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9109 //
9110 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9111 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9112 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9113 assert((uint)predicate < 5, "sanity");
9114 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9115
9116 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9117 Node* src = argument(1); // byte[] array
9118 Node* ofs = argument(2); // type int
9119 Node* limit = argument(3); // type int
9120
9121 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9122 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9123 // failed array check
9124 return false;
9125 }
9126 // Figure out the size and type of the elements we will be copying.
9127 BasicType src_elem = src_type->elem()->array_element_basic_type();
9128 if (src_elem != T_BYTE) {
9129 return false;
9130 }
9131 // 'src_start' points to src array + offset
9132 src = must_be_not_null(src, false);
9133 Node* src_start = array_element_address(src, ofs, src_elem);
9134
9135 const char* klass_digestBase_name = nullptr;
9136 const char* stub_name = nullptr;
9137 address stub_addr = nullptr;
9138 BasicType elem_type = T_INT;
9139
9140 switch (predicate) {
9141 case 0:
9142 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9143 klass_digestBase_name = "sun/security/provider/MD5";
9144 stub_name = "md5_implCompressMB";
9145 stub_addr = StubRoutines::md5_implCompressMB();
9146 }
9147 break;
9148 case 1:
9149 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9150 klass_digestBase_name = "sun/security/provider/SHA";
9151 stub_name = "sha1_implCompressMB";
9152 stub_addr = StubRoutines::sha1_implCompressMB();
9153 }
9154 break;
9155 case 2:
9156 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9157 klass_digestBase_name = "sun/security/provider/SHA2";
9158 stub_name = "sha256_implCompressMB";
9159 stub_addr = StubRoutines::sha256_implCompressMB();
9160 }
9161 break;
9162 case 3:
9163 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9164 klass_digestBase_name = "sun/security/provider/SHA5";
9165 stub_name = "sha512_implCompressMB";
9166 stub_addr = StubRoutines::sha512_implCompressMB();
9167 elem_type = T_LONG;
9168 }
9169 break;
9170 case 4:
9171 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9172 klass_digestBase_name = "sun/security/provider/SHA3";
9173 stub_name = "sha3_implCompressMB";
9174 stub_addr = StubRoutines::sha3_implCompressMB();
9175 elem_type = T_LONG;
9176 }
9177 break;
9178 default:
9179 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9180 }
9181 if (klass_digestBase_name != nullptr) {
9182 assert(stub_addr != nullptr, "Stub is generated");
9183 if (stub_addr == nullptr) return false;
9184
9185 // get DigestBase klass to lookup for SHA klass
9186 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9187 assert(tinst != nullptr, "digestBase_obj is not instance???");
9188 assert(tinst->is_loaded(), "DigestBase is not loaded");
9189
9190 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9191 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9192 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9193 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9194 }
9195 return false;
9196 }
9197
9198 //------------------------------inline_digestBase_implCompressMB-----------------------
9199 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9200 BasicType elem_type, address stubAddr, const char *stubName,
9201 Node* src_start, Node* ofs, Node* limit) {
9202 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9203 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9204 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9205 digest_obj = _gvn.transform(digest_obj);
9206
9207 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9208 if (state == nullptr) return false;
9209
9210 Node* block_size = nullptr;
9211 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9212 block_size = get_block_size_from_digest_object(digest_obj);
9213 if (block_size == nullptr) return false;
9214 }
9215
9216 // Call the stub.
9217 Node* call;
9218 if (block_size == nullptr) {
9219 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9220 OptoRuntime::digestBase_implCompressMB_Type(false),
9221 stubAddr, stubName, TypePtr::BOTTOM,
9222 src_start, state, ofs, limit);
9223 } else {
9224 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9225 OptoRuntime::digestBase_implCompressMB_Type(true),
9226 stubAddr, stubName, TypePtr::BOTTOM,
9227 src_start, state, block_size, ofs, limit);
9228 }
9229
9230 // return ofs (int)
9231 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9232 set_result(result);
9233
9234 return true;
9235 }
9236
9237 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9238 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9239 assert(UseAES, "need AES instruction support");
9240 address stubAddr = nullptr;
9241 const char *stubName = nullptr;
9242 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9243 stubName = "galoisCounterMode_AESCrypt";
9244
9245 if (stubAddr == nullptr) return false;
9246
9247 Node* in = argument(0);
9248 Node* inOfs = argument(1);
9249 Node* len = argument(2);
9250 Node* ct = argument(3);
9251 Node* ctOfs = argument(4);
9252 Node* out = argument(5);
9253 Node* outOfs = argument(6);
9254 Node* gctr_object = argument(7);
9255 Node* ghash_object = argument(8);
9256
9257 // (1) in, ct and out are arrays.
9258 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9259 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9260 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9261 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9262 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9263 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9264
9265 // checks are the responsibility of the caller
9266 Node* in_start = in;
9267 Node* ct_start = ct;
9268 Node* out_start = out;
9269 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9270 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9271 in_start = array_element_address(in, inOfs, T_BYTE);
9272 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9273 out_start = array_element_address(out, outOfs, T_BYTE);
9274 }
9275
9276 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9277 // (because of the predicated logic executed earlier).
9278 // so we cast it here safely.
9279 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9280 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9281 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9282 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9283 Node* state = load_field_from_object(ghash_object, "state", "[J");
9284
9285 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9286 return false;
9287 }
9288 // cast it to what we know it will be at runtime
9289 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9290 assert(tinst != nullptr, "GCTR obj is null");
9291 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9292 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9293 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9294 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9295 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9296 const TypeOopPtr* xtype = aklass->as_instance_type();
9297 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9298 aescrypt_object = _gvn.transform(aescrypt_object);
9299 // we need to get the start of the aescrypt_object's expanded key array
9300 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9301 if (k_start == nullptr) return false;
9302 // similarly, get the start address of the r vector
9303 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9304 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9305 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9306
9307
9308 // Call the stub, passing params
9309 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9310 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9311 stubAddr, stubName, TypePtr::BOTTOM,
9312 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9313
9314 // return cipher length (int)
9315 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9316 set_result(retvalue);
9317
9318 return true;
9319 }
9320
9321 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9322 // Return node representing slow path of predicate check.
9323 // the pseudo code we want to emulate with this predicate is:
9324 // for encryption:
9325 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9326 // for decryption:
9327 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9328 // note cipher==plain is more conservative than the original java code but that's OK
9329 //
9330
9331 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9332 // The receiver was checked for null already.
9333 Node* objGCTR = argument(7);
9334 // Load embeddedCipher field of GCTR object.
9335 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9336 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9337
9338 // get AESCrypt klass for instanceOf check
9339 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9340 // will have same classloader as CipherBlockChaining object
9341 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9342 assert(tinst != nullptr, "GCTR obj is null");
9343 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9344
9345 // we want to do an instanceof comparison against the AESCrypt class
9346 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
9347 if (!klass_AESCrypt->is_loaded()) {
9348 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9349 Node* ctrl = control();
9350 set_control(top()); // no regular fast path
9351 return ctrl;
9352 }
9353
9354 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9355 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9356 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9357 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9358 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9359
9360 return instof_false; // even if it is null
9361 }
9362
9363 //------------------------------get_state_from_digest_object-----------------------
9364 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9365 const char* state_type;
9366 switch (elem_type) {
9367 case T_BYTE: state_type = "[B"; break;
9368 case T_INT: state_type = "[I"; break;
9369 case T_LONG: state_type = "[J"; break;
9370 default: ShouldNotReachHere();
9371 }
9372 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9373 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9374 if (digest_state == nullptr) return (Node *) nullptr;
9375
9376 // now have the array, need to get the start address of the state array
9377 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9378 return state;
9379 }
9380
9381 //------------------------------get_block_size_from_sha3_object----------------------------------
9382 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9383 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9384 assert (block_size != nullptr, "sanity");
9385 return block_size;
9386 }
9387
9388 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9389 // Return node representing slow path of predicate check.
9390 // the pseudo code we want to emulate with this predicate is:
9391 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9392 //
9393 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9394 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9395 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9396 assert((uint)predicate < 5, "sanity");
9397
9398 // The receiver was checked for null already.
9399 Node* digestBaseObj = argument(0);
9400
9401 // get DigestBase klass for instanceOf check
9402 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9403 assert(tinst != nullptr, "digestBaseObj is null");
9404 assert(tinst->is_loaded(), "DigestBase is not loaded");
9405
9406 const char* klass_name = nullptr;
9407 switch (predicate) {
9408 case 0:
9409 if (UseMD5Intrinsics) {
9410 // we want to do an instanceof comparison against the MD5 class
9411 klass_name = "sun/security/provider/MD5";
9412 }
9413 break;
9414 case 1:
9415 if (UseSHA1Intrinsics) {
9416 // we want to do an instanceof comparison against the SHA class
9417 klass_name = "sun/security/provider/SHA";
9418 }
9419 break;
9420 case 2:
9421 if (UseSHA256Intrinsics) {
9422 // we want to do an instanceof comparison against the SHA2 class
9423 klass_name = "sun/security/provider/SHA2";
9424 }
9425 break;
9426 case 3:
9427 if (UseSHA512Intrinsics) {
9428 // we want to do an instanceof comparison against the SHA5 class
9429 klass_name = "sun/security/provider/SHA5";
9430 }
9431 break;
9432 case 4:
9433 if (UseSHA3Intrinsics) {
9434 // we want to do an instanceof comparison against the SHA3 class
9435 klass_name = "sun/security/provider/SHA3";
9436 }
9437 break;
9438 default:
9439 fatal("unknown SHA intrinsic predicate: %d", predicate);
9440 }
9441
9442 ciKlass* klass = nullptr;
9443 if (klass_name != nullptr) {
9444 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9445 }
9446 if ((klass == nullptr) || !klass->is_loaded()) {
9447 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9448 Node* ctrl = control();
9449 set_control(top()); // no intrinsic path
9450 return ctrl;
9451 }
9452 ciInstanceKlass* instklass = klass->as_instance_klass();
9453
9454 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9455 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9456 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9457 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9458
9459 return instof_false; // even if it is null
9460 }
9461
9462 //-------------inline_fma-----------------------------------
9463 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9464 Node *a = nullptr;
9465 Node *b = nullptr;
9466 Node *c = nullptr;
9467 Node* result = nullptr;
9468 switch (id) {
9469 case vmIntrinsics::_fmaD:
9470 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9471 // no receiver since it is static method
9472 a = argument(0);
9473 b = argument(2);
9474 c = argument(4);
9475 result = _gvn.transform(new FmaDNode(a, b, c));
9476 break;
9477 case vmIntrinsics::_fmaF:
9478 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9479 a = argument(0);
9480 b = argument(1);
9481 c = argument(2);
9482 result = _gvn.transform(new FmaFNode(a, b, c));
9483 break;
9484 default:
9485 fatal_unexpected_iid(id); break;
9486 }
9487 set_result(result);
9488 return true;
9489 }
9490
9491 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9492 // argument(0) is receiver
9493 Node* codePoint = argument(1);
9494 Node* n = nullptr;
9495
9496 switch (id) {
9497 case vmIntrinsics::_isDigit :
9498 n = new DigitNode(control(), codePoint);
9499 break;
9500 case vmIntrinsics::_isLowerCase :
9501 n = new LowerCaseNode(control(), codePoint);
9502 break;
9503 case vmIntrinsics::_isUpperCase :
9504 n = new UpperCaseNode(control(), codePoint);
9505 break;
9506 case vmIntrinsics::_isWhitespace :
9507 n = new WhitespaceNode(control(), codePoint);
9508 break;
9509 default:
9510 fatal_unexpected_iid(id);
9511 }
9512
9513 set_result(_gvn.transform(n));
9514 return true;
9515 }
9516
9517 bool LibraryCallKit::inline_profileBoolean() {
9518 Node* counts = argument(1);
9519 const TypeAryPtr* ary = nullptr;
9520 ciArray* aobj = nullptr;
9521 if (counts->is_Con()
9522 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9523 && (aobj = ary->const_oop()->as_array()) != nullptr
9524 && (aobj->length() == 2)) {
9525 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9526 jint false_cnt = aobj->element_value(0).as_int();
9527 jint true_cnt = aobj->element_value(1).as_int();
9528
9529 if (C->log() != nullptr) {
9530 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9531 false_cnt, true_cnt);
9532 }
9533
9534 if (false_cnt + true_cnt == 0) {
9535 // According to profile, never executed.
9536 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9537 Deoptimization::Action_reinterpret);
9538 return true;
9539 }
9540
9541 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9542 // is a number of each value occurrences.
9543 Node* result = argument(0);
9544 if (false_cnt == 0 || true_cnt == 0) {
9545 // According to profile, one value has been never seen.
9546 int expected_val = (false_cnt == 0) ? 1 : 0;
9547
9548 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9549 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9550
9551 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9552 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9553 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9554
9555 { // Slow path: uncommon trap for never seen value and then reexecute
9556 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9557 // the value has been seen at least once.
9558 PreserveJVMState pjvms(this);
9559 PreserveReexecuteState preexecs(this);
9560 jvms()->set_should_reexecute(true);
9561
9562 set_control(slow_path);
9563 set_i_o(i_o());
9564
9565 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9566 Deoptimization::Action_reinterpret);
9567 }
9568 // The guard for never seen value enables sharpening of the result and
9569 // returning a constant. It allows to eliminate branches on the same value
9570 // later on.
9571 set_control(fast_path);
9572 result = intcon(expected_val);
9573 }
9574 // Stop profiling.
9575 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9576 // By replacing method body with profile data (represented as ProfileBooleanNode
9577 // on IR level) we effectively disable profiling.
9578 // It enables full speed execution once optimized code is generated.
9579 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9580 C->record_for_igvn(profile);
9581 set_result(profile);
9582 return true;
9583 } else {
9584 // Continue profiling.
9585 // Profile data isn't available at the moment. So, execute method's bytecode version.
9586 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9587 // is compiled and counters aren't available since corresponding MethodHandle
9588 // isn't a compile-time constant.
9589 return false;
9590 }
9591 }
9592
9593 bool LibraryCallKit::inline_isCompileConstant() {
9594 Node* n = argument(0);
9595 set_result(n->is_Con() ? intcon(1) : intcon(0));
9596 return true;
9597 }
9598
9599 //------------------------------- inline_getObjectSize --------------------------------------
9600 //
9601 // Calculate the runtime size of the object/array.
9602 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9603 //
9604 bool LibraryCallKit::inline_getObjectSize() {
9605 Node* obj = argument(3);
9606 Node* klass_node = load_object_klass(obj);
9607
9608 jint layout_con = Klass::_lh_neutral_value;
9609 Node* layout_val = get_layout_helper(klass_node, layout_con);
9610 int layout_is_con = (layout_val == nullptr);
9611
9612 if (layout_is_con) {
9613 // Layout helper is constant, can figure out things at compile time.
9614
9615 if (Klass::layout_helper_is_instance(layout_con)) {
9616 // Instance case: layout_con contains the size itself.
9617 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9618 set_result(size);
9619 } else {
9620 // Array case: size is round(header + element_size*arraylength).
9621 // Since arraylength is different for every array instance, we have to
9622 // compute the whole thing at runtime.
9623
9624 Node* arr_length = load_array_length(obj);
9625
9626 int round_mask = MinObjAlignmentInBytes - 1;
9627 int hsize = Klass::layout_helper_header_size(layout_con);
9628 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9629
9630 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9631 round_mask = 0; // strength-reduce it if it goes away completely
9632 }
9633 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9634 Node* header_size = intcon(hsize + round_mask);
9635
9636 Node* lengthx = ConvI2X(arr_length);
9637 Node* headerx = ConvI2X(header_size);
9638
9639 Node* abody = lengthx;
9640 if (eshift != 0) {
9641 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9642 }
9643 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9644 if (round_mask != 0) {
9645 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9646 }
9647 size = ConvX2L(size);
9648 set_result(size);
9649 }
9650 } else {
9651 // Layout helper is not constant, need to test for array-ness at runtime.
9652
9653 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9654 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9655 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9656 record_for_igvn(result_reg);
9657
9658 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9659 if (array_ctl != nullptr) {
9660 // Array case: size is round(header + element_size*arraylength).
9661 // Since arraylength is different for every array instance, we have to
9662 // compute the whole thing at runtime.
9663
9664 PreserveJVMState pjvms(this);
9665 set_control(array_ctl);
9666 Node* arr_length = load_array_length(obj);
9667
9668 int round_mask = MinObjAlignmentInBytes - 1;
9669 Node* mask = intcon(round_mask);
9670
9671 Node* hss = intcon(Klass::_lh_header_size_shift);
9672 Node* hsm = intcon(Klass::_lh_header_size_mask);
9673 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9674 header_size = _gvn.transform(new AndINode(header_size, hsm));
9675 header_size = _gvn.transform(new AddINode(header_size, mask));
9676
9677 // There is no need to mask or shift this value.
9678 // The semantics of LShiftINode include an implicit mask to 0x1F.
9679 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9680 Node* elem_shift = layout_val;
9681
9682 Node* lengthx = ConvI2X(arr_length);
9683 Node* headerx = ConvI2X(header_size);
9684
9685 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9686 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9687 if (round_mask != 0) {
9688 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9689 }
9690 size = ConvX2L(size);
9691
9692 result_reg->init_req(_array_path, control());
9693 result_val->init_req(_array_path, size);
9694 }
9695
9696 if (!stopped()) {
9697 // Instance case: the layout helper gives us instance size almost directly,
9698 // but we need to mask out the _lh_instance_slow_path_bit.
9699 Node* size = ConvI2X(layout_val);
9700 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9701 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9702 size = _gvn.transform(new AndXNode(size, mask));
9703 size = ConvX2L(size);
9704
9705 result_reg->init_req(_instance_path, control());
9706 result_val->init_req(_instance_path, size);
9707 }
9708
9709 set_result(result_reg, result_val);
9710 }
9711
9712 return true;
9713 }
9714
9715 //------------------------------- inline_blackhole --------------------------------------
9716 //
9717 // Make sure all arguments to this node are alive.
9718 // This matches methods that were requested to be blackholed through compile commands.
9719 //
9720 bool LibraryCallKit::inline_blackhole() {
9721 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9722 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9723 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9724
9725 // Blackhole node pinches only the control, not memory. This allows
9726 // the blackhole to be pinned in the loop that computes blackholed
9727 // values, but have no other side effects, like breaking the optimizations
9728 // across the blackhole.
9729
9730 Node* bh = _gvn.transform(new BlackholeNode(control()));
9731 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9732
9733 // Bind call arguments as blackhole arguments to keep them alive
9734 uint nargs = callee()->arg_size();
9735 for (uint i = 0; i < nargs; i++) {
9736 bh->add_req(argument(i));
9737 }
9738
9739 return true;
9740 }
9741
9742 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9743 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9744 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9745 return nullptr; // box klass is not Float16
9746 }
9747
9748 // Null check; get notnull casted pointer
9749 Node* null_ctl = top();
9750 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9751 // If not_null_box is dead, only null-path is taken
9752 if (stopped()) {
9753 set_control(null_ctl);
9754 return nullptr;
9755 }
9756 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9757 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9758 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9759 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9760 }
9761
9762 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9763 PreserveReexecuteState preexecs(this);
9764 jvms()->set_should_reexecute(true);
9765
9766 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9767 Node* klass_node = makecon(klass_type);
9768 Node* box = new_instance(klass_node);
9769
9770 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9771 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9772
9773 Node* field_store = _gvn.transform(access_store_at(box,
9774 value_field,
9775 value_adr_type,
9776 value,
9777 TypeInt::SHORT,
9778 T_SHORT,
9779 IN_HEAP));
9780 set_memory(field_store, value_adr_type);
9781 return box;
9782 }
9783
9784 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9785 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9786 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9787 return false;
9788 }
9789
9790 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9791 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9792 return false;
9793 }
9794
9795 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9796 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9797 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9798 ciSymbols::short_signature(),
9799 false);
9800 assert(field != nullptr, "");
9801
9802 // Transformed nodes
9803 Node* fld1 = nullptr;
9804 Node* fld2 = nullptr;
9805 Node* fld3 = nullptr;
9806 switch(num_args) {
9807 case 3:
9808 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9809 if (fld3 == nullptr) {
9810 return false;
9811 }
9812 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9813 // fall-through
9814 case 2:
9815 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9816 if (fld2 == nullptr) {
9817 return false;
9818 }
9819 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9820 // fall-through
9821 case 1:
9822 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9823 if (fld1 == nullptr) {
9824 return false;
9825 }
9826 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9827 break;
9828 default: fatal("Unsupported number of arguments %d", num_args);
9829 }
9830
9831 Node* result = nullptr;
9832 switch (id) {
9833 // Unary operations
9834 case vmIntrinsics::_sqrt_float16:
9835 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9836 break;
9837 // Ternary operations
9838 case vmIntrinsics::_fma_float16:
9839 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9840 break;
9841 default:
9842 fatal_unexpected_iid(id);
9843 break;
9844 }
9845 result = _gvn.transform(new ReinterpretHF2SNode(result));
9846 set_result(box_fp16_value(float16_box_type, field, result));
9847 return true;
9848 }
9849