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::_arrayInstanceBaseOffset: return inline_arrayInstanceBaseOffset();
489 case vmIntrinsics::_arrayInstanceIndexScale: return inline_arrayInstanceIndexScale();
490 case vmIntrinsics::_arrayLayout: return inline_arrayLayout();
491
492 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
493
494 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
495 case vmIntrinsics::_currentThread: return inline_native_currentThread();
496 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
497
498 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
499 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
500
501 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
502 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
503
504 #if INCLUDE_JVMTI
505 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
506 "notifyJvmtiStart", true, false);
507 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
508 "notifyJvmtiEnd", false, true);
509 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
510 "notifyJvmtiMount", false, false);
511 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
512 "notifyJvmtiUnmount", false, false);
513 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
514 #endif
515
516 #ifdef JFR_HAVE_INTRINSICS
517 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
518 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
519 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
520 #endif
521 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
522 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
523 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
524 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
525 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
526 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
527 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
528 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
529 case vmIntrinsics::_getLength: return inline_native_getLength();
530 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
531 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
532 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
533 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
534 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
535 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
536 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
537
538 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
539 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
540 case vmIntrinsics::_newNullRestrictedNonAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ false);
541 case vmIntrinsics::_newNullRestrictedAtomicArray: return inline_newArray(/* null_free */ true, /* atomic */ true);
542 case vmIntrinsics::_newNullableAtomicArray: return inline_newArray(/* null_free */ false, /* atomic */ true);
543 case vmIntrinsics::_isFlatArray: return inline_getArrayProperties(IsFlat);
544 case vmIntrinsics::_isNullRestrictedArray: return inline_getArrayProperties(IsNullRestricted);
545 case vmIntrinsics::_isAtomicArray: return inline_getArrayProperties(IsAtomic);
546
547 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
548
549 case vmIntrinsics::_isInstance:
550 case vmIntrinsics::_isHidden:
551 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
552
553 case vmIntrinsics::_floatToRawIntBits:
554 case vmIntrinsics::_floatToIntBits:
555 case vmIntrinsics::_intBitsToFloat:
556 case vmIntrinsics::_doubleToRawLongBits:
557 case vmIntrinsics::_doubleToLongBits:
558 case vmIntrinsics::_longBitsToDouble:
559 case vmIntrinsics::_floatToFloat16:
560 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
561 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
562 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
563 case vmIntrinsics::_floatIsFinite:
564 case vmIntrinsics::_floatIsInfinite:
565 case vmIntrinsics::_doubleIsFinite:
566 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
567
568 case vmIntrinsics::_numberOfLeadingZeros_i:
569 case vmIntrinsics::_numberOfLeadingZeros_l:
570 case vmIntrinsics::_numberOfTrailingZeros_i:
571 case vmIntrinsics::_numberOfTrailingZeros_l:
572 case vmIntrinsics::_bitCount_i:
573 case vmIntrinsics::_bitCount_l:
574 case vmIntrinsics::_reverse_i:
575 case vmIntrinsics::_reverse_l:
576 case vmIntrinsics::_reverseBytes_i:
577 case vmIntrinsics::_reverseBytes_l:
578 case vmIntrinsics::_reverseBytes_s:
579 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
580
581 case vmIntrinsics::_compress_i:
582 case vmIntrinsics::_compress_l:
583 case vmIntrinsics::_expand_i:
584 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
585
586 case vmIntrinsics::_compareUnsigned_i:
587 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
588
589 case vmIntrinsics::_divideUnsigned_i:
590 case vmIntrinsics::_divideUnsigned_l:
591 case vmIntrinsics::_remainderUnsigned_i:
592 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
593
594 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
595
596 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
597 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
598 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
599 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
600 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
601
602 case vmIntrinsics::_Class_cast: return inline_Class_cast();
603
604 case vmIntrinsics::_aescrypt_encryptBlock:
605 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
606
607 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
608 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
609 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
610
611 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
612 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
613 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
614
615 case vmIntrinsics::_counterMode_AESCrypt:
616 return inline_counterMode_AESCrypt(intrinsic_id());
617
618 case vmIntrinsics::_galoisCounterMode_AESCrypt:
619 return inline_galoisCounterMode_AESCrypt();
620
621 case vmIntrinsics::_md5_implCompress:
622 case vmIntrinsics::_sha_implCompress:
623 case vmIntrinsics::_sha2_implCompress:
624 case vmIntrinsics::_sha5_implCompress:
625 case vmIntrinsics::_sha3_implCompress:
626 return inline_digestBase_implCompress(intrinsic_id());
627 case vmIntrinsics::_double_keccak:
628 return inline_double_keccak();
629
630 case vmIntrinsics::_digestBase_implCompressMB:
631 return inline_digestBase_implCompressMB(predicate);
632
633 case vmIntrinsics::_multiplyToLen:
634 return inline_multiplyToLen();
635
636 case vmIntrinsics::_squareToLen:
637 return inline_squareToLen();
638
639 case vmIntrinsics::_mulAdd:
640 return inline_mulAdd();
641
642 case vmIntrinsics::_montgomeryMultiply:
643 return inline_montgomeryMultiply();
644 case vmIntrinsics::_montgomerySquare:
645 return inline_montgomerySquare();
646
647 case vmIntrinsics::_bigIntegerRightShiftWorker:
648 return inline_bigIntegerShift(true);
649 case vmIntrinsics::_bigIntegerLeftShiftWorker:
650 return inline_bigIntegerShift(false);
651
652 case vmIntrinsics::_vectorizedMismatch:
653 return inline_vectorizedMismatch();
654
655 case vmIntrinsics::_ghash_processBlocks:
656 return inline_ghash_processBlocks();
657 case vmIntrinsics::_chacha20Block:
658 return inline_chacha20Block();
659 case vmIntrinsics::_kyberNtt:
660 return inline_kyberNtt();
661 case vmIntrinsics::_kyberInverseNtt:
662 return inline_kyberInverseNtt();
663 case vmIntrinsics::_kyberNttMult:
664 return inline_kyberNttMult();
665 case vmIntrinsics::_kyberAddPoly_2:
666 return inline_kyberAddPoly_2();
667 case vmIntrinsics::_kyberAddPoly_3:
668 return inline_kyberAddPoly_3();
669 case vmIntrinsics::_kyber12To16:
670 return inline_kyber12To16();
671 case vmIntrinsics::_kyberBarrettReduce:
672 return inline_kyberBarrettReduce();
673 case vmIntrinsics::_dilithiumAlmostNtt:
674 return inline_dilithiumAlmostNtt();
675 case vmIntrinsics::_dilithiumAlmostInverseNtt:
676 return inline_dilithiumAlmostInverseNtt();
677 case vmIntrinsics::_dilithiumNttMult:
678 return inline_dilithiumNttMult();
679 case vmIntrinsics::_dilithiumMontMulByConstant:
680 return inline_dilithiumMontMulByConstant();
681 case vmIntrinsics::_dilithiumDecomposePoly:
682 return inline_dilithiumDecomposePoly();
683 case vmIntrinsics::_base64_encodeBlock:
684 return inline_base64_encodeBlock();
685 case vmIntrinsics::_base64_decodeBlock:
686 return inline_base64_decodeBlock();
687 case vmIntrinsics::_poly1305_processBlocks:
688 return inline_poly1305_processBlocks();
689 case vmIntrinsics::_intpoly_montgomeryMult_P256:
690 return inline_intpoly_montgomeryMult_P256();
691 case vmIntrinsics::_intpoly_assign:
692 return inline_intpoly_assign();
693 case vmIntrinsics::_encodeISOArray:
694 case vmIntrinsics::_encodeByteISOArray:
695 return inline_encodeISOArray(false);
696 case vmIntrinsics::_encodeAsciiArray:
697 return inline_encodeISOArray(true);
698
699 case vmIntrinsics::_updateCRC32:
700 return inline_updateCRC32();
701 case vmIntrinsics::_updateBytesCRC32:
702 return inline_updateBytesCRC32();
703 case vmIntrinsics::_updateByteBufferCRC32:
704 return inline_updateByteBufferCRC32();
705
706 case vmIntrinsics::_updateBytesCRC32C:
707 return inline_updateBytesCRC32C();
708 case vmIntrinsics::_updateDirectByteBufferCRC32C:
709 return inline_updateDirectByteBufferCRC32C();
710
711 case vmIntrinsics::_updateBytesAdler32:
712 return inline_updateBytesAdler32();
713 case vmIntrinsics::_updateByteBufferAdler32:
714 return inline_updateByteBufferAdler32();
715
716 case vmIntrinsics::_profileBoolean:
717 return inline_profileBoolean();
718 case vmIntrinsics::_isCompileConstant:
719 return inline_isCompileConstant();
720
721 case vmIntrinsics::_countPositives:
722 return inline_countPositives();
723
724 case vmIntrinsics::_fmaD:
725 case vmIntrinsics::_fmaF:
726 return inline_fma(intrinsic_id());
727
728 case vmIntrinsics::_isDigit:
729 case vmIntrinsics::_isLowerCase:
730 case vmIntrinsics::_isUpperCase:
731 case vmIntrinsics::_isWhitespace:
732 return inline_character_compare(intrinsic_id());
733
734 case vmIntrinsics::_min:
735 case vmIntrinsics::_max:
736 case vmIntrinsics::_min_strict:
737 case vmIntrinsics::_max_strict:
738 case vmIntrinsics::_minL:
739 case vmIntrinsics::_maxL:
740 case vmIntrinsics::_minF:
741 case vmIntrinsics::_maxF:
742 case vmIntrinsics::_minD:
743 case vmIntrinsics::_maxD:
744 case vmIntrinsics::_minF_strict:
745 case vmIntrinsics::_maxF_strict:
746 case vmIntrinsics::_minD_strict:
747 case vmIntrinsics::_maxD_strict:
748 return inline_min_max(intrinsic_id());
749
750 case vmIntrinsics::_VectorUnaryOp:
751 return inline_vector_nary_operation(1);
752 case vmIntrinsics::_VectorBinaryOp:
753 return inline_vector_nary_operation(2);
754 case vmIntrinsics::_VectorUnaryLibOp:
755 return inline_vector_call(1);
756 case vmIntrinsics::_VectorBinaryLibOp:
757 return inline_vector_call(2);
758 case vmIntrinsics::_VectorTernaryOp:
759 return inline_vector_nary_operation(3);
760 case vmIntrinsics::_VectorFromBitsCoerced:
761 return inline_vector_frombits_coerced();
762 case vmIntrinsics::_VectorMaskOp:
763 return inline_vector_mask_operation();
764 case vmIntrinsics::_VectorLoadOp:
765 return inline_vector_mem_operation(/*is_store=*/false);
766 case vmIntrinsics::_VectorLoadMaskedOp:
767 return inline_vector_mem_masked_operation(/*is_store*/false);
768 case vmIntrinsics::_VectorStoreOp:
769 return inline_vector_mem_operation(/*is_store=*/true);
770 case vmIntrinsics::_VectorStoreMaskedOp:
771 return inline_vector_mem_masked_operation(/*is_store=*/true);
772 case vmIntrinsics::_VectorGatherOp:
773 return inline_vector_gather_scatter(/*is_scatter*/ false);
774 case vmIntrinsics::_VectorScatterOp:
775 return inline_vector_gather_scatter(/*is_scatter*/ true);
776 case vmIntrinsics::_VectorReductionCoerced:
777 return inline_vector_reduction();
778 case vmIntrinsics::_VectorTest:
779 return inline_vector_test();
780 case vmIntrinsics::_VectorBlend:
781 return inline_vector_blend();
782 case vmIntrinsics::_VectorRearrange:
783 return inline_vector_rearrange();
784 case vmIntrinsics::_VectorSelectFrom:
785 return inline_vector_select_from();
786 case vmIntrinsics::_VectorCompare:
787 return inline_vector_compare();
788 case vmIntrinsics::_VectorBroadcastInt:
789 return inline_vector_broadcast_int();
790 case vmIntrinsics::_VectorConvert:
791 return inline_vector_convert();
792 case vmIntrinsics::_VectorInsert:
793 return inline_vector_insert();
794 case vmIntrinsics::_VectorExtract:
795 return inline_vector_extract();
796 case vmIntrinsics::_VectorCompressExpand:
797 return inline_vector_compress_expand();
798 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
799 return inline_vector_select_from_two_vectors();
800 case vmIntrinsics::_IndexVector:
801 return inline_index_vector();
802 case vmIntrinsics::_IndexPartiallyInUpperRange:
803 return inline_index_partially_in_upper_range();
804
805 case vmIntrinsics::_getObjectSize:
806 return inline_getObjectSize();
807
808 case vmIntrinsics::_blackhole:
809 return inline_blackhole();
810
811 default:
812 // If you get here, it may be that someone has added a new intrinsic
813 // to the list in vmIntrinsics.hpp without implementing it here.
814 #ifndef PRODUCT
815 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
816 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
817 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
818 }
819 #endif
820 return false;
821 }
822 }
823
824 Node* LibraryCallKit::try_to_predicate(int predicate) {
825 if (!jvms()->has_method()) {
826 // Root JVMState has a null method.
827 assert(map()->memory()->Opcode() == Op_Parm, "");
828 // Insert the memory aliasing node
829 set_all_memory(reset_memory());
830 }
831 assert(merged_memory(), "");
832
833 switch (intrinsic_id()) {
834 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
835 return inline_cipherBlockChaining_AESCrypt_predicate(false);
836 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
837 return inline_cipherBlockChaining_AESCrypt_predicate(true);
838 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
839 return inline_electronicCodeBook_AESCrypt_predicate(false);
840 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
841 return inline_electronicCodeBook_AESCrypt_predicate(true);
842 case vmIntrinsics::_counterMode_AESCrypt:
843 return inline_counterMode_AESCrypt_predicate();
844 case vmIntrinsics::_digestBase_implCompressMB:
845 return inline_digestBase_implCompressMB_predicate(predicate);
846 case vmIntrinsics::_galoisCounterMode_AESCrypt:
847 return inline_galoisCounterMode_AESCrypt_predicate();
848
849 default:
850 // If you get here, it may be that someone has added a new intrinsic
851 // to the list in vmIntrinsics.hpp without implementing it here.
852 #ifndef PRODUCT
853 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
854 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
855 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
856 }
857 #endif
858 Node* slow_ctl = control();
859 set_control(top()); // No fast path intrinsic
860 return slow_ctl;
861 }
862 }
863
864 //------------------------------set_result-------------------------------
865 // Helper function for finishing intrinsics.
866 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
867 record_for_igvn(region);
868 set_control(_gvn.transform(region));
869 set_result( _gvn.transform(value));
870 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
871 }
872
873 //------------------------------generate_guard---------------------------
874 // Helper function for generating guarded fast-slow graph structures.
875 // The given 'test', if true, guards a slow path. If the test fails
876 // then a fast path can be taken. (We generally hope it fails.)
877 // In all cases, GraphKit::control() is updated to the fast path.
878 // The returned value represents the control for the slow path.
879 // The return value is never 'top'; it is either a valid control
880 // or null if it is obvious that the slow path can never be taken.
881 // Also, if region and the slow control are not null, the slow edge
882 // is appended to the region.
883 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
884 if (stopped()) {
885 // Already short circuited.
886 return nullptr;
887 }
888
889 // Build an if node and its projections.
890 // If test is true we take the slow path, which we assume is uncommon.
891 if (_gvn.type(test) == TypeInt::ZERO) {
892 // The slow branch is never taken. No need to build this guard.
893 return nullptr;
894 }
895
896 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
897
898 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
899 if (if_slow == top()) {
900 // The slow branch is never taken. No need to build this guard.
901 return nullptr;
902 }
903
904 if (region != nullptr)
905 region->add_req(if_slow);
906
907 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
908 set_control(if_fast);
909
910 return if_slow;
911 }
912
913 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
914 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
915 }
916 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
917 return generate_guard(test, region, PROB_FAIR);
918 }
919
920 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
921 Node* *pos_index) {
922 if (stopped())
923 return nullptr; // already stopped
924 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
925 return nullptr; // index is already adequately typed
926 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
927 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
928 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
929 if (is_neg != nullptr && pos_index != nullptr) {
930 // Emulate effect of Parse::adjust_map_after_if.
931 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
932 (*pos_index) = _gvn.transform(ccast);
933 }
934 return is_neg;
935 }
936
937 // Make sure that 'position' is a valid limit index, in [0..length].
938 // There are two equivalent plans for checking this:
939 // A. (offset + copyLength) unsigned<= arrayLength
940 // B. offset <= (arrayLength - copyLength)
941 // We require that all of the values above, except for the sum and
942 // difference, are already known to be non-negative.
943 // Plan A is robust in the face of overflow, if offset and copyLength
944 // are both hugely positive.
945 //
946 // Plan B is less direct and intuitive, but it does not overflow at
947 // all, since the difference of two non-negatives is always
948 // representable. Whenever Java methods must perform the equivalent
949 // check they generally use Plan B instead of Plan A.
950 // For the moment we use Plan A.
951 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
952 Node* subseq_length,
953 Node* array_length,
954 RegionNode* region) {
955 if (stopped())
956 return nullptr; // already stopped
957 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
958 if (zero_offset && subseq_length->eqv_uncast(array_length))
959 return nullptr; // common case of whole-array copy
960 Node* last = subseq_length;
961 if (!zero_offset) // last += offset
962 last = _gvn.transform(new AddINode(last, offset));
963 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
964 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
965 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
966 return is_over;
967 }
968
969 // Emit range checks for the given String.value byte array
970 void LibraryCallKit::generate_string_range_check(Node* array,
971 Node* offset,
972 Node* count,
973 bool char_count,
974 bool halt_on_oob) {
975 if (stopped()) {
976 return; // already stopped
977 }
978 RegionNode* bailout = new RegionNode(1);
979 record_for_igvn(bailout);
980 if (char_count) {
981 // Convert char count to byte count
982 count = _gvn.transform(new LShiftINode(count, intcon(1)));
983 }
984
985 // Offset and count must not be negative
986 generate_negative_guard(offset, bailout);
987 generate_negative_guard(count, bailout);
988 // Offset + count must not exceed length of array
989 generate_limit_guard(offset, count, load_array_length(array), bailout);
990
991 if (bailout->req() > 1) {
992 if (halt_on_oob) {
993 bailout = _gvn.transform(bailout)->as_Region();
994 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
995 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
996 C->root()->add_req(halt);
997 } else {
998 PreserveJVMState pjvms(this);
999 set_control(_gvn.transform(bailout));
1000 uncommon_trap(Deoptimization::Reason_intrinsic,
1001 Deoptimization::Action_maybe_recompile);
1002 }
1003 }
1004 }
1005
1006 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
1007 bool is_immutable) {
1008 ciKlass* thread_klass = env()->Thread_klass();
1009 const Type* thread_type
1010 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
1011
1012 Node* thread = _gvn.transform(new ThreadLocalNode());
1013 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
1014 tls_output = thread;
1015
1016 Node* thread_obj_handle
1017 = (is_immutable
1018 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
1019 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
1020 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1021 thread_obj_handle = _gvn.transform(thread_obj_handle);
1022
1023 DecoratorSet decorators = IN_NATIVE;
1024 if (is_immutable) {
1025 decorators |= C2_IMMUTABLE_MEMORY;
1026 }
1027 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1028 }
1029
1030 //--------------------------generate_current_thread--------------------
1031 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1032 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1033 /*is_immutable*/false);
1034 }
1035
1036 //--------------------------generate_virtual_thread--------------------
1037 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1038 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1039 !C->method()->changes_current_thread());
1040 }
1041
1042 //------------------------------make_string_method_node------------------------
1043 // Helper method for String intrinsic functions. This version is called with
1044 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1045 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1046 // containing the lengths of str1 and str2.
1047 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1048 Node* result = nullptr;
1049 switch (opcode) {
1050 case Op_StrIndexOf:
1051 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1052 str1_start, cnt1, str2_start, cnt2, ae);
1053 break;
1054 case Op_StrComp:
1055 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1056 str1_start, cnt1, str2_start, cnt2, ae);
1057 break;
1058 case Op_StrEquals:
1059 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1060 // Use the constant length if there is one because optimized match rule may exist.
1061 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1062 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1063 break;
1064 default:
1065 ShouldNotReachHere();
1066 return nullptr;
1067 }
1068
1069 // All these intrinsics have checks.
1070 C->set_has_split_ifs(true); // Has chance for split-if optimization
1071 clear_upper_avx();
1072
1073 return _gvn.transform(result);
1074 }
1075
1076 //------------------------------inline_string_compareTo------------------------
1077 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1078 Node* arg1 = argument(0);
1079 Node* arg2 = argument(1);
1080
1081 arg1 = must_be_not_null(arg1, true);
1082 arg2 = must_be_not_null(arg2, true);
1083
1084 // Get start addr and length of first argument
1085 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1086 Node* arg1_cnt = load_array_length(arg1);
1087
1088 // Get start addr and length of second argument
1089 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1090 Node* arg2_cnt = load_array_length(arg2);
1091
1092 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1093 set_result(result);
1094 return true;
1095 }
1096
1097 //------------------------------inline_string_equals------------------------
1098 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1099 Node* arg1 = argument(0);
1100 Node* arg2 = argument(1);
1101
1102 // paths (plus control) merge
1103 RegionNode* region = new RegionNode(3);
1104 Node* phi = new PhiNode(region, TypeInt::BOOL);
1105
1106 if (!stopped()) {
1107
1108 arg1 = must_be_not_null(arg1, true);
1109 arg2 = must_be_not_null(arg2, true);
1110
1111 // Get start addr and length of first argument
1112 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1113 Node* arg1_cnt = load_array_length(arg1);
1114
1115 // Get start addr and length of second argument
1116 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1117 Node* arg2_cnt = load_array_length(arg2);
1118
1119 // Check for arg1_cnt != arg2_cnt
1120 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1121 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1122 Node* if_ne = generate_slow_guard(bol, nullptr);
1123 if (if_ne != nullptr) {
1124 phi->init_req(2, intcon(0));
1125 region->init_req(2, if_ne);
1126 }
1127
1128 // Check for count == 0 is done by assembler code for StrEquals.
1129
1130 if (!stopped()) {
1131 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1132 phi->init_req(1, equals);
1133 region->init_req(1, control());
1134 }
1135 }
1136
1137 // post merge
1138 set_control(_gvn.transform(region));
1139 record_for_igvn(region);
1140
1141 set_result(_gvn.transform(phi));
1142 return true;
1143 }
1144
1145 //------------------------------inline_array_equals----------------------------
1146 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1147 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1148 Node* arg1 = argument(0);
1149 Node* arg2 = argument(1);
1150
1151 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1152 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1153 clear_upper_avx();
1154
1155 return true;
1156 }
1157
1158
1159 //------------------------------inline_countPositives------------------------------
1160 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1161 bool LibraryCallKit::inline_countPositives() {
1162 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1163 return false;
1164 }
1165
1166 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1167 // no receiver since it is static method
1168 Node* ba = argument(0);
1169 Node* offset = argument(1);
1170 Node* len = argument(2);
1171
1172 if (VerifyIntrinsicChecks) {
1173 ba = must_be_not_null(ba, true);
1174 generate_string_range_check(ba, offset, len, false, true);
1175 if (stopped()) {
1176 return true;
1177 }
1178 }
1179
1180 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1181 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1182 set_result(_gvn.transform(result));
1183 clear_upper_avx();
1184 return true;
1185 }
1186
1187 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1188 Node* index = argument(0);
1189 Node* length = bt == T_INT ? argument(1) : argument(2);
1190 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1191 return false;
1192 }
1193
1194 // check that length is positive
1195 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1196 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1197
1198 {
1199 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1200 uncommon_trap(Deoptimization::Reason_intrinsic,
1201 Deoptimization::Action_make_not_entrant);
1202 }
1203
1204 if (stopped()) {
1205 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1206 return true;
1207 }
1208
1209 // length is now known positive, add a cast node to make this explicit
1210 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1211 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1212 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1213 ConstraintCastNode::RegularDependency, bt);
1214 casted_length = _gvn.transform(casted_length);
1215 replace_in_map(length, casted_length);
1216 length = casted_length;
1217
1218 // Use an unsigned comparison for the range check itself
1219 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1220 BoolTest::mask btest = BoolTest::lt;
1221 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1222 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1223 _gvn.set_type(rc, rc->Value(&_gvn));
1224 if (!rc_bool->is_Con()) {
1225 record_for_igvn(rc);
1226 }
1227 set_control(_gvn.transform(new IfTrueNode(rc)));
1228 {
1229 PreserveJVMState pjvms(this);
1230 set_control(_gvn.transform(new IfFalseNode(rc)));
1231 uncommon_trap(Deoptimization::Reason_range_check,
1232 Deoptimization::Action_make_not_entrant);
1233 }
1234
1235 if (stopped()) {
1236 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1237 return true;
1238 }
1239
1240 // index is now known to be >= 0 and < length, cast it
1241 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1242 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1243 ConstraintCastNode::RegularDependency, bt);
1244 result = _gvn.transform(result);
1245 set_result(result);
1246 replace_in_map(index, result);
1247 return true;
1248 }
1249
1250 //------------------------------inline_string_indexOf------------------------
1251 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1252 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1253 return false;
1254 }
1255 Node* src = argument(0);
1256 Node* tgt = argument(1);
1257
1258 // Make the merge point
1259 RegionNode* result_rgn = new RegionNode(4);
1260 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1261
1262 src = must_be_not_null(src, true);
1263 tgt = must_be_not_null(tgt, true);
1264
1265 // Get start addr and length of source string
1266 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1267 Node* src_count = load_array_length(src);
1268
1269 // Get start addr and length of substring
1270 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1271 Node* tgt_count = load_array_length(tgt);
1272
1273 Node* result = nullptr;
1274 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1275
1276 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1277 // Divide src size by 2 if String is UTF16 encoded
1278 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1279 }
1280 if (ae == StrIntrinsicNode::UU) {
1281 // Divide substring size by 2 if String is UTF16 encoded
1282 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1283 }
1284
1285 if (call_opt_stub) {
1286 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1287 StubRoutines::_string_indexof_array[ae],
1288 "stringIndexOf", TypePtr::BOTTOM, src_start,
1289 src_count, tgt_start, tgt_count);
1290 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1291 } else {
1292 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1293 result_rgn, result_phi, ae);
1294 }
1295 if (result != nullptr) {
1296 result_phi->init_req(3, result);
1297 result_rgn->init_req(3, control());
1298 }
1299 set_control(_gvn.transform(result_rgn));
1300 record_for_igvn(result_rgn);
1301 set_result(_gvn.transform(result_phi));
1302
1303 return true;
1304 }
1305
1306 //-----------------------------inline_string_indexOfI-----------------------
1307 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1308 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1309 return false;
1310 }
1311 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1312 return false;
1313 }
1314
1315 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1316 Node* src = argument(0); // byte[]
1317 Node* src_count = argument(1); // char count
1318 Node* tgt = argument(2); // byte[]
1319 Node* tgt_count = argument(3); // char count
1320 Node* from_index = argument(4); // char index
1321
1322 src = must_be_not_null(src, true);
1323 tgt = must_be_not_null(tgt, true);
1324
1325 // Multiply byte array index by 2 if String is UTF16 encoded
1326 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1327 src_count = _gvn.transform(new SubINode(src_count, from_index));
1328 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1329 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1330
1331 // Range checks
1332 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1333 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1334 if (stopped()) {
1335 return true;
1336 }
1337
1338 RegionNode* region = new RegionNode(5);
1339 Node* phi = new PhiNode(region, TypeInt::INT);
1340 Node* result = nullptr;
1341
1342 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1343
1344 if (call_opt_stub) {
1345 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1346 StubRoutines::_string_indexof_array[ae],
1347 "stringIndexOf", TypePtr::BOTTOM, src_start,
1348 src_count, tgt_start, tgt_count);
1349 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1350 } else {
1351 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1352 region, phi, ae);
1353 }
1354 if (result != nullptr) {
1355 // The result is index relative to from_index if substring was found, -1 otherwise.
1356 // Generate code which will fold into cmove.
1357 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1359
1360 Node* if_lt = generate_slow_guard(bol, nullptr);
1361 if (if_lt != nullptr) {
1362 // result == -1
1363 phi->init_req(3, result);
1364 region->init_req(3, if_lt);
1365 }
1366 if (!stopped()) {
1367 result = _gvn.transform(new AddINode(result, from_index));
1368 phi->init_req(4, result);
1369 region->init_req(4, control());
1370 }
1371 }
1372
1373 set_control(_gvn.transform(region));
1374 record_for_igvn(region);
1375 set_result(_gvn.transform(phi));
1376 clear_upper_avx();
1377
1378 return true;
1379 }
1380
1381 // Create StrIndexOfNode with fast path checks
1382 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1383 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1384 // Check for substr count > string count
1385 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1386 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1387 Node* if_gt = generate_slow_guard(bol, nullptr);
1388 if (if_gt != nullptr) {
1389 phi->init_req(1, intcon(-1));
1390 region->init_req(1, if_gt);
1391 }
1392 if (!stopped()) {
1393 // Check for substr count == 0
1394 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1395 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1396 Node* if_zero = generate_slow_guard(bol, nullptr);
1397 if (if_zero != nullptr) {
1398 phi->init_req(2, intcon(0));
1399 region->init_req(2, if_zero);
1400 }
1401 }
1402 if (!stopped()) {
1403 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1404 }
1405 return nullptr;
1406 }
1407
1408 //-----------------------------inline_string_indexOfChar-----------------------
1409 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1410 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1411 return false;
1412 }
1413 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1414 return false;
1415 }
1416 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1417 Node* src = argument(0); // byte[]
1418 Node* int_ch = argument(1);
1419 Node* from_index = argument(2);
1420 Node* max = argument(3);
1421
1422 src = must_be_not_null(src, true);
1423
1424 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1425 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1426 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1427
1428 // Range checks
1429 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1430
1431 // Check for int_ch >= 0
1432 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1433 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1434 {
1435 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1436 uncommon_trap(Deoptimization::Reason_intrinsic,
1437 Deoptimization::Action_maybe_recompile);
1438 }
1439 if (stopped()) {
1440 return true;
1441 }
1442
1443 RegionNode* region = new RegionNode(3);
1444 Node* phi = new PhiNode(region, TypeInt::INT);
1445
1446 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1447 C->set_has_split_ifs(true); // Has chance for split-if optimization
1448 _gvn.transform(result);
1449
1450 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1451 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1452
1453 Node* if_lt = generate_slow_guard(bol, nullptr);
1454 if (if_lt != nullptr) {
1455 // result == -1
1456 phi->init_req(2, result);
1457 region->init_req(2, if_lt);
1458 }
1459 if (!stopped()) {
1460 result = _gvn.transform(new AddINode(result, from_index));
1461 phi->init_req(1, result);
1462 region->init_req(1, control());
1463 }
1464 set_control(_gvn.transform(region));
1465 record_for_igvn(region);
1466 set_result(_gvn.transform(phi));
1467 clear_upper_avx();
1468
1469 return true;
1470 }
1471 //---------------------------inline_string_copy---------------------
1472 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1473 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1474 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1475 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1476 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1477 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1478 bool LibraryCallKit::inline_string_copy(bool compress) {
1479 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1480 return false;
1481 }
1482 int nargs = 5; // 2 oops, 3 ints
1483 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1484
1485 Node* src = argument(0);
1486 Node* src_offset = argument(1);
1487 Node* dst = argument(2);
1488 Node* dst_offset = argument(3);
1489 Node* length = argument(4);
1490
1491 // Check for allocation before we add nodes that would confuse
1492 // tightly_coupled_allocation()
1493 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1494
1495 // Figure out the size and type of the elements we will be copying.
1496 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1497 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1498 if (src_type == nullptr || dst_type == nullptr) {
1499 return false;
1500 }
1501 BasicType src_elem = src_type->elem()->array_element_basic_type();
1502 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1503 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1504 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1505 "Unsupported array types for inline_string_copy");
1506
1507 src = must_be_not_null(src, true);
1508 dst = must_be_not_null(dst, true);
1509
1510 // Convert char[] offsets to byte[] offsets
1511 bool convert_src = (compress && src_elem == T_BYTE);
1512 bool convert_dst = (!compress && dst_elem == T_BYTE);
1513 if (convert_src) {
1514 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1515 } else if (convert_dst) {
1516 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1517 }
1518
1519 // Range checks
1520 generate_string_range_check(src, src_offset, length, convert_src);
1521 generate_string_range_check(dst, dst_offset, length, convert_dst);
1522 if (stopped()) {
1523 return true;
1524 }
1525
1526 Node* src_start = array_element_address(src, src_offset, src_elem);
1527 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1528 // 'src_start' points to src array + scaled offset
1529 // 'dst_start' points to dst array + scaled offset
1530 Node* count = nullptr;
1531 if (compress) {
1532 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1533 } else {
1534 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1535 }
1536
1537 if (alloc != nullptr) {
1538 if (alloc->maybe_set_complete(&_gvn)) {
1539 // "You break it, you buy it."
1540 InitializeNode* init = alloc->initialization();
1541 assert(init->is_complete(), "we just did this");
1542 init->set_complete_with_arraycopy();
1543 assert(dst->is_CheckCastPP(), "sanity");
1544 assert(dst->in(0)->in(0) == init, "dest pinned");
1545 }
1546 // Do not let stores that initialize this object be reordered with
1547 // a subsequent store that would make this object accessible by
1548 // other threads.
1549 // Record what AllocateNode this StoreStore protects so that
1550 // escape analysis can go from the MemBarStoreStoreNode to the
1551 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1552 // based on the escape status of the AllocateNode.
1553 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1554 }
1555 if (compress) {
1556 set_result(_gvn.transform(count));
1557 }
1558 clear_upper_avx();
1559
1560 return true;
1561 }
1562
1563 #ifdef _LP64
1564 #define XTOP ,top() /*additional argument*/
1565 #else //_LP64
1566 #define XTOP /*no additional argument*/
1567 #endif //_LP64
1568
1569 //------------------------inline_string_toBytesU--------------------------
1570 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1571 bool LibraryCallKit::inline_string_toBytesU() {
1572 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1573 return false;
1574 }
1575 // Get the arguments.
1576 Node* value = argument(0);
1577 Node* offset = argument(1);
1578 Node* length = argument(2);
1579
1580 Node* newcopy = nullptr;
1581
1582 // Set the original stack and the reexecute bit for the interpreter to reexecute
1583 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1584 { PreserveReexecuteState preexecs(this);
1585 jvms()->set_should_reexecute(true);
1586
1587 // Check if a null path was taken unconditionally.
1588 value = null_check(value);
1589
1590 RegionNode* bailout = new RegionNode(1);
1591 record_for_igvn(bailout);
1592
1593 // Range checks
1594 generate_negative_guard(offset, bailout);
1595 generate_negative_guard(length, bailout);
1596 generate_limit_guard(offset, length, load_array_length(value), bailout);
1597 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1598 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1599
1600 if (bailout->req() > 1) {
1601 PreserveJVMState pjvms(this);
1602 set_control(_gvn.transform(bailout));
1603 uncommon_trap(Deoptimization::Reason_intrinsic,
1604 Deoptimization::Action_maybe_recompile);
1605 }
1606 if (stopped()) {
1607 return true;
1608 }
1609
1610 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1611 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1612 newcopy = new_array(klass_node, size, 0); // no arguments to push
1613 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1614 guarantee(alloc != nullptr, "created above");
1615
1616 // Calculate starting addresses.
1617 Node* src_start = array_element_address(value, offset, T_CHAR);
1618 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1619
1620 // Check if dst array address is aligned to HeapWordSize
1621 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1622 // If true, then check if src array address is aligned to HeapWordSize
1623 if (aligned) {
1624 const TypeInt* toffset = gvn().type(offset)->is_int();
1625 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1626 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1627 }
1628
1629 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1630 const char* copyfunc_name = "arraycopy";
1631 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1632 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1633 OptoRuntime::fast_arraycopy_Type(),
1634 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1635 src_start, dst_start, ConvI2X(length) XTOP);
1636 // Do not let reads from the cloned object float above the arraycopy.
1637 if (alloc->maybe_set_complete(&_gvn)) {
1638 // "You break it, you buy it."
1639 InitializeNode* init = alloc->initialization();
1640 assert(init->is_complete(), "we just did this");
1641 init->set_complete_with_arraycopy();
1642 assert(newcopy->is_CheckCastPP(), "sanity");
1643 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1644 }
1645 // Do not let stores that initialize this object be reordered with
1646 // a subsequent store that would make this object accessible by
1647 // other threads.
1648 // Record what AllocateNode this StoreStore protects so that
1649 // escape analysis can go from the MemBarStoreStoreNode to the
1650 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1651 // based on the escape status of the AllocateNode.
1652 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1653 } // original reexecute is set back here
1654
1655 C->set_has_split_ifs(true); // Has chance for split-if optimization
1656 if (!stopped()) {
1657 set_result(newcopy);
1658 }
1659 clear_upper_avx();
1660
1661 return true;
1662 }
1663
1664 //------------------------inline_string_getCharsU--------------------------
1665 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1666 bool LibraryCallKit::inline_string_getCharsU() {
1667 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1668 return false;
1669 }
1670
1671 // Get the arguments.
1672 Node* src = argument(0);
1673 Node* src_begin = argument(1);
1674 Node* src_end = argument(2); // exclusive offset (i < src_end)
1675 Node* dst = argument(3);
1676 Node* dst_begin = argument(4);
1677
1678 // Check for allocation before we add nodes that would confuse
1679 // tightly_coupled_allocation()
1680 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1681
1682 // Check if a null path was taken unconditionally.
1683 src = null_check(src);
1684 dst = null_check(dst);
1685 if (stopped()) {
1686 return true;
1687 }
1688
1689 // Get length and convert char[] offset to byte[] offset
1690 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1691 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1692
1693 // Range checks
1694 generate_string_range_check(src, src_begin, length, true);
1695 generate_string_range_check(dst, dst_begin, length, false);
1696 if (stopped()) {
1697 return true;
1698 }
1699
1700 if (!stopped()) {
1701 // Calculate starting addresses.
1702 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1703 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1704
1705 // Check if array addresses are aligned to HeapWordSize
1706 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1707 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1708 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1709 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1710
1711 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1712 const char* copyfunc_name = "arraycopy";
1713 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1714 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1715 OptoRuntime::fast_arraycopy_Type(),
1716 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1717 src_start, dst_start, ConvI2X(length) XTOP);
1718 // Do not let reads from the cloned object float above the arraycopy.
1719 if (alloc != nullptr) {
1720 if (alloc->maybe_set_complete(&_gvn)) {
1721 // "You break it, you buy it."
1722 InitializeNode* init = alloc->initialization();
1723 assert(init->is_complete(), "we just did this");
1724 init->set_complete_with_arraycopy();
1725 assert(dst->is_CheckCastPP(), "sanity");
1726 assert(dst->in(0)->in(0) == init, "dest pinned");
1727 }
1728 // Do not let stores that initialize this object be reordered with
1729 // a subsequent store that would make this object accessible by
1730 // other threads.
1731 // Record what AllocateNode this StoreStore protects so that
1732 // escape analysis can go from the MemBarStoreStoreNode to the
1733 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1734 // based on the escape status of the AllocateNode.
1735 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1736 } else {
1737 insert_mem_bar(Op_MemBarCPUOrder);
1738 }
1739 }
1740
1741 C->set_has_split_ifs(true); // Has chance for split-if optimization
1742 return true;
1743 }
1744
1745 //----------------------inline_string_char_access----------------------------
1746 // Store/Load char to/from byte[] array.
1747 // static void StringUTF16.putChar(byte[] val, int index, int c)
1748 // static char StringUTF16.getChar(byte[] val, int index)
1749 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1750 Node* value = argument(0);
1751 Node* index = argument(1);
1752 Node* ch = is_store ? argument(2) : nullptr;
1753
1754 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1755 // correctly requires matched array shapes.
1756 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1757 "sanity: byte[] and char[] bases agree");
1758 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1759 "sanity: byte[] and char[] scales agree");
1760
1761 // Bail when getChar over constants is requested: constant folding would
1762 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1763 // Java method would constant fold nicely instead.
1764 if (!is_store && value->is_Con() && index->is_Con()) {
1765 return false;
1766 }
1767
1768 // Save state and restore on bailout
1769 SavedState old_state(this);
1770
1771 value = must_be_not_null(value, true);
1772
1773 Node* adr = array_element_address(value, index, T_CHAR);
1774 if (adr->is_top()) {
1775 return false;
1776 }
1777 old_state.discard();
1778 if (is_store) {
1779 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1780 } else {
1781 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);
1782 set_result(ch);
1783 }
1784 return true;
1785 }
1786
1787
1788 //------------------------------inline_math-----------------------------------
1789 // public static double Math.abs(double)
1790 // public static double Math.sqrt(double)
1791 // public static double Math.log(double)
1792 // public static double Math.log10(double)
1793 // public static double Math.round(double)
1794 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1795 Node* arg = argument(0);
1796 Node* n = nullptr;
1797 switch (id) {
1798 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1799 case vmIntrinsics::_dsqrt:
1800 case vmIntrinsics::_dsqrt_strict:
1801 n = new SqrtDNode(C, control(), arg); break;
1802 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1803 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1804 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1805 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1806 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1807 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1808 default: fatal_unexpected_iid(id); break;
1809 }
1810 set_result(_gvn.transform(n));
1811 return true;
1812 }
1813
1814 //------------------------------inline_math-----------------------------------
1815 // public static float Math.abs(float)
1816 // public static int Math.abs(int)
1817 // public static long Math.abs(long)
1818 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1819 Node* arg = argument(0);
1820 Node* n = nullptr;
1821 switch (id) {
1822 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1823 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1824 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1825 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1826 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1827 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1828 default: fatal_unexpected_iid(id); break;
1829 }
1830 set_result(_gvn.transform(n));
1831 return true;
1832 }
1833
1834 //------------------------------runtime_math-----------------------------
1835 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1836 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1837 "must be (DD)D or (D)D type");
1838
1839 // Inputs
1840 Node* a = argument(0);
1841 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1842
1843 const TypePtr* no_memory_effects = nullptr;
1844 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1845 no_memory_effects,
1846 a, top(), b, b ? top() : nullptr);
1847 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1848 #ifdef ASSERT
1849 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1850 assert(value_top == top(), "second value must be top");
1851 #endif
1852
1853 set_result(value);
1854 return true;
1855 }
1856
1857 //------------------------------inline_math_pow-----------------------------
1858 bool LibraryCallKit::inline_math_pow() {
1859 Node* exp = argument(2);
1860 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1861 if (d != nullptr) {
1862 if (d->getd() == 2.0) {
1863 // Special case: pow(x, 2.0) => x * x
1864 Node* base = argument(0);
1865 set_result(_gvn.transform(new MulDNode(base, base)));
1866 return true;
1867 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1868 // Special case: pow(x, 0.5) => sqrt(x)
1869 Node* base = argument(0);
1870 Node* zero = _gvn.zerocon(T_DOUBLE);
1871
1872 RegionNode* region = new RegionNode(3);
1873 Node* phi = new PhiNode(region, Type::DOUBLE);
1874
1875 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1876 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1877 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1878 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1879 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1880
1881 Node* if_pow = generate_slow_guard(test, nullptr);
1882 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1883 phi->init_req(1, value_sqrt);
1884 region->init_req(1, control());
1885
1886 if (if_pow != nullptr) {
1887 set_control(if_pow);
1888 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1889 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1890 const TypePtr* no_memory_effects = nullptr;
1891 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1892 no_memory_effects, base, top(), exp, top());
1893 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1894 #ifdef ASSERT
1895 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1896 assert(value_top == top(), "second value must be top");
1897 #endif
1898 phi->init_req(2, value_pow);
1899 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1900 }
1901
1902 C->set_has_split_ifs(true); // Has chance for split-if optimization
1903 set_control(_gvn.transform(region));
1904 record_for_igvn(region);
1905 set_result(_gvn.transform(phi));
1906
1907 return true;
1908 }
1909 }
1910
1911 return StubRoutines::dpow() != nullptr ?
1912 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1913 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1914 }
1915
1916 //------------------------------inline_math_native-----------------------------
1917 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1918 switch (id) {
1919 case vmIntrinsics::_dsin:
1920 return StubRoutines::dsin() != nullptr ?
1921 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1922 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1923 case vmIntrinsics::_dcos:
1924 return StubRoutines::dcos() != nullptr ?
1925 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1926 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1927 case vmIntrinsics::_dtan:
1928 return StubRoutines::dtan() != nullptr ?
1929 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1930 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1931 case vmIntrinsics::_dsinh:
1932 return StubRoutines::dsinh() != nullptr ?
1933 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1934 case vmIntrinsics::_dtanh:
1935 return StubRoutines::dtanh() != nullptr ?
1936 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1937 case vmIntrinsics::_dcbrt:
1938 return StubRoutines::dcbrt() != nullptr ?
1939 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1940 case vmIntrinsics::_dexp:
1941 return StubRoutines::dexp() != nullptr ?
1942 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1943 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1944 case vmIntrinsics::_dlog:
1945 return StubRoutines::dlog() != nullptr ?
1946 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1947 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1948 case vmIntrinsics::_dlog10:
1949 return StubRoutines::dlog10() != nullptr ?
1950 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1951 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1952
1953 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1954 case vmIntrinsics::_ceil:
1955 case vmIntrinsics::_floor:
1956 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1957
1958 case vmIntrinsics::_dsqrt:
1959 case vmIntrinsics::_dsqrt_strict:
1960 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1961 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1962 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1963 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1964 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1965
1966 case vmIntrinsics::_dpow: return inline_math_pow();
1967 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1968 case vmIntrinsics::_fcopySign: return inline_math(id);
1969 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1970 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1971 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1972
1973 // These intrinsics are not yet correctly implemented
1974 case vmIntrinsics::_datan2:
1975 return false;
1976
1977 default:
1978 fatal_unexpected_iid(id);
1979 return false;
1980 }
1981 }
1982
1983 //----------------------------inline_notify-----------------------------------*
1984 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1985 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1986 address func;
1987 if (id == vmIntrinsics::_notify) {
1988 func = OptoRuntime::monitor_notify_Java();
1989 } else {
1990 func = OptoRuntime::monitor_notifyAll_Java();
1991 }
1992 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1993 make_slow_call_ex(call, env()->Throwable_klass(), false);
1994 return true;
1995 }
1996
1997
1998 //----------------------------inline_min_max-----------------------------------
1999 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
2000 Node* a = nullptr;
2001 Node* b = nullptr;
2002 Node* n = nullptr;
2003 switch (id) {
2004 case vmIntrinsics::_min:
2005 case vmIntrinsics::_max:
2006 case vmIntrinsics::_minF:
2007 case vmIntrinsics::_maxF:
2008 case vmIntrinsics::_minF_strict:
2009 case vmIntrinsics::_maxF_strict:
2010 case vmIntrinsics::_min_strict:
2011 case vmIntrinsics::_max_strict:
2012 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
2013 a = argument(0);
2014 b = argument(1);
2015 break;
2016 case vmIntrinsics::_minD:
2017 case vmIntrinsics::_maxD:
2018 case vmIntrinsics::_minD_strict:
2019 case vmIntrinsics::_maxD_strict:
2020 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
2021 a = argument(0);
2022 b = argument(2);
2023 break;
2024 case vmIntrinsics::_minL:
2025 case vmIntrinsics::_maxL:
2026 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
2027 a = argument(0);
2028 b = argument(2);
2029 break;
2030 default:
2031 fatal_unexpected_iid(id);
2032 break;
2033 }
2034
2035 switch (id) {
2036 case vmIntrinsics::_min:
2037 case vmIntrinsics::_min_strict:
2038 n = new MinINode(a, b);
2039 break;
2040 case vmIntrinsics::_max:
2041 case vmIntrinsics::_max_strict:
2042 n = new MaxINode(a, b);
2043 break;
2044 case vmIntrinsics::_minF:
2045 case vmIntrinsics::_minF_strict:
2046 n = new MinFNode(a, b);
2047 break;
2048 case vmIntrinsics::_maxF:
2049 case vmIntrinsics::_maxF_strict:
2050 n = new MaxFNode(a, b);
2051 break;
2052 case vmIntrinsics::_minD:
2053 case vmIntrinsics::_minD_strict:
2054 n = new MinDNode(a, b);
2055 break;
2056 case vmIntrinsics::_maxD:
2057 case vmIntrinsics::_maxD_strict:
2058 n = new MaxDNode(a, b);
2059 break;
2060 case vmIntrinsics::_minL:
2061 n = new MinLNode(_gvn.C, a, b);
2062 break;
2063 case vmIntrinsics::_maxL:
2064 n = new MaxLNode(_gvn.C, a, b);
2065 break;
2066 default:
2067 fatal_unexpected_iid(id);
2068 break;
2069 }
2070
2071 set_result(_gvn.transform(n));
2072 return true;
2073 }
2074
2075 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2076 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2077 env()->ArithmeticException_instance())) {
2078 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2079 // so let's bail out intrinsic rather than risking deopting again.
2080 return false;
2081 }
2082
2083 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2084 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2085 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2086 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2087
2088 {
2089 PreserveJVMState pjvms(this);
2090 PreserveReexecuteState preexecs(this);
2091 jvms()->set_should_reexecute(true);
2092
2093 set_control(slow_path);
2094 set_i_o(i_o());
2095
2096 builtin_throw(Deoptimization::Reason_intrinsic,
2097 env()->ArithmeticException_instance(),
2098 /*allow_too_many_traps*/ false);
2099 }
2100
2101 set_control(fast_path);
2102 set_result(math);
2103 return true;
2104 }
2105
2106 template <typename OverflowOp>
2107 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2108 typedef typename OverflowOp::MathOp MathOp;
2109
2110 MathOp* mathOp = new MathOp(arg1, arg2);
2111 Node* operation = _gvn.transform( mathOp );
2112 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2113 return inline_math_mathExact(operation, ofcheck);
2114 }
2115
2116 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2117 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2118 }
2119
2120 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2121 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2122 }
2123
2124 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2125 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2126 }
2127
2128 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2129 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2130 }
2131
2132 bool LibraryCallKit::inline_math_negateExactI() {
2133 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2134 }
2135
2136 bool LibraryCallKit::inline_math_negateExactL() {
2137 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2138 }
2139
2140 bool LibraryCallKit::inline_math_multiplyExactI() {
2141 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2142 }
2143
2144 bool LibraryCallKit::inline_math_multiplyExactL() {
2145 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2146 }
2147
2148 bool LibraryCallKit::inline_math_multiplyHigh() {
2149 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2150 return true;
2151 }
2152
2153 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2154 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2155 return true;
2156 }
2157
2158 inline int
2159 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2160 const TypePtr* base_type = TypePtr::NULL_PTR;
2161 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2162 if (base_type == nullptr) {
2163 // Unknown type.
2164 return Type::AnyPtr;
2165 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2166 // Since this is a null+long form, we have to switch to a rawptr.
2167 base = _gvn.transform(new CastX2PNode(offset));
2168 offset = MakeConX(0);
2169 return Type::RawPtr;
2170 } else if (base_type->base() == Type::RawPtr) {
2171 return Type::RawPtr;
2172 } else if (base_type->isa_oopptr()) {
2173 // Base is never null => always a heap address.
2174 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2175 return Type::OopPtr;
2176 }
2177 // Offset is small => always a heap address.
2178 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2179 if (offset_type != nullptr &&
2180 base_type->offset() == 0 && // (should always be?)
2181 offset_type->_lo >= 0 &&
2182 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2183 return Type::OopPtr;
2184 } else if (type == T_OBJECT) {
2185 // off heap access to an oop doesn't make any sense. Has to be on
2186 // heap.
2187 return Type::OopPtr;
2188 }
2189 // Otherwise, it might either be oop+off or null+addr.
2190 return Type::AnyPtr;
2191 } else {
2192 // No information:
2193 return Type::AnyPtr;
2194 }
2195 }
2196
2197 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2198 Node* uncasted_base = base;
2199 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2200 if (kind == Type::RawPtr) {
2201 return basic_plus_adr(top(), uncasted_base, offset);
2202 } else if (kind == Type::AnyPtr) {
2203 assert(base == uncasted_base, "unexpected base change");
2204 if (can_cast) {
2205 if (!_gvn.type(base)->speculative_maybe_null() &&
2206 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2207 // According to profiling, this access is always on
2208 // heap. Casting the base to not null and thus avoiding membars
2209 // around the access should allow better optimizations
2210 Node* null_ctl = top();
2211 base = null_check_oop(base, &null_ctl, true, true, true);
2212 assert(null_ctl->is_top(), "no null control here");
2213 return basic_plus_adr(base, offset);
2214 } else if (_gvn.type(base)->speculative_always_null() &&
2215 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2216 // According to profiling, this access is always off
2217 // heap.
2218 base = null_assert(base);
2219 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2220 offset = MakeConX(0);
2221 return basic_plus_adr(top(), raw_base, offset);
2222 }
2223 }
2224 // We don't know if it's an on heap or off heap access. Fall back
2225 // to raw memory access.
2226 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2227 return basic_plus_adr(top(), raw, offset);
2228 } else {
2229 assert(base == uncasted_base, "unexpected base change");
2230 // We know it's an on heap access so base can't be null
2231 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2232 base = must_be_not_null(base, true);
2233 }
2234 return basic_plus_adr(base, offset);
2235 }
2236 }
2237
2238 //--------------------------inline_number_methods-----------------------------
2239 // inline int Integer.numberOfLeadingZeros(int)
2240 // inline int Long.numberOfLeadingZeros(long)
2241 //
2242 // inline int Integer.numberOfTrailingZeros(int)
2243 // inline int Long.numberOfTrailingZeros(long)
2244 //
2245 // inline int Integer.bitCount(int)
2246 // inline int Long.bitCount(long)
2247 //
2248 // inline char Character.reverseBytes(char)
2249 // inline short Short.reverseBytes(short)
2250 // inline int Integer.reverseBytes(int)
2251 // inline long Long.reverseBytes(long)
2252 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2253 Node* arg = argument(0);
2254 Node* n = nullptr;
2255 switch (id) {
2256 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2257 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2258 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2259 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2260 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2261 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2262 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2263 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2264 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2265 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2266 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2267 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2268 default: fatal_unexpected_iid(id); break;
2269 }
2270 set_result(_gvn.transform(n));
2271 return true;
2272 }
2273
2274 //--------------------------inline_bitshuffle_methods-----------------------------
2275 // inline int Integer.compress(int, int)
2276 // inline int Integer.expand(int, int)
2277 // inline long Long.compress(long, long)
2278 // inline long Long.expand(long, long)
2279 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2280 Node* n = nullptr;
2281 switch (id) {
2282 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2283 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2284 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2285 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2286 default: fatal_unexpected_iid(id); break;
2287 }
2288 set_result(_gvn.transform(n));
2289 return true;
2290 }
2291
2292 //--------------------------inline_number_methods-----------------------------
2293 // inline int Integer.compareUnsigned(int, int)
2294 // inline int Long.compareUnsigned(long, long)
2295 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2296 Node* arg1 = argument(0);
2297 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2298 Node* n = nullptr;
2299 switch (id) {
2300 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2301 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2302 default: fatal_unexpected_iid(id); break;
2303 }
2304 set_result(_gvn.transform(n));
2305 return true;
2306 }
2307
2308 //--------------------------inline_unsigned_divmod_methods-----------------------------
2309 // inline int Integer.divideUnsigned(int, int)
2310 // inline int Integer.remainderUnsigned(int, int)
2311 // inline long Long.divideUnsigned(long, long)
2312 // inline long Long.remainderUnsigned(long, long)
2313 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2314 Node* n = nullptr;
2315 switch (id) {
2316 case vmIntrinsics::_divideUnsigned_i: {
2317 zero_check_int(argument(1));
2318 // Compile-time detect of null-exception
2319 if (stopped()) {
2320 return true; // keep the graph constructed so far
2321 }
2322 n = new UDivINode(control(), argument(0), argument(1));
2323 break;
2324 }
2325 case vmIntrinsics::_divideUnsigned_l: {
2326 zero_check_long(argument(2));
2327 // Compile-time detect of null-exception
2328 if (stopped()) {
2329 return true; // keep the graph constructed so far
2330 }
2331 n = new UDivLNode(control(), argument(0), argument(2));
2332 break;
2333 }
2334 case vmIntrinsics::_remainderUnsigned_i: {
2335 zero_check_int(argument(1));
2336 // Compile-time detect of null-exception
2337 if (stopped()) {
2338 return true; // keep the graph constructed so far
2339 }
2340 n = new UModINode(control(), argument(0), argument(1));
2341 break;
2342 }
2343 case vmIntrinsics::_remainderUnsigned_l: {
2344 zero_check_long(argument(2));
2345 // Compile-time detect of null-exception
2346 if (stopped()) {
2347 return true; // keep the graph constructed so far
2348 }
2349 n = new UModLNode(control(), argument(0), argument(2));
2350 break;
2351 }
2352 default: fatal_unexpected_iid(id); break;
2353 }
2354 set_result(_gvn.transform(n));
2355 return true;
2356 }
2357
2358 //----------------------------inline_unsafe_access----------------------------
2359
2360 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2361 // Attempt to infer a sharper value type from the offset and base type.
2362 ciKlass* sharpened_klass = nullptr;
2363 bool null_free = false;
2364
2365 // See if it is an instance field, with an object type.
2366 if (alias_type->field() != nullptr) {
2367 if (alias_type->field()->type()->is_klass()) {
2368 sharpened_klass = alias_type->field()->type()->as_klass();
2369 null_free = alias_type->field()->is_null_free();
2370 }
2371 }
2372
2373 const TypeOopPtr* result = nullptr;
2374 // See if it is a narrow oop array.
2375 if (adr_type->isa_aryptr()) {
2376 if (adr_type->offset() >= refArrayOopDesc::base_offset_in_bytes()) {
2377 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2378 null_free = adr_type->is_aryptr()->is_null_free();
2379 if (elem_type != nullptr && elem_type->is_loaded()) {
2380 // Sharpen the value type.
2381 result = elem_type;
2382 }
2383 }
2384 }
2385
2386 // The sharpened class might be unloaded if there is no class loader
2387 // contraint in place.
2388 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2389 // Sharpen the value type.
2390 result = TypeOopPtr::make_from_klass(sharpened_klass);
2391 if (null_free) {
2392 result = result->join_speculative(TypePtr::NOTNULL)->is_oopptr();
2393 }
2394 }
2395 if (result != nullptr) {
2396 #ifndef PRODUCT
2397 if (C->print_intrinsics() || C->print_inlining()) {
2398 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2399 tty->print(" sharpened value: "); result->dump(); tty->cr();
2400 }
2401 #endif
2402 }
2403 return result;
2404 }
2405
2406 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2407 switch (kind) {
2408 case Relaxed:
2409 return MO_UNORDERED;
2410 case Opaque:
2411 return MO_RELAXED;
2412 case Acquire:
2413 return MO_ACQUIRE;
2414 case Release:
2415 return MO_RELEASE;
2416 case Volatile:
2417 return MO_SEQ_CST;
2418 default:
2419 ShouldNotReachHere();
2420 return 0;
2421 }
2422 }
2423
2424 LibraryCallKit::SavedState::SavedState(LibraryCallKit* kit) :
2425 _kit(kit),
2426 _sp(kit->sp()),
2427 _jvms(kit->jvms()),
2428 _map(kit->clone_map()),
2429 _discarded(false)
2430 {
2431 for (DUIterator_Fast imax, i = kit->control()->fast_outs(imax); i < imax; i++) {
2432 Node* out = kit->control()->fast_out(i);
2433 if (out->is_CFG()) {
2434 _ctrl_succ.push(out);
2435 }
2436 }
2437 }
2438
2439 LibraryCallKit::SavedState::~SavedState() {
2440 if (_discarded) {
2441 _kit->destruct_map_clone(_map);
2442 return;
2443 }
2444 _kit->jvms()->set_map(_map);
2445 _kit->jvms()->set_sp(_sp);
2446 _map->set_jvms(_kit->jvms());
2447 _kit->set_map(_map);
2448 _kit->set_sp(_sp);
2449 for (DUIterator_Fast imax, i = _kit->control()->fast_outs(imax); i < imax; i++) {
2450 Node* out = _kit->control()->fast_out(i);
2451 if (out->is_CFG() && out->in(0) == _kit->control() && out != _kit->map() && !_ctrl_succ.member(out)) {
2452 _kit->_gvn.hash_delete(out);
2453 out->set_req(0, _kit->C->top());
2454 _kit->C->record_for_igvn(out);
2455 --i; --imax;
2456 _kit->_gvn.hash_find_insert(out);
2457 }
2458 }
2459 }
2460
2461 void LibraryCallKit::SavedState::discard() {
2462 _discarded = true;
2463 }
2464
2465 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned, const bool is_flat) {
2466 if (callee()->is_static()) return false; // caller must have the capability!
2467 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2468 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2469 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2470 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2471
2472 if (is_reference_type(type)) {
2473 decorators |= ON_UNKNOWN_OOP_REF;
2474 }
2475
2476 if (unaligned) {
2477 decorators |= C2_UNALIGNED;
2478 }
2479
2480 #ifndef PRODUCT
2481 {
2482 ResourceMark rm;
2483 // Check the signatures.
2484 ciSignature* sig = callee()->signature();
2485 #ifdef ASSERT
2486 if (!is_store) {
2487 // Object getReference(Object base, int/long offset), etc.
2488 BasicType rtype = sig->return_type()->basic_type();
2489 assert(rtype == type, "getter must return the expected value");
2490 assert(sig->count() == 2 || (is_flat && sig->count() == 3), "oop getter has 2 or 3 arguments");
2491 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2492 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2493 } else {
2494 // void putReference(Object base, int/long offset, Object x), etc.
2495 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2496 assert(sig->count() == 3 || (is_flat && sig->count() == 4), "oop putter has 3 arguments");
2497 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2498 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2499 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2500 assert(vtype == type, "putter must accept the expected value");
2501 }
2502 #endif // ASSERT
2503 }
2504 #endif //PRODUCT
2505
2506 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2507
2508 Node* receiver = argument(0); // type: oop
2509
2510 // Build address expression.
2511 Node* heap_base_oop = top();
2512
2513 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2514 Node* base = argument(1); // type: oop
2515 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2516 Node* offset = argument(2); // type: long
2517 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2518 // to be plain byte offsets, which are also the same as those accepted
2519 // by oopDesc::field_addr.
2520 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2521 "fieldOffset must be byte-scaled");
2522
2523 ciInlineKlass* inline_klass = nullptr;
2524 if (is_flat) {
2525 const TypeInstPtr* cls = _gvn.type(argument(4))->isa_instptr();
2526 if (cls == nullptr || cls->const_oop() == nullptr) {
2527 return false;
2528 }
2529 ciType* mirror_type = cls->const_oop()->as_instance()->java_mirror_type();
2530 if (!mirror_type->is_inlinetype()) {
2531 return false;
2532 }
2533 inline_klass = mirror_type->as_inline_klass();
2534 }
2535
2536 if (base->is_InlineType()) {
2537 assert(!is_store, "InlineTypeNodes are non-larval value objects");
2538 InlineTypeNode* vt = base->as_InlineType();
2539 if (offset->is_Con()) {
2540 long off = find_long_con(offset, 0);
2541 ciInlineKlass* vk = vt->type()->inline_klass();
2542 if ((long)(int)off != off || !vk->contains_field_offset(off)) {
2543 return false;
2544 }
2545
2546 ciField* field = vk->get_non_flat_field_by_offset(off);
2547 if (field != nullptr) {
2548 BasicType bt = type2field[field->type()->basic_type()];
2549 if (bt == T_ARRAY || bt == T_NARROWOOP) {
2550 bt = T_OBJECT;
2551 }
2552 if (bt == type && (!field->is_flat() || field->type() == inline_klass)) {
2553 Node* value = vt->field_value_by_offset(off, false);
2554 if (value->is_InlineType()) {
2555 value = value->as_InlineType()->adjust_scalarization_depth(this);
2556 }
2557 set_result(value);
2558 return true;
2559 }
2560 }
2561 }
2562 {
2563 // Re-execute the unsafe access if allocation triggers deoptimization.
2564 PreserveReexecuteState preexecs(this);
2565 jvms()->set_should_reexecute(true);
2566 vt = vt->buffer(this);
2567 }
2568 base = vt->get_oop();
2569 }
2570
2571 // 32-bit machines ignore the high half!
2572 offset = ConvL2X(offset);
2573
2574 // Save state and restore on bailout
2575 SavedState old_state(this);
2576
2577 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2578 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2579
2580 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2581 if (type != T_OBJECT && (inline_klass == nullptr || !inline_klass->has_object_fields())) {
2582 decorators |= IN_NATIVE; // off-heap primitive access
2583 } else {
2584 return false; // off-heap oop accesses are not supported
2585 }
2586 } else {
2587 heap_base_oop = base; // on-heap or mixed access
2588 }
2589
2590 // Can base be null? Otherwise, always on-heap access.
2591 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2592
2593 if (!can_access_non_heap) {
2594 decorators |= IN_HEAP;
2595 }
2596
2597 Node* val = is_store ? argument(4 + (is_flat ? 1 : 0)) : nullptr;
2598
2599 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2600 if (adr_type == TypePtr::NULL_PTR) {
2601 return false; // off-heap access with zero address
2602 }
2603
2604 // Try to categorize the address.
2605 Compile::AliasType* alias_type = C->alias_type(adr_type);
2606 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2607
2608 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2609 alias_type->adr_type() == TypeAryPtr::RANGE) {
2610 return false; // not supported
2611 }
2612
2613 bool mismatched = false;
2614 BasicType bt = T_ILLEGAL;
2615 ciField* field = nullptr;
2616 if (adr_type->isa_instptr()) {
2617 const TypeInstPtr* instptr = adr_type->is_instptr();
2618 ciInstanceKlass* k = instptr->instance_klass();
2619 int off = instptr->offset();
2620 if (instptr->const_oop() != nullptr &&
2621 k == ciEnv::current()->Class_klass() &&
2622 instptr->offset() >= (k->size_helper() * wordSize)) {
2623 k = instptr->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
2624 field = k->get_field_by_offset(off, true);
2625 } else {
2626 field = k->get_non_flat_field_by_offset(off);
2627 }
2628 if (field != nullptr) {
2629 bt = type2field[field->type()->basic_type()];
2630 }
2631 if (bt != alias_type->basic_type()) {
2632 // Type mismatch. Is it an access to a nested flat field?
2633 field = k->get_field_by_offset(off, false);
2634 if (field != nullptr) {
2635 bt = type2field[field->type()->basic_type()];
2636 }
2637 }
2638 assert(bt == alias_type->basic_type() || is_flat, "should match");
2639 } else {
2640 bt = alias_type->basic_type();
2641 }
2642
2643 if (bt != T_ILLEGAL) {
2644 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2645 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2646 // Alias type doesn't differentiate between byte[] and boolean[]).
2647 // Use address type to get the element type.
2648 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2649 }
2650 if (is_reference_type(bt, true)) {
2651 // accessing an array field with getReference is not a mismatch
2652 bt = T_OBJECT;
2653 }
2654 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2655 // Don't intrinsify mismatched object accesses
2656 return false;
2657 }
2658 mismatched = (bt != type);
2659 } else if (alias_type->adr_type()->isa_oopptr()) {
2660 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2661 }
2662
2663 if (is_flat) {
2664 if (adr_type->isa_instptr()) {
2665 if (field == nullptr || field->type() != inline_klass) {
2666 mismatched = true;
2667 }
2668 } else if (adr_type->isa_aryptr()) {
2669 const Type* elem = adr_type->is_aryptr()->elem();
2670 if (!adr_type->is_flat() || elem->inline_klass() != inline_klass) {
2671 mismatched = true;
2672 }
2673 } else {
2674 mismatched = true;
2675 }
2676 if (is_store) {
2677 const Type* val_t = _gvn.type(val);
2678 if (!val_t->is_inlinetypeptr() || val_t->inline_klass() != inline_klass) {
2679 return false;
2680 }
2681 }
2682 }
2683
2684 old_state.discard();
2685 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2686
2687 if (mismatched) {
2688 decorators |= C2_MISMATCHED;
2689 }
2690
2691 // First guess at the value type.
2692 const Type *value_type = Type::get_const_basic_type(type);
2693
2694 // Figure out the memory ordering.
2695 decorators |= mo_decorator_for_access_kind(kind);
2696
2697 if (!is_store) {
2698 if (type == T_OBJECT && !is_flat) {
2699 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2700 if (tjp != nullptr) {
2701 value_type = tjp;
2702 }
2703 }
2704 }
2705
2706 receiver = null_check(receiver);
2707 if (stopped()) {
2708 return true;
2709 }
2710 // Heap pointers get a null-check from the interpreter,
2711 // as a courtesy. However, this is not guaranteed by Unsafe,
2712 // and it is not possible to fully distinguish unintended nulls
2713 // from intended ones in this API.
2714
2715 if (!is_store) {
2716 Node* p = nullptr;
2717 // Try to constant fold a load from a constant field
2718
2719 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !field->is_flat() && !mismatched) {
2720 // final or stable field
2721 p = make_constant_from_field(field, heap_base_oop);
2722 }
2723
2724 if (p == nullptr) { // Could not constant fold the load
2725 if (is_flat) {
2726 p = InlineTypeNode::make_from_flat(this, inline_klass, base, adr, adr_type, false, false, true);
2727 } else {
2728 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2729 const TypeOopPtr* ptr = value_type->make_oopptr();
2730 if (ptr != nullptr && ptr->is_inlinetypeptr()) {
2731 // Load a non-flattened inline type from memory
2732 p = InlineTypeNode::make_from_oop(this, p, ptr->inline_klass());
2733 }
2734 }
2735 // Normalize the value returned by getBoolean in the following cases
2736 if (type == T_BOOLEAN &&
2737 (mismatched ||
2738 heap_base_oop == top() || // - heap_base_oop is null or
2739 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2740 // and the unsafe access is made to large offset
2741 // (i.e., larger than the maximum offset necessary for any
2742 // field access)
2743 ) {
2744 IdealKit ideal = IdealKit(this);
2745 #define __ ideal.
2746 IdealVariable normalized_result(ideal);
2747 __ declarations_done();
2748 __ set(normalized_result, p);
2749 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2750 __ set(normalized_result, ideal.ConI(1));
2751 ideal.end_if();
2752 final_sync(ideal);
2753 p = __ value(normalized_result);
2754 #undef __
2755 }
2756 }
2757 if (type == T_ADDRESS) {
2758 p = gvn().transform(new CastP2XNode(nullptr, p));
2759 p = ConvX2UL(p);
2760 }
2761 // The load node has the control of the preceding MemBarCPUOrder. All
2762 // following nodes will have the control of the MemBarCPUOrder inserted at
2763 // the end of this method. So, pushing the load onto the stack at a later
2764 // point is fine.
2765 set_result(p);
2766 } else {
2767 if (bt == T_ADDRESS) {
2768 // Repackage the long as a pointer.
2769 val = ConvL2X(val);
2770 val = gvn().transform(new CastX2PNode(val));
2771 }
2772 if (is_flat) {
2773 val->as_InlineType()->store_flat(this, base, adr, false, false, true, decorators);
2774 } else {
2775 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2776 }
2777 }
2778
2779 return true;
2780 }
2781
2782 bool LibraryCallKit::inline_unsafe_flat_access(bool is_store, AccessKind kind) {
2783 #ifdef ASSERT
2784 {
2785 ResourceMark rm;
2786 // Check the signatures.
2787 ciSignature* sig = callee()->signature();
2788 assert(sig->type_at(0)->basic_type() == T_OBJECT, "base should be object, but is %s", type2name(sig->type_at(0)->basic_type()));
2789 assert(sig->type_at(1)->basic_type() == T_LONG, "offset should be long, but is %s", type2name(sig->type_at(1)->basic_type()));
2790 assert(sig->type_at(2)->basic_type() == T_INT, "layout kind should be int, but is %s", type2name(sig->type_at(3)->basic_type()));
2791 assert(sig->type_at(3)->basic_type() == T_OBJECT, "value klass should be object, but is %s", type2name(sig->type_at(4)->basic_type()));
2792 if (is_store) {
2793 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value, but returns %s", type2name(sig->return_type()->basic_type()));
2794 assert(sig->count() == 5, "flat putter should have 5 arguments, but has %d", sig->count());
2795 assert(sig->type_at(4)->basic_type() == T_OBJECT, "put value should be object, but is %s", type2name(sig->type_at(5)->basic_type()));
2796 } else {
2797 assert(sig->return_type()->basic_type() == T_OBJECT, "getter must return an object, but returns %s", type2name(sig->return_type()->basic_type()));
2798 assert(sig->count() == 4, "flat getter should have 4 arguments, but has %d", sig->count());
2799 }
2800 }
2801 #endif // ASSERT
2802
2803 assert(kind == Relaxed, "Only plain accesses for now");
2804 if (callee()->is_static()) {
2805 // caller must have the capability!
2806 return false;
2807 }
2808 C->set_has_unsafe_access(true);
2809
2810 const TypeInstPtr* value_klass_node = _gvn.type(argument(5))->isa_instptr();
2811 if (value_klass_node == nullptr || value_klass_node->const_oop() == nullptr) {
2812 // parameter valueType is not a constant
2813 return false;
2814 }
2815 ciType* mirror_type = value_klass_node->const_oop()->as_instance()->java_mirror_type();
2816 if (!mirror_type->is_inlinetype()) {
2817 // Dead code
2818 return false;
2819 }
2820 ciInlineKlass* value_klass = mirror_type->as_inline_klass();
2821
2822 const TypeInt* layout_type = _gvn.type(argument(4))->isa_int();
2823 if (layout_type == nullptr || !layout_type->is_con()) {
2824 // parameter layoutKind is not a constant
2825 return false;
2826 }
2827 assert(layout_type->get_con() >= static_cast<int>(LayoutKind::REFERENCE) &&
2828 layout_type->get_con() <= static_cast<int>(LayoutKind::UNKNOWN),
2829 "invalid layoutKind %d", layout_type->get_con());
2830 LayoutKind layout = static_cast<LayoutKind>(layout_type->get_con());
2831 assert(layout == LayoutKind::REFERENCE || layout == LayoutKind::NON_ATOMIC_FLAT ||
2832 layout == LayoutKind::ATOMIC_FLAT || layout == LayoutKind::NULLABLE_ATOMIC_FLAT,
2833 "unexpected layoutKind %d", layout_type->get_con());
2834
2835 null_check(argument(0));
2836 if (stopped()) {
2837 return true;
2838 }
2839
2840 Node* base = must_be_not_null(argument(1), true);
2841 Node* offset = argument(2);
2842 const Type* base_type = _gvn.type(base);
2843
2844 Node* ptr;
2845 bool immutable_memory = false;
2846 DecoratorSet decorators = C2_UNSAFE_ACCESS | IN_HEAP | MO_UNORDERED;
2847 if (base_type->isa_instptr()) {
2848 const TypeLong* offset_type = _gvn.type(offset)->isa_long();
2849 if (offset_type == nullptr || !offset_type->is_con()) {
2850 // Offset into a non-array should be a constant
2851 decorators |= C2_MISMATCHED;
2852 } else {
2853 int offset_con = checked_cast<int>(offset_type->get_con());
2854 ciInstanceKlass* base_klass = base_type->is_instptr()->instance_klass();
2855 ciField* field = base_klass->get_non_flat_field_by_offset(offset_con);
2856 if (field == nullptr) {
2857 assert(!base_klass->is_final(), "non-existence field at offset %d of class %s", offset_con, base_klass->name()->as_utf8());
2858 decorators |= C2_MISMATCHED;
2859 } else {
2860 assert(field->type() == value_klass, "field at offset %d of %s is of type %s, but valueType is %s",
2861 offset_con, base_klass->name()->as_utf8(), field->type()->name(), value_klass->name()->as_utf8());
2862 immutable_memory = field->is_strict() && field->is_final();
2863
2864 if (base->is_InlineType()) {
2865 assert(!is_store, "Cannot store into a non-larval value object");
2866 set_result(base->as_InlineType()->field_value_by_offset(offset_con, false));
2867 return true;
2868 }
2869 }
2870 }
2871
2872 if (base->is_InlineType()) {
2873 assert(!is_store, "Cannot store into a non-larval value object");
2874 base = base->as_InlineType()->buffer(this, true);
2875 }
2876 ptr = basic_plus_adr(base, ConvL2X(offset));
2877 } else if (base_type->isa_aryptr()) {
2878 decorators |= IS_ARRAY;
2879 if (layout == LayoutKind::REFERENCE) {
2880 if (!base_type->is_aryptr()->is_not_flat()) {
2881 const TypeAryPtr* array_type = base_type->is_aryptr()->cast_to_not_flat();
2882 Node* new_base = _gvn.transform(new CastPPNode(control(), base, array_type, ConstraintCastNode::StrongDependency));
2883 replace_in_map(base, new_base);
2884 base = new_base;
2885 }
2886 ptr = basic_plus_adr(base, ConvL2X(offset));
2887 } else {
2888 if (UseArrayFlattening) {
2889 // Flat array must have an exact type
2890 bool is_null_free = !LayoutKindHelper::is_nullable_flat(layout);
2891 bool is_atomic = LayoutKindHelper::is_atomic_flat(layout);
2892 Node* new_base = cast_to_flat_array(base, value_klass, is_null_free, !is_null_free, is_atomic);
2893 replace_in_map(base, new_base);
2894 base = new_base;
2895 ptr = basic_plus_adr(base, ConvL2X(offset));
2896 const TypeAryPtr* ptr_type = _gvn.type(ptr)->is_aryptr();
2897 if (ptr_type->field_offset().get() != 0) {
2898 ptr = _gvn.transform(new CastPPNode(control(), ptr, ptr_type->with_field_offset(0), ConstraintCastNode::StrongDependency));
2899 }
2900 } else {
2901 uncommon_trap(Deoptimization::Reason_intrinsic,
2902 Deoptimization::Action_none);
2903 return true;
2904 }
2905 }
2906 } else {
2907 decorators |= C2_MISMATCHED;
2908 ptr = basic_plus_adr(base, ConvL2X(offset));
2909 }
2910
2911 if (is_store) {
2912 Node* value = argument(6);
2913 const Type* value_type = _gvn.type(value);
2914 if (!value_type->is_inlinetypeptr()) {
2915 value_type = Type::get_const_type(value_klass)->filter_speculative(value_type);
2916 Node* new_value = _gvn.transform(new CastPPNode(control(), value, value_type, ConstraintCastNode::StrongDependency));
2917 new_value = InlineTypeNode::make_from_oop(this, new_value, value_klass);
2918 replace_in_map(value, new_value);
2919 value = new_value;
2920 }
2921
2922 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());
2923 if (layout == LayoutKind::REFERENCE) {
2924 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2925 access_store_at(base, ptr, ptr_type, value, value_type, T_OBJECT, decorators);
2926 } else {
2927 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2928 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2929 value->as_InlineType()->store_flat(this, base, ptr, atomic, immutable_memory, null_free, decorators);
2930 }
2931
2932 return true;
2933 } else {
2934 decorators |= (C2_CONTROL_DEPENDENT_LOAD | C2_UNKNOWN_CONTROL_LOAD);
2935 InlineTypeNode* result;
2936 if (layout == LayoutKind::REFERENCE) {
2937 const TypePtr* ptr_type = (decorators & C2_MISMATCHED) != 0 ? TypeRawPtr::BOTTOM : _gvn.type(ptr)->is_ptr();
2938 Node* oop = access_load_at(base, ptr, ptr_type, Type::get_const_type(value_klass), T_OBJECT, decorators);
2939 result = InlineTypeNode::make_from_oop(this, oop, value_klass);
2940 } else {
2941 bool atomic = LayoutKindHelper::is_atomic_flat(layout);
2942 bool null_free = !LayoutKindHelper::is_nullable_flat(layout);
2943 result = InlineTypeNode::make_from_flat(this, value_klass, base, ptr, atomic, immutable_memory, null_free, decorators);
2944 }
2945
2946 set_result(result);
2947 return true;
2948 }
2949 }
2950
2951 bool LibraryCallKit::inline_unsafe_make_private_buffer() {
2952 Node* receiver = argument(0);
2953 Node* value = argument(1);
2954
2955 const Type* type = gvn().type(value);
2956 if (!type->is_inlinetypeptr()) {
2957 C->record_method_not_compilable("value passed to Unsafe::makePrivateBuffer is not of a constant value type");
2958 return false;
2959 }
2960
2961 null_check(receiver);
2962 if (stopped()) {
2963 return true;
2964 }
2965
2966 value = null_check(value);
2967 if (stopped()) {
2968 return true;
2969 }
2970
2971 ciInlineKlass* vk = type->inline_klass();
2972 Node* klass = makecon(TypeKlassPtr::make(vk));
2973 Node* obj = new_instance(klass);
2974 AllocateNode::Ideal_allocation(obj)->_larval = true;
2975
2976 assert(value->is_InlineType(), "must be an InlineTypeNode");
2977 Node* payload_ptr = basic_plus_adr(obj, vk->payload_offset());
2978 value->as_InlineType()->store_flat(this, obj, payload_ptr, false, true, true, IN_HEAP | MO_UNORDERED);
2979
2980 set_result(obj);
2981 return true;
2982 }
2983
2984 bool LibraryCallKit::inline_unsafe_finish_private_buffer() {
2985 Node* receiver = argument(0);
2986 Node* buffer = argument(1);
2987
2988 const Type* type = gvn().type(buffer);
2989 if (!type->is_inlinetypeptr()) {
2990 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer is not of a constant value type");
2991 return false;
2992 }
2993
2994 AllocateNode* alloc = AllocateNode::Ideal_allocation(buffer);
2995 if (alloc == nullptr) {
2996 C->record_method_not_compilable("value passed to Unsafe::finishPrivateBuffer must be allocated by Unsafe::makePrivateBuffer");
2997 return false;
2998 }
2999
3000 null_check(receiver);
3001 if (stopped()) {
3002 return true;
3003 }
3004
3005 // Unset the larval bit in the object header
3006 Node* old_header = make_load(control(), buffer, TypeX_X, TypeX_X->basic_type(), MemNode::unordered, LoadNode::Pinned);
3007 Node* new_header = gvn().transform(new AndXNode(old_header, MakeConX(~markWord::larval_bit_in_place)));
3008 access_store_at(buffer, buffer, type->is_ptr(), new_header, TypeX_X, TypeX_X->basic_type(), MO_UNORDERED | IN_HEAP);
3009
3010 // We must ensure that the buffer is properly published
3011 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out(AllocateNode::RawAddress));
3012 assert(!type->maybe_null(), "result of an allocation should not be null");
3013 set_result(InlineTypeNode::make_from_oop(this, buffer, type->inline_klass()));
3014 return true;
3015 }
3016
3017 //----------------------------inline_unsafe_load_store----------------------------
3018 // This method serves a couple of different customers (depending on LoadStoreKind):
3019 //
3020 // LS_cmp_swap:
3021 //
3022 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
3023 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
3024 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
3025 //
3026 // LS_cmp_swap_weak:
3027 //
3028 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
3029 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
3030 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
3031 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
3032 //
3033 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
3034 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
3035 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
3036 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
3037 //
3038 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
3039 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
3040 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
3041 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
3042 //
3043 // LS_cmp_exchange:
3044 //
3045 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
3046 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
3047 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
3048 //
3049 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
3050 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
3051 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
3052 //
3053 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
3054 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
3055 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
3056 //
3057 // LS_get_add:
3058 //
3059 // int getAndAddInt( Object o, long offset, int delta)
3060 // long getAndAddLong(Object o, long offset, long delta)
3061 //
3062 // LS_get_set:
3063 //
3064 // int getAndSet(Object o, long offset, int newValue)
3065 // long getAndSet(Object o, long offset, long newValue)
3066 // Object getAndSet(Object o, long offset, Object newValue)
3067 //
3068 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
3069 // This basic scheme here is the same as inline_unsafe_access, but
3070 // differs in enough details that combining them would make the code
3071 // overly confusing. (This is a true fact! I originally combined
3072 // them, but even I was confused by it!) As much code/comments as
3073 // possible are retained from inline_unsafe_access though to make
3074 // the correspondences clearer. - dl
3075
3076 if (callee()->is_static()) return false; // caller must have the capability!
3077
3078 DecoratorSet decorators = C2_UNSAFE_ACCESS;
3079 decorators |= mo_decorator_for_access_kind(access_kind);
3080
3081 #ifndef PRODUCT
3082 BasicType rtype;
3083 {
3084 ResourceMark rm;
3085 // Check the signatures.
3086 ciSignature* sig = callee()->signature();
3087 rtype = sig->return_type()->basic_type();
3088 switch(kind) {
3089 case LS_get_add:
3090 case LS_get_set: {
3091 // Check the signatures.
3092 #ifdef ASSERT
3093 assert(rtype == type, "get and set must return the expected type");
3094 assert(sig->count() == 3, "get and set has 3 arguments");
3095 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
3096 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
3097 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
3098 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
3099 #endif // ASSERT
3100 break;
3101 }
3102 case LS_cmp_swap:
3103 case LS_cmp_swap_weak: {
3104 // Check the signatures.
3105 #ifdef ASSERT
3106 assert(rtype == T_BOOLEAN, "CAS must return boolean");
3107 assert(sig->count() == 4, "CAS has 4 arguments");
3108 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3109 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3110 #endif // ASSERT
3111 break;
3112 }
3113 case LS_cmp_exchange: {
3114 // Check the signatures.
3115 #ifdef ASSERT
3116 assert(rtype == type, "CAS must return the expected type");
3117 assert(sig->count() == 4, "CAS has 4 arguments");
3118 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
3119 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
3120 #endif // ASSERT
3121 break;
3122 }
3123 default:
3124 ShouldNotReachHere();
3125 }
3126 }
3127 #endif //PRODUCT
3128
3129 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
3130
3131 // Get arguments:
3132 Node* receiver = nullptr;
3133 Node* base = nullptr;
3134 Node* offset = nullptr;
3135 Node* oldval = nullptr;
3136 Node* newval = nullptr;
3137 switch(kind) {
3138 case LS_cmp_swap:
3139 case LS_cmp_swap_weak:
3140 case LS_cmp_exchange: {
3141 const bool two_slot_type = type2size[type] == 2;
3142 receiver = argument(0); // type: oop
3143 base = argument(1); // type: oop
3144 offset = argument(2); // type: long
3145 oldval = argument(4); // type: oop, int, or long
3146 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
3147 break;
3148 }
3149 case LS_get_add:
3150 case LS_get_set: {
3151 receiver = argument(0); // type: oop
3152 base = argument(1); // type: oop
3153 offset = argument(2); // type: long
3154 oldval = nullptr;
3155 newval = argument(4); // type: oop, int, or long
3156 break;
3157 }
3158 default:
3159 ShouldNotReachHere();
3160 }
3161
3162 // Build field offset expression.
3163 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
3164 // to be plain byte offsets, which are also the same as those accepted
3165 // by oopDesc::field_addr.
3166 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
3167 // 32-bit machines ignore the high half of long offsets
3168 offset = ConvL2X(offset);
3169 // Save state and restore on bailout
3170 SavedState old_state(this);
3171 Node* adr = make_unsafe_address(base, offset,type, false);
3172 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
3173
3174 Compile::AliasType* alias_type = C->alias_type(adr_type);
3175 BasicType bt = alias_type->basic_type();
3176 if (bt != T_ILLEGAL &&
3177 (is_reference_type(bt) != (type == T_OBJECT))) {
3178 // Don't intrinsify mismatched object accesses.
3179 return false;
3180 }
3181
3182 old_state.discard();
3183
3184 // For CAS, unlike inline_unsafe_access, there seems no point in
3185 // trying to refine types. Just use the coarse types here.
3186 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
3187 const Type *value_type = Type::get_const_basic_type(type);
3188
3189 switch (kind) {
3190 case LS_get_set:
3191 case LS_cmp_exchange: {
3192 if (type == T_OBJECT) {
3193 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
3194 if (tjp != nullptr) {
3195 value_type = tjp;
3196 }
3197 }
3198 break;
3199 }
3200 case LS_cmp_swap:
3201 case LS_cmp_swap_weak:
3202 case LS_get_add:
3203 break;
3204 default:
3205 ShouldNotReachHere();
3206 }
3207
3208 // Null check receiver.
3209 receiver = null_check(receiver);
3210 if (stopped()) {
3211 return true;
3212 }
3213
3214 int alias_idx = C->get_alias_index(adr_type);
3215
3216 if (is_reference_type(type)) {
3217 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
3218
3219 if (oldval != nullptr && oldval->is_InlineType()) {
3220 // Re-execute the unsafe access if allocation triggers deoptimization.
3221 PreserveReexecuteState preexecs(this);
3222 jvms()->set_should_reexecute(true);
3223 oldval = oldval->as_InlineType()->buffer(this)->get_oop();
3224 }
3225 if (newval != nullptr && newval->is_InlineType()) {
3226 // Re-execute the unsafe access if allocation triggers deoptimization.
3227 PreserveReexecuteState preexecs(this);
3228 jvms()->set_should_reexecute(true);
3229 newval = newval->as_InlineType()->buffer(this)->get_oop();
3230 }
3231
3232 // Transformation of a value which could be null pointer (CastPP #null)
3233 // could be delayed during Parse (for example, in adjust_map_after_if()).
3234 // Execute transformation here to avoid barrier generation in such case.
3235 if (_gvn.type(newval) == TypePtr::NULL_PTR)
3236 newval = _gvn.makecon(TypePtr::NULL_PTR);
3237
3238 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
3239 // Refine the value to a null constant, when it is known to be null
3240 oldval = _gvn.makecon(TypePtr::NULL_PTR);
3241 }
3242 }
3243
3244 Node* result = nullptr;
3245 switch (kind) {
3246 case LS_cmp_exchange: {
3247 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
3248 oldval, newval, value_type, type, decorators);
3249 break;
3250 }
3251 case LS_cmp_swap_weak:
3252 decorators |= C2_WEAK_CMPXCHG;
3253 case LS_cmp_swap: {
3254 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
3255 oldval, newval, value_type, type, decorators);
3256 break;
3257 }
3258 case LS_get_set: {
3259 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
3260 newval, value_type, type, decorators);
3261 break;
3262 }
3263 case LS_get_add: {
3264 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
3265 newval, value_type, type, decorators);
3266 break;
3267 }
3268 default:
3269 ShouldNotReachHere();
3270 }
3271
3272 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
3273 set_result(result);
3274 return true;
3275 }
3276
3277 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
3278 // Regardless of form, don't allow previous ld/st to move down,
3279 // then issue acquire, release, or volatile mem_bar.
3280 insert_mem_bar(Op_MemBarCPUOrder);
3281 switch(id) {
3282 case vmIntrinsics::_loadFence:
3283 insert_mem_bar(Op_LoadFence);
3284 return true;
3285 case vmIntrinsics::_storeFence:
3286 insert_mem_bar(Op_StoreFence);
3287 return true;
3288 case vmIntrinsics::_storeStoreFence:
3289 insert_mem_bar(Op_StoreStoreFence);
3290 return true;
3291 case vmIntrinsics::_fullFence:
3292 insert_mem_bar(Op_MemBarVolatile);
3293 return true;
3294 default:
3295 fatal_unexpected_iid(id);
3296 return false;
3297 }
3298 }
3299
3300 // private native int arrayInstanceBaseOffset0(Object[] array);
3301 bool LibraryCallKit::inline_arrayInstanceBaseOffset() {
3302 Node* array = argument(1);
3303 Node* klass_node = load_object_klass(array);
3304
3305 jint layout_con = Klass::_lh_neutral_value;
3306 Node* layout_val = get_layout_helper(klass_node, layout_con);
3307 int layout_is_con = (layout_val == nullptr);
3308
3309 Node* header_size = nullptr;
3310 if (layout_is_con) {
3311 int hsize = Klass::layout_helper_header_size(layout_con);
3312 header_size = intcon(hsize);
3313 } else {
3314 Node* hss = intcon(Klass::_lh_header_size_shift);
3315 Node* hsm = intcon(Klass::_lh_header_size_mask);
3316 header_size = _gvn.transform(new URShiftINode(layout_val, hss));
3317 header_size = _gvn.transform(new AndINode(header_size, hsm));
3318 }
3319 set_result(header_size);
3320 return true;
3321 }
3322
3323 // private native int arrayInstanceIndexScale0(Object[] array);
3324 bool LibraryCallKit::inline_arrayInstanceIndexScale() {
3325 Node* array = argument(1);
3326 Node* klass_node = load_object_klass(array);
3327
3328 jint layout_con = Klass::_lh_neutral_value;
3329 Node* layout_val = get_layout_helper(klass_node, layout_con);
3330 int layout_is_con = (layout_val == nullptr);
3331
3332 Node* element_size = nullptr;
3333 if (layout_is_con) {
3334 int log_element_size = Klass::layout_helper_log2_element_size(layout_con);
3335 int elem_size = 1 << log_element_size;
3336 element_size = intcon(elem_size);
3337 } else {
3338 Node* ess = intcon(Klass::_lh_log2_element_size_shift);
3339 Node* esm = intcon(Klass::_lh_log2_element_size_mask);
3340 Node* log_element_size = _gvn.transform(new URShiftINode(layout_val, ess));
3341 log_element_size = _gvn.transform(new AndINode(log_element_size, esm));
3342 element_size = _gvn.transform(new LShiftINode(intcon(1), log_element_size));
3343 }
3344 set_result(element_size);
3345 return true;
3346 }
3347
3348 // private native int arrayLayout0(Object[] array);
3349 bool LibraryCallKit::inline_arrayLayout() {
3350 RegionNode* region = new RegionNode(2);
3351 Node* phi = new PhiNode(region, TypeInt::POS);
3352
3353 Node* array = argument(1);
3354 Node* klass_node = load_object_klass(array);
3355 generate_refArray_guard(klass_node, region);
3356 if (region->req() == 3) {
3357 phi->add_req(intcon((jint)LayoutKind::REFERENCE));
3358 }
3359
3360 int layout_kind_offset = in_bytes(FlatArrayKlass::layout_kind_offset());
3361 Node* layout_kind_addr = basic_plus_adr(klass_node, klass_node, layout_kind_offset);
3362 Node* layout_kind = make_load(nullptr, layout_kind_addr, TypeInt::POS, T_INT, MemNode::unordered);
3363
3364 region->init_req(1, control());
3365 phi->init_req(1, layout_kind);
3366
3367 set_control(_gvn.transform(region));
3368 set_result(_gvn.transform(phi));
3369 return true;
3370 }
3371
3372 bool LibraryCallKit::inline_onspinwait() {
3373 insert_mem_bar(Op_OnSpinWait);
3374 return true;
3375 }
3376
3377 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
3378 if (!kls->is_Con()) {
3379 return true;
3380 }
3381 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
3382 if (klsptr == nullptr) {
3383 return true;
3384 }
3385 ciInstanceKlass* ik = klsptr->instance_klass();
3386 // don't need a guard for a klass that is already initialized
3387 return !ik->is_initialized();
3388 }
3389
3390 //----------------------------inline_unsafe_writeback0-------------------------
3391 // public native void Unsafe.writeback0(long address)
3392 bool LibraryCallKit::inline_unsafe_writeback0() {
3393 if (!Matcher::has_match_rule(Op_CacheWB)) {
3394 return false;
3395 }
3396 #ifndef PRODUCT
3397 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
3398 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
3399 ciSignature* sig = callee()->signature();
3400 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
3401 #endif
3402 null_check_receiver(); // null-check, then ignore
3403 Node *addr = argument(1);
3404 addr = new CastX2PNode(addr);
3405 addr = _gvn.transform(addr);
3406 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
3407 flush = _gvn.transform(flush);
3408 set_memory(flush, TypeRawPtr::BOTTOM);
3409 return true;
3410 }
3411
3412 //----------------------------inline_unsafe_writeback0-------------------------
3413 // public native void Unsafe.writeback0(long address)
3414 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
3415 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
3416 return false;
3417 }
3418 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
3419 return false;
3420 }
3421 #ifndef PRODUCT
3422 assert(Matcher::has_match_rule(Op_CacheWB),
3423 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
3424 : "found match rule for CacheWBPostSync but not CacheWB"));
3425
3426 #endif
3427 null_check_receiver(); // null-check, then ignore
3428 Node *sync;
3429 if (is_pre) {
3430 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3431 } else {
3432 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
3433 }
3434 sync = _gvn.transform(sync);
3435 set_memory(sync, TypeRawPtr::BOTTOM);
3436 return true;
3437 }
3438
3439 //----------------------------inline_unsafe_allocate---------------------------
3440 // public native Object Unsafe.allocateInstance(Class<?> cls);
3441 bool LibraryCallKit::inline_unsafe_allocate() {
3442
3443 #if INCLUDE_JVMTI
3444 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
3445 return false;
3446 }
3447 #endif //INCLUDE_JVMTI
3448
3449 if (callee()->is_static()) return false; // caller must have the capability!
3450
3451 null_check_receiver(); // null-check, then ignore
3452 Node* cls = null_check(argument(1));
3453 if (stopped()) return true;
3454
3455 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
3456 kls = null_check(kls);
3457 if (stopped()) return true; // argument was like int.class
3458
3459 #if INCLUDE_JVMTI
3460 // Don't try to access new allocated obj in the intrinsic.
3461 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
3462 // Deoptimize and allocate in interpreter instead.
3463 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
3464 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
3465 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
3466 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
3467 {
3468 BuildCutout unless(this, tst, PROB_MAX);
3469 uncommon_trap(Deoptimization::Reason_intrinsic,
3470 Deoptimization::Action_make_not_entrant);
3471 }
3472 if (stopped()) {
3473 return true;
3474 }
3475 #endif //INCLUDE_JVMTI
3476
3477 Node* test = nullptr;
3478 if (LibraryCallKit::klass_needs_init_guard(kls)) {
3479 // Note: The argument might still be an illegal value like
3480 // Serializable.class or Object[].class. The runtime will handle it.
3481 // But we must make an explicit check for initialization.
3482 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
3483 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
3484 // can generate code to load it as unsigned byte.
3485 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
3486 Node* bits = intcon(InstanceKlass::fully_initialized);
3487 test = _gvn.transform(new SubINode(inst, bits));
3488 // The 'test' is non-zero if we need to take a slow path.
3489 }
3490 Node* obj = nullptr;
3491 const TypeInstKlassPtr* tkls = _gvn.type(kls)->isa_instklassptr();
3492 if (tkls != nullptr && tkls->instance_klass()->is_inlinetype()) {
3493 obj = InlineTypeNode::make_all_zero(_gvn, tkls->instance_klass()->as_inline_klass())->buffer(this);
3494 } else {
3495 obj = new_instance(kls, test);
3496 }
3497 set_result(obj);
3498 return true;
3499 }
3500
3501 //------------------------inline_native_time_funcs--------------
3502 // inline code for System.currentTimeMillis() and System.nanoTime()
3503 // these have the same type and signature
3504 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
3505 const TypeFunc* tf = OptoRuntime::void_long_Type();
3506 const TypePtr* no_memory_effects = nullptr;
3507 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
3508 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
3509 #ifdef ASSERT
3510 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
3511 assert(value_top == top(), "second value must be top");
3512 #endif
3513 set_result(value);
3514 return true;
3515 }
3516
3517
3518 #if INCLUDE_JVMTI
3519
3520 // When notifications are disabled then just update the VTMS transition bit and return.
3521 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
3522 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
3523 if (!DoJVMTIVirtualThreadTransitions) {
3524 return true;
3525 }
3526 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3527 IdealKit ideal(this);
3528
3529 Node* ONE = ideal.ConI(1);
3530 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3531 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3532 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3533
3534 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3535 sync_kit(ideal);
3536 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3537 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3538 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3539 ideal.sync_kit(this);
3540 } ideal.else_(); {
3541 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3542 Node* thread = ideal.thread();
3543 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3544 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3545
3546 sync_kit(ideal);
3547 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3548 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3549
3550 ideal.sync_kit(this);
3551 } ideal.end_if();
3552 final_sync(ideal);
3553
3554 return true;
3555 }
3556
3557 // Always update the is_disable_suspend bit.
3558 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3559 if (!DoJVMTIVirtualThreadTransitions) {
3560 return true;
3561 }
3562 IdealKit ideal(this);
3563
3564 {
3565 // unconditionally update the is_disable_suspend bit in current JavaThread
3566 Node* thread = ideal.thread();
3567 Node* arg = _gvn.transform(argument(0)); // argument for notification
3568 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3569 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3570
3571 sync_kit(ideal);
3572 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3573 ideal.sync_kit(this);
3574 }
3575 final_sync(ideal);
3576
3577 return true;
3578 }
3579
3580 #endif // INCLUDE_JVMTI
3581
3582 #ifdef JFR_HAVE_INTRINSICS
3583
3584 /**
3585 * if oop->klass != null
3586 * // normal class
3587 * epoch = _epoch_state ? 2 : 1
3588 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3589 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3590 * }
3591 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3592 * else
3593 * // primitive class
3594 * if oop->array_klass != null
3595 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3596 * else
3597 * id = LAST_TYPE_ID + 1 // void class path
3598 * if (!signaled)
3599 * signaled = true
3600 */
3601 bool LibraryCallKit::inline_native_classID() {
3602 Node* cls = argument(0);
3603
3604 IdealKit ideal(this);
3605 #define __ ideal.
3606 IdealVariable result(ideal); __ declarations_done();
3607 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3608 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3609 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3610
3611
3612 __ if_then(kls, BoolTest::ne, null()); {
3613 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3614 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3615
3616 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3617 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3618 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3619 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3620 mask = _gvn.transform(new OrLNode(mask, epoch));
3621 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3622
3623 float unlikely = PROB_UNLIKELY(0.999);
3624 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3625 sync_kit(ideal);
3626 make_runtime_call(RC_LEAF,
3627 OptoRuntime::class_id_load_barrier_Type(),
3628 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3629 "class id load barrier",
3630 TypePtr::BOTTOM,
3631 kls);
3632 ideal.sync_kit(this);
3633 } __ end_if();
3634
3635 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3636 } __ else_(); {
3637 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3638 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3639 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3640 __ if_then(array_kls, BoolTest::ne, null()); {
3641 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3642 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3643 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3644 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3645 } __ else_(); {
3646 // void class case
3647 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3648 } __ end_if();
3649
3650 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3651 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3652 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3653 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3654 } __ end_if();
3655 } __ end_if();
3656
3657 final_sync(ideal);
3658 set_result(ideal.value(result));
3659 #undef __
3660 return true;
3661 }
3662
3663 //------------------------inline_native_jvm_commit------------------
3664 bool LibraryCallKit::inline_native_jvm_commit() {
3665 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3666
3667 // Save input memory and i_o state.
3668 Node* input_memory_state = reset_memory();
3669 set_all_memory(input_memory_state);
3670 Node* input_io_state = i_o();
3671
3672 // TLS.
3673 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3674 // Jfr java buffer.
3675 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3676 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3677 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3678
3679 // Load the current value of the notified field in the JfrThreadLocal.
3680 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3681 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3682
3683 // Test for notification.
3684 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3685 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3686 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3687
3688 // True branch, is notified.
3689 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3690 set_control(is_notified);
3691
3692 // Reset notified state.
3693 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3694 Node* notified_reset_memory = reset_memory();
3695
3696 // 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.
3697 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3698 // Convert the machine-word to a long.
3699 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3700
3701 // False branch, not notified.
3702 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3703 set_control(not_notified);
3704 set_all_memory(input_memory_state);
3705
3706 // Arg is the next position as a long.
3707 Node* arg = argument(0);
3708 // Convert long to machine-word.
3709 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3710
3711 // Store the next_position to the underlying jfr java buffer.
3712 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3713
3714 Node* commit_memory = reset_memory();
3715 set_all_memory(commit_memory);
3716
3717 // 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.
3718 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3719 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3720 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3721
3722 // And flags with lease constant.
3723 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3724
3725 // Branch on lease to conditionalize returning the leased java buffer.
3726 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3727 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3728 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3729
3730 // False branch, not a lease.
3731 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3732
3733 // True branch, is lease.
3734 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3735 set_control(is_lease);
3736
3737 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3738 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3739 OptoRuntime::void_void_Type(),
3740 SharedRuntime::jfr_return_lease(),
3741 "return_lease", TypePtr::BOTTOM);
3742 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3743
3744 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3745 record_for_igvn(lease_compare_rgn);
3746 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3747 record_for_igvn(lease_compare_mem);
3748 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3749 record_for_igvn(lease_compare_io);
3750 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3751 record_for_igvn(lease_result_value);
3752
3753 // Update control and phi nodes.
3754 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3755 lease_compare_rgn->init_req(_false_path, not_lease);
3756
3757 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3758 lease_compare_mem->init_req(_false_path, commit_memory);
3759
3760 lease_compare_io->init_req(_true_path, i_o());
3761 lease_compare_io->init_req(_false_path, input_io_state);
3762
3763 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3764 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3765
3766 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3767 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3768 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3769 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3770
3771 // Update control and phi nodes.
3772 result_rgn->init_req(_true_path, is_notified);
3773 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3774
3775 result_mem->init_req(_true_path, notified_reset_memory);
3776 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3777
3778 result_io->init_req(_true_path, input_io_state);
3779 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3780
3781 result_value->init_req(_true_path, current_pos);
3782 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3783
3784 // Set output state.
3785 set_control(_gvn.transform(result_rgn));
3786 set_all_memory(_gvn.transform(result_mem));
3787 set_i_o(_gvn.transform(result_io));
3788 set_result(result_rgn, result_value);
3789 return true;
3790 }
3791
3792 /*
3793 * The intrinsic is a model of this pseudo-code:
3794 *
3795 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3796 * jobject h_event_writer = tl->java_event_writer();
3797 * if (h_event_writer == nullptr) {
3798 * return nullptr;
3799 * }
3800 * oop threadObj = Thread::threadObj();
3801 * oop vthread = java_lang_Thread::vthread(threadObj);
3802 * traceid tid;
3803 * bool pinVirtualThread;
3804 * bool excluded;
3805 * if (vthread != threadObj) { // i.e. current thread is virtual
3806 * tid = java_lang_Thread::tid(vthread);
3807 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3808 * pinVirtualThread = VMContinuations;
3809 * excluded = vthread_epoch_raw & excluded_mask;
3810 * if (!excluded) {
3811 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3812 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3813 * if (vthread_epoch != current_epoch) {
3814 * write_checkpoint();
3815 * }
3816 * }
3817 * } else {
3818 * tid = java_lang_Thread::tid(threadObj);
3819 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3820 * pinVirtualThread = false;
3821 * excluded = thread_epoch_raw & excluded_mask;
3822 * }
3823 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3824 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3825 * if (tid_in_event_writer != tid) {
3826 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3827 * setField(event_writer, "excluded", excluded);
3828 * setField(event_writer, "threadID", tid);
3829 * }
3830 * return event_writer
3831 */
3832 bool LibraryCallKit::inline_native_getEventWriter() {
3833 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3834
3835 // Save input memory and i_o state.
3836 Node* input_memory_state = reset_memory();
3837 set_all_memory(input_memory_state);
3838 Node* input_io_state = i_o();
3839
3840 // The most significant bit of the u2 is used to denote thread exclusion
3841 Node* excluded_shift = _gvn.intcon(15);
3842 Node* excluded_mask = _gvn.intcon(1 << 15);
3843 // The epoch generation is the range [1-32767]
3844 Node* epoch_mask = _gvn.intcon(32767);
3845
3846 // TLS
3847 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3848
3849 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3850 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3851
3852 // Load the eventwriter jobject handle.
3853 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3854
3855 // Null check the jobject handle.
3856 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3857 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3858 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3859
3860 // False path, jobj is null.
3861 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3862
3863 // True path, jobj is not null.
3864 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3865
3866 set_control(jobj_is_not_null);
3867
3868 // Load the threadObj for the CarrierThread.
3869 Node* threadObj = generate_current_thread(tls_ptr);
3870
3871 // Load the vthread.
3872 Node* vthread = generate_virtual_thread(tls_ptr);
3873
3874 // If vthread != threadObj, this is a virtual thread.
3875 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3876 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3877 IfNode* iff_vthread_not_equal_threadObj =
3878 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3879
3880 // False branch, fallback to threadObj.
3881 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3882 set_control(vthread_equal_threadObj);
3883
3884 // Load the tid field from the vthread object.
3885 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3886
3887 // Load the raw epoch value from the threadObj.
3888 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3889 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3890 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3891 TypeInt::CHAR, T_CHAR,
3892 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3893
3894 // Mask off the excluded information from the epoch.
3895 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3896
3897 // True branch, this is a virtual thread.
3898 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3899 set_control(vthread_not_equal_threadObj);
3900
3901 // Load the tid field from the vthread object.
3902 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3903
3904 // Continuation support determines if a virtual thread should be pinned.
3905 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3906 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3907
3908 // Load the raw epoch value from the vthread.
3909 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3910 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3911 TypeInt::CHAR, T_CHAR,
3912 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3913
3914 // Mask off the excluded information from the epoch.
3915 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3916
3917 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3918 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3919 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3920 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3921
3922 // False branch, vthread is excluded, no need to write epoch info.
3923 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3924
3925 // True branch, vthread is included, update epoch info.
3926 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3927 set_control(included);
3928
3929 // Get epoch value.
3930 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3931
3932 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3933 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3934 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3935
3936 // Compare the epoch in the vthread to the current epoch generation.
3937 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3938 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3939 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3940
3941 // False path, epoch is equal, checkpoint information is valid.
3942 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3943
3944 // True path, epoch is not equal, write a checkpoint for the vthread.
3945 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3946
3947 set_control(epoch_is_not_equal);
3948
3949 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3950 // The call also updates the native thread local thread id and the vthread with the current epoch.
3951 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3952 OptoRuntime::jfr_write_checkpoint_Type(),
3953 SharedRuntime::jfr_write_checkpoint(),
3954 "write_checkpoint", TypePtr::BOTTOM);
3955 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3956
3957 // vthread epoch != current epoch
3958 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3959 record_for_igvn(epoch_compare_rgn);
3960 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3961 record_for_igvn(epoch_compare_mem);
3962 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3963 record_for_igvn(epoch_compare_io);
3964
3965 // Update control and phi nodes.
3966 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3967 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3968 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3969 epoch_compare_mem->init_req(_false_path, input_memory_state);
3970 epoch_compare_io->init_req(_true_path, i_o());
3971 epoch_compare_io->init_req(_false_path, input_io_state);
3972
3973 // excluded != true
3974 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3975 record_for_igvn(exclude_compare_rgn);
3976 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3977 record_for_igvn(exclude_compare_mem);
3978 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3979 record_for_igvn(exclude_compare_io);
3980
3981 // Update control and phi nodes.
3982 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3983 exclude_compare_rgn->init_req(_false_path, excluded);
3984 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3985 exclude_compare_mem->init_req(_false_path, input_memory_state);
3986 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3987 exclude_compare_io->init_req(_false_path, input_io_state);
3988
3989 // vthread != threadObj
3990 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3991 record_for_igvn(vthread_compare_rgn);
3992 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3993 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3994 record_for_igvn(vthread_compare_io);
3995 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3996 record_for_igvn(tid);
3997 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3998 record_for_igvn(exclusion);
3999 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
4000 record_for_igvn(pinVirtualThread);
4001
4002 // Update control and phi nodes.
4003 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
4004 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
4005 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
4006 vthread_compare_mem->init_req(_false_path, input_memory_state);
4007 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
4008 vthread_compare_io->init_req(_false_path, input_io_state);
4009 tid->init_req(_true_path, _gvn.transform(vthread_tid));
4010 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
4011 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4012 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
4013 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
4014 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
4015
4016 // Update branch state.
4017 set_control(_gvn.transform(vthread_compare_rgn));
4018 set_all_memory(_gvn.transform(vthread_compare_mem));
4019 set_i_o(_gvn.transform(vthread_compare_io));
4020
4021 // Load the event writer oop by dereferencing the jobject handle.
4022 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
4023 assert(klass_EventWriter->is_loaded(), "invariant");
4024 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
4025 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
4026 const TypeOopPtr* const xtype = aklass->as_instance_type();
4027 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
4028 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
4029
4030 // Load the current thread id from the event writer object.
4031 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
4032 // Get the field offset to, conditionally, store an updated tid value later.
4033 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
4034 // Get the field offset to, conditionally, store an updated exclusion value later.
4035 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
4036 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
4037 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
4038
4039 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
4040 record_for_igvn(event_writer_tid_compare_rgn);
4041 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4042 record_for_igvn(event_writer_tid_compare_mem);
4043 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
4044 record_for_igvn(event_writer_tid_compare_io);
4045
4046 // Compare the current tid from the thread object to what is currently stored in the event writer object.
4047 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
4048 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
4049 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
4050
4051 // False path, tids are the same.
4052 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
4053
4054 // True path, tid is not equal, need to update the tid in the event writer.
4055 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
4056 record_for_igvn(tid_is_not_equal);
4057
4058 // Store the pin state to the event writer.
4059 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
4060
4061 // Store the exclusion state to the event writer.
4062 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
4063 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
4064
4065 // Store the tid to the event writer.
4066 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
4067
4068 // Update control and phi nodes.
4069 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
4070 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
4071 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4072 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
4073 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
4074 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
4075
4076 // Result of top level CFG, Memory, IO and Value.
4077 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4078 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4079 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
4080 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
4081
4082 // Result control.
4083 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
4084 result_rgn->init_req(_false_path, jobj_is_null);
4085
4086 // Result memory.
4087 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
4088 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4089
4090 // Result IO.
4091 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
4092 result_io->init_req(_false_path, _gvn.transform(input_io_state));
4093
4094 // Result value.
4095 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
4096 result_value->init_req(_false_path, null()); // return null
4097
4098 // Set output state.
4099 set_control(_gvn.transform(result_rgn));
4100 set_all_memory(_gvn.transform(result_mem));
4101 set_i_o(_gvn.transform(result_io));
4102 set_result(result_rgn, result_value);
4103 return true;
4104 }
4105
4106 /*
4107 * The intrinsic is a model of this pseudo-code:
4108 *
4109 * JfrThreadLocal* const tl = thread->jfr_thread_local();
4110 * if (carrierThread != thread) { // is virtual thread
4111 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
4112 * bool excluded = vthread_epoch_raw & excluded_mask;
4113 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
4114 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
4115 * if (!excluded) {
4116 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
4117 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
4118 * }
4119 * AtomicAccess::release_store(&tl->_vthread, true);
4120 * return;
4121 * }
4122 * AtomicAccess::release_store(&tl->_vthread, false);
4123 */
4124 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
4125 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4126
4127 Node* input_memory_state = reset_memory();
4128 set_all_memory(input_memory_state);
4129
4130 // The most significant bit of the u2 is used to denote thread exclusion
4131 Node* excluded_mask = _gvn.intcon(1 << 15);
4132 // The epoch generation is the range [1-32767]
4133 Node* epoch_mask = _gvn.intcon(32767);
4134
4135 Node* const carrierThread = generate_current_thread(jt);
4136 // If thread != carrierThread, this is a virtual thread.
4137 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
4138 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
4139 IfNode* iff_thread_not_equal_carrierThread =
4140 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
4141
4142 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
4143
4144 // False branch, is carrierThread.
4145 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
4146 // Store release
4147 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
4148
4149 set_all_memory(input_memory_state);
4150
4151 // True branch, is virtual thread.
4152 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
4153 set_control(thread_not_equal_carrierThread);
4154
4155 // Load the raw epoch value from the vthread.
4156 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
4157 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
4158 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
4159
4160 // Mask off the excluded information from the epoch.
4161 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
4162
4163 // Load the tid field from the thread.
4164 Node* tid = load_field_from_object(thread, "tid", "J");
4165
4166 // Store the vthread tid to the jfr thread local.
4167 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
4168 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
4169
4170 // Branch is_excluded to conditionalize updating the epoch .
4171 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
4172 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
4173 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
4174
4175 // True branch, vthread is excluded, no need to write epoch info.
4176 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
4177 set_control(excluded);
4178 Node* vthread_is_excluded = _gvn.intcon(1);
4179
4180 // False branch, vthread is included, update epoch info.
4181 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
4182 set_control(included);
4183 Node* vthread_is_included = _gvn.intcon(0);
4184
4185 // Get epoch value.
4186 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
4187
4188 // Store the vthread epoch to the jfr thread local.
4189 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
4190 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
4191
4192 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
4193 record_for_igvn(excluded_rgn);
4194 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
4195 record_for_igvn(excluded_mem);
4196 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
4197 record_for_igvn(exclusion);
4198
4199 // Merge the excluded control and memory.
4200 excluded_rgn->init_req(_true_path, excluded);
4201 excluded_rgn->init_req(_false_path, included);
4202 excluded_mem->init_req(_true_path, tid_memory);
4203 excluded_mem->init_req(_false_path, included_memory);
4204 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
4205 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
4206
4207 // Set intermediate state.
4208 set_control(_gvn.transform(excluded_rgn));
4209 set_all_memory(excluded_mem);
4210
4211 // Store the vthread exclusion state to the jfr thread local.
4212 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
4213 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
4214
4215 // Store release
4216 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
4217
4218 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
4219 record_for_igvn(thread_compare_rgn);
4220 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
4221 record_for_igvn(thread_compare_mem);
4222 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
4223 record_for_igvn(vthread);
4224
4225 // Merge the thread_compare control and memory.
4226 thread_compare_rgn->init_req(_true_path, control());
4227 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
4228 thread_compare_mem->init_req(_true_path, vthread_true_memory);
4229 thread_compare_mem->init_req(_false_path, vthread_false_memory);
4230
4231 // Set output state.
4232 set_control(_gvn.transform(thread_compare_rgn));
4233 set_all_memory(_gvn.transform(thread_compare_mem));
4234 }
4235
4236 #endif // JFR_HAVE_INTRINSICS
4237
4238 //------------------------inline_native_currentCarrierThread------------------
4239 bool LibraryCallKit::inline_native_currentCarrierThread() {
4240 Node* junk = nullptr;
4241 set_result(generate_current_thread(junk));
4242 return true;
4243 }
4244
4245 //------------------------inline_native_currentThread------------------
4246 bool LibraryCallKit::inline_native_currentThread() {
4247 Node* junk = nullptr;
4248 set_result(generate_virtual_thread(junk));
4249 return true;
4250 }
4251
4252 //------------------------inline_native_setVthread------------------
4253 bool LibraryCallKit::inline_native_setCurrentThread() {
4254 assert(C->method()->changes_current_thread(),
4255 "method changes current Thread but is not annotated ChangesCurrentThread");
4256 Node* arr = argument(1);
4257 Node* thread = _gvn.transform(new ThreadLocalNode());
4258 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
4259 Node* thread_obj_handle
4260 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
4261 thread_obj_handle = _gvn.transform(thread_obj_handle);
4262 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
4263 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
4264
4265 // Change the _monitor_owner_id of the JavaThread
4266 Node* tid = load_field_from_object(arr, "tid", "J");
4267 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
4268 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
4269
4270 JFR_ONLY(extend_setCurrentThread(thread, arr);)
4271 return true;
4272 }
4273
4274 const Type* LibraryCallKit::scopedValueCache_type() {
4275 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
4276 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
4277 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS, /* stable= */ false, /* flat= */ false, /* not_flat= */ true, /* not_null_free= */ true);
4278
4279 // Because we create the scopedValue cache lazily we have to make the
4280 // type of the result BotPTR.
4281 bool xk = etype->klass_is_exact();
4282 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, TypeAryPtr::Offset(0));
4283 return objects_type;
4284 }
4285
4286 Node* LibraryCallKit::scopedValueCache_helper() {
4287 Node* thread = _gvn.transform(new ThreadLocalNode());
4288 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
4289 // We cannot use immutable_memory() because we might flip onto a
4290 // different carrier thread, at which point we'll need to use that
4291 // carrier thread's cache.
4292 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
4293 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
4294 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
4295 }
4296
4297 //------------------------inline_native_scopedValueCache------------------
4298 bool LibraryCallKit::inline_native_scopedValueCache() {
4299 Node* cache_obj_handle = scopedValueCache_helper();
4300 const Type* objects_type = scopedValueCache_type();
4301 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
4302
4303 return true;
4304 }
4305
4306 //------------------------inline_native_setScopedValueCache------------------
4307 bool LibraryCallKit::inline_native_setScopedValueCache() {
4308 Node* arr = argument(0);
4309 Node* cache_obj_handle = scopedValueCache_helper();
4310 const Type* objects_type = scopedValueCache_type();
4311
4312 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
4313 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
4314
4315 return true;
4316 }
4317
4318 //------------------------inline_native_Continuation_pin and unpin-----------
4319
4320 // Shared implementation routine for both pin and unpin.
4321 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
4322 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
4323
4324 // Save input memory.
4325 Node* input_memory_state = reset_memory();
4326 set_all_memory(input_memory_state);
4327
4328 // TLS
4329 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
4330 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
4331 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
4332
4333 // Null check the last continuation object.
4334 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
4335 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
4336 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
4337
4338 // False path, last continuation is null.
4339 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
4340
4341 // True path, last continuation is not null.
4342 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
4343
4344 set_control(continuation_is_not_null);
4345
4346 // Load the pin count from the last continuation.
4347 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
4348 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
4349
4350 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
4351 Node* pin_count_rhs;
4352 if (unpin) {
4353 pin_count_rhs = _gvn.intcon(0);
4354 } else {
4355 pin_count_rhs = _gvn.intcon(UINT32_MAX);
4356 }
4357 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
4358 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
4359 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
4360
4361 // True branch, pin count over/underflow.
4362 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
4363 {
4364 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
4365 // which will throw IllegalStateException for pin count over/underflow.
4366 // No memory changed so far - we can use memory create by reset_memory()
4367 // at the beginning of this intrinsic. No need to call reset_memory() again.
4368 PreserveJVMState pjvms(this);
4369 set_control(pin_count_over_underflow);
4370 uncommon_trap(Deoptimization::Reason_intrinsic,
4371 Deoptimization::Action_none);
4372 assert(stopped(), "invariant");
4373 }
4374
4375 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
4376 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
4377 set_control(valid_pin_count);
4378
4379 Node* next_pin_count;
4380 if (unpin) {
4381 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
4382 } else {
4383 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
4384 }
4385
4386 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
4387
4388 // Result of top level CFG and Memory.
4389 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
4390 record_for_igvn(result_rgn);
4391 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
4392 record_for_igvn(result_mem);
4393
4394 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
4395 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
4396 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
4397 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
4398
4399 // Set output state.
4400 set_control(_gvn.transform(result_rgn));
4401 set_all_memory(_gvn.transform(result_mem));
4402
4403 return true;
4404 }
4405
4406 //-----------------------load_klass_from_mirror_common-------------------------
4407 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
4408 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
4409 // and branch to the given path on the region.
4410 // If never_see_null, take an uncommon trap on null, so we can optimistically
4411 // compile for the non-null case.
4412 // If the region is null, force never_see_null = true.
4413 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
4414 bool never_see_null,
4415 RegionNode* region,
4416 int null_path,
4417 int offset) {
4418 if (region == nullptr) never_see_null = true;
4419 Node* p = basic_plus_adr(mirror, offset);
4420 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4421 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
4422 Node* null_ctl = top();
4423 kls = null_check_oop(kls, &null_ctl, never_see_null);
4424 if (region != nullptr) {
4425 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
4426 region->init_req(null_path, null_ctl);
4427 } else {
4428 assert(null_ctl == top(), "no loose ends");
4429 }
4430 return kls;
4431 }
4432
4433 //--------------------(inline_native_Class_query helpers)---------------------
4434 // Use this for JVM_ACC_INTERFACE.
4435 // Fall through if (mods & mask) == bits, take the guard otherwise.
4436 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
4437 ByteSize offset, const Type* type, BasicType bt) {
4438 // Branch around if the given klass has the given modifier bit set.
4439 // Like generate_guard, adds a new path onto the region.
4440 Node* modp = basic_plus_adr(kls, in_bytes(offset));
4441 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
4442 Node* mask = intcon(modifier_mask);
4443 Node* bits = intcon(modifier_bits);
4444 Node* mbit = _gvn.transform(new AndINode(mods, mask));
4445 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
4446 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
4447 return generate_fair_guard(bol, region);
4448 }
4449
4450 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
4451 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
4452 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
4453 }
4454
4455 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
4456 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
4457 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
4458 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
4459 }
4460
4461 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
4462 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
4463 }
4464
4465 //-------------------------inline_native_Class_query-------------------
4466 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
4467 const Type* return_type = TypeInt::BOOL;
4468 Node* prim_return_value = top(); // what happens if it's a primitive class?
4469 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4470 bool expect_prim = false; // most of these guys expect to work on refs
4471
4472 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
4473
4474 Node* mirror = argument(0);
4475 Node* obj = top();
4476
4477 switch (id) {
4478 case vmIntrinsics::_isInstance:
4479 // nothing is an instance of a primitive type
4480 prim_return_value = intcon(0);
4481 obj = argument(1);
4482 break;
4483 case vmIntrinsics::_isHidden:
4484 prim_return_value = intcon(0);
4485 break;
4486 case vmIntrinsics::_getSuperclass:
4487 prim_return_value = null();
4488 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4489 break;
4490 default:
4491 fatal_unexpected_iid(id);
4492 break;
4493 }
4494
4495 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4496 if (mirror_con == nullptr) return false; // cannot happen?
4497
4498 #ifndef PRODUCT
4499 if (C->print_intrinsics() || C->print_inlining()) {
4500 ciType* k = mirror_con->java_mirror_type();
4501 if (k) {
4502 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4503 k->print_name();
4504 tty->cr();
4505 }
4506 }
4507 #endif
4508
4509 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4510 RegionNode* region = new RegionNode(PATH_LIMIT);
4511 record_for_igvn(region);
4512 PhiNode* phi = new PhiNode(region, return_type);
4513
4514 // The mirror will never be null of Reflection.getClassAccessFlags, however
4515 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4516 // if it is. See bug 4774291.
4517
4518 // For Reflection.getClassAccessFlags(), the null check occurs in
4519 // the wrong place; see inline_unsafe_access(), above, for a similar
4520 // situation.
4521 mirror = null_check(mirror);
4522 // If mirror or obj is dead, only null-path is taken.
4523 if (stopped()) return true;
4524
4525 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4526
4527 // Now load the mirror's klass metaobject, and null-check it.
4528 // Side-effects region with the control path if the klass is null.
4529 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4530 // If kls is null, we have a primitive mirror.
4531 phi->init_req(_prim_path, prim_return_value);
4532 if (stopped()) { set_result(region, phi); return true; }
4533 bool safe_for_replace = (region->in(_prim_path) == top());
4534
4535 Node* p; // handy temp
4536 Node* null_ctl;
4537
4538 // Now that we have the non-null klass, we can perform the real query.
4539 // For constant classes, the query will constant-fold in LoadNode::Value.
4540 Node* query_value = top();
4541 switch (id) {
4542 case vmIntrinsics::_isInstance:
4543 // nothing is an instance of a primitive type
4544 query_value = gen_instanceof(obj, kls, safe_for_replace);
4545 break;
4546
4547 case vmIntrinsics::_isHidden:
4548 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4549 if (generate_hidden_class_guard(kls, region) != nullptr)
4550 // A guard was added. If the guard is taken, it was an hidden class.
4551 phi->add_req(intcon(1));
4552 // If we fall through, it's a plain class.
4553 query_value = intcon(0);
4554 break;
4555
4556
4557 case vmIntrinsics::_getSuperclass:
4558 // The rules here are somewhat unfortunate, but we can still do better
4559 // with random logic than with a JNI call.
4560 // Interfaces store null or Object as _super, but must report null.
4561 // Arrays store an intermediate super as _super, but must report Object.
4562 // Other types can report the actual _super.
4563 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4564 if (generate_interface_guard(kls, region) != nullptr)
4565 // A guard was added. If the guard is taken, it was an interface.
4566 phi->add_req(null());
4567 if (generate_array_guard(kls, region) != nullptr)
4568 // A guard was added. If the guard is taken, it was an array.
4569 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4570 // If we fall through, it's a plain class. Get its _super.
4571 if (!stopped()) {
4572 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4573 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4574 null_ctl = top();
4575 kls = null_check_oop(kls, &null_ctl);
4576 if (null_ctl != top()) {
4577 // If the guard is taken, Object.superClass is null (both klass and mirror).
4578 region->add_req(null_ctl);
4579 phi ->add_req(null());
4580 }
4581 if (!stopped()) {
4582 query_value = load_mirror_from_klass(kls);
4583 }
4584 }
4585 break;
4586
4587 default:
4588 fatal_unexpected_iid(id);
4589 break;
4590 }
4591
4592 // Fall-through is the normal case of a query to a real class.
4593 phi->init_req(1, query_value);
4594 region->init_req(1, control());
4595
4596 C->set_has_split_ifs(true); // Has chance for split-if optimization
4597 set_result(region, phi);
4598 return true;
4599 }
4600
4601
4602 //-------------------------inline_Class_cast-------------------
4603 bool LibraryCallKit::inline_Class_cast() {
4604 Node* mirror = argument(0); // Class
4605 Node* obj = argument(1);
4606 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4607 if (mirror_con == nullptr) {
4608 return false; // dead path (mirror->is_top()).
4609 }
4610 if (obj == nullptr || obj->is_top()) {
4611 return false; // dead path
4612 }
4613 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4614
4615 // First, see if Class.cast() can be folded statically.
4616 // java_mirror_type() returns non-null for compile-time Class constants.
4617 ciType* tm = mirror_con->java_mirror_type();
4618 if (tm != nullptr && tm->is_klass() &&
4619 tp != nullptr) {
4620 if (!tp->is_loaded()) {
4621 // Don't use intrinsic when class is not loaded.
4622 return false;
4623 } else {
4624 const TypeKlassPtr* tklass = TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces);
4625 int static_res = C->static_subtype_check(tklass, tp->as_klass_type());
4626 if (static_res == Compile::SSC_always_true) {
4627 // isInstance() is true - fold the code.
4628 set_result(obj);
4629 return true;
4630 } else if (static_res == Compile::SSC_always_false) {
4631 // Don't use intrinsic, have to throw ClassCastException.
4632 // If the reference is null, the non-intrinsic bytecode will
4633 // be optimized appropriately.
4634 return false;
4635 }
4636 }
4637 }
4638
4639 // Bailout intrinsic and do normal inlining if exception path is frequent.
4640 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4641 return false;
4642 }
4643
4644 // Generate dynamic checks.
4645 // Class.cast() is java implementation of _checkcast bytecode.
4646 // Do checkcast (Parse::do_checkcast()) optimizations here.
4647
4648 mirror = null_check(mirror);
4649 // If mirror is dead, only null-path is taken.
4650 if (stopped()) {
4651 return true;
4652 }
4653
4654 // Not-subtype or the mirror's klass ptr is nullptr (in case it is a primitive).
4655 enum { _bad_type_path = 1, _prim_path = 2, _npe_path = 3, PATH_LIMIT };
4656 RegionNode* region = new RegionNode(PATH_LIMIT);
4657 record_for_igvn(region);
4658
4659 // Now load the mirror's klass metaobject, and null-check it.
4660 // If kls is null, we have a primitive mirror and
4661 // nothing is an instance of a primitive type.
4662 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4663
4664 Node* res = top();
4665 Node* io = i_o();
4666 Node* mem = merged_memory();
4667 if (!stopped()) {
4668
4669 Node* bad_type_ctrl = top();
4670 // Do checkcast optimizations.
4671 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4672 region->init_req(_bad_type_path, bad_type_ctrl);
4673 }
4674 if (region->in(_prim_path) != top() ||
4675 region->in(_bad_type_path) != top() ||
4676 region->in(_npe_path) != top()) {
4677 // Let Interpreter throw ClassCastException.
4678 PreserveJVMState pjvms(this);
4679 set_control(_gvn.transform(region));
4680 // Set IO and memory because gen_checkcast may override them when buffering inline types
4681 set_i_o(io);
4682 set_all_memory(mem);
4683 uncommon_trap(Deoptimization::Reason_intrinsic,
4684 Deoptimization::Action_maybe_recompile);
4685 }
4686 if (!stopped()) {
4687 set_result(res);
4688 }
4689 return true;
4690 }
4691
4692
4693 //--------------------------inline_native_subtype_check------------------------
4694 // This intrinsic takes the JNI calls out of the heart of
4695 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4696 bool LibraryCallKit::inline_native_subtype_check() {
4697 // Pull both arguments off the stack.
4698 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4699 args[0] = argument(0);
4700 args[1] = argument(1);
4701 Node* klasses[2]; // corresponding Klasses: superk, subk
4702 klasses[0] = klasses[1] = top();
4703
4704 enum {
4705 // A full decision tree on {superc is prim, subc is prim}:
4706 _prim_0_path = 1, // {P,N} => false
4707 // {P,P} & superc!=subc => false
4708 _prim_same_path, // {P,P} & superc==subc => true
4709 _prim_1_path, // {N,P} => false
4710 _ref_subtype_path, // {N,N} & subtype check wins => true
4711 _both_ref_path, // {N,N} & subtype check loses => false
4712 PATH_LIMIT
4713 };
4714
4715 RegionNode* region = new RegionNode(PATH_LIMIT);
4716 RegionNode* prim_region = new RegionNode(2);
4717 Node* phi = new PhiNode(region, TypeInt::BOOL);
4718 record_for_igvn(region);
4719 record_for_igvn(prim_region);
4720
4721 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4722 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4723 int class_klass_offset = java_lang_Class::klass_offset();
4724
4725 // First null-check both mirrors and load each mirror's klass metaobject.
4726 int which_arg;
4727 for (which_arg = 0; which_arg <= 1; which_arg++) {
4728 Node* arg = args[which_arg];
4729 arg = null_check(arg);
4730 if (stopped()) break;
4731 args[which_arg] = arg;
4732
4733 Node* p = basic_plus_adr(arg, class_klass_offset);
4734 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4735 klasses[which_arg] = _gvn.transform(kls);
4736 }
4737
4738 // Having loaded both klasses, test each for null.
4739 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4740 for (which_arg = 0; which_arg <= 1; which_arg++) {
4741 Node* kls = klasses[which_arg];
4742 Node* null_ctl = top();
4743 kls = null_check_oop(kls, &null_ctl, never_see_null);
4744 if (which_arg == 0) {
4745 prim_region->init_req(1, null_ctl);
4746 } else {
4747 region->init_req(_prim_1_path, null_ctl);
4748 }
4749 if (stopped()) break;
4750 klasses[which_arg] = kls;
4751 }
4752
4753 if (!stopped()) {
4754 // now we have two reference types, in klasses[0..1]
4755 Node* subk = klasses[1]; // the argument to isAssignableFrom
4756 Node* superk = klasses[0]; // the receiver
4757 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4758 region->set_req(_ref_subtype_path, control());
4759 }
4760
4761 // If both operands are primitive (both klasses null), then
4762 // we must return true when they are identical primitives.
4763 // It is convenient to test this after the first null klass check.
4764 // This path is also used if superc is a value mirror.
4765 set_control(_gvn.transform(prim_region));
4766 if (!stopped()) {
4767 // Since superc is primitive, make a guard for the superc==subc case.
4768 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4769 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4770 generate_fair_guard(bol_eq, region);
4771 if (region->req() == PATH_LIMIT+1) {
4772 // A guard was added. If the added guard is taken, superc==subc.
4773 region->swap_edges(PATH_LIMIT, _prim_same_path);
4774 region->del_req(PATH_LIMIT);
4775 }
4776 region->set_req(_prim_0_path, control()); // Not equal after all.
4777 }
4778
4779 // these are the only paths that produce 'true':
4780 phi->set_req(_prim_same_path, intcon(1));
4781 phi->set_req(_ref_subtype_path, intcon(1));
4782
4783 // pull together the cases:
4784 assert(region->req() == PATH_LIMIT, "sane region");
4785 for (uint i = 1; i < region->req(); i++) {
4786 Node* ctl = region->in(i);
4787 if (ctl == nullptr || ctl == top()) {
4788 region->set_req(i, top());
4789 phi ->set_req(i, top());
4790 } else if (phi->in(i) == nullptr) {
4791 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4792 }
4793 }
4794
4795 set_control(_gvn.transform(region));
4796 set_result(_gvn.transform(phi));
4797 return true;
4798 }
4799
4800 //---------------------generate_array_guard_common------------------------
4801 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region, ArrayKind kind, Node** obj) {
4802
4803 if (stopped()) {
4804 return nullptr;
4805 }
4806
4807 // Like generate_guard, adds a new path onto the region.
4808 jint layout_con = 0;
4809 Node* layout_val = get_layout_helper(kls, layout_con);
4810 if (layout_val == nullptr) {
4811 bool query = 0;
4812 switch(kind) {
4813 case RefArray: query = Klass::layout_helper_is_refArray(layout_con); break;
4814 case NonRefArray: query = !Klass::layout_helper_is_refArray(layout_con); break;
4815 case TypeArray: query = Klass::layout_helper_is_typeArray(layout_con); break;
4816 case AnyArray: query = Klass::layout_helper_is_array(layout_con); break;
4817 case NonArray: query = !Klass::layout_helper_is_array(layout_con); break;
4818 default:
4819 ShouldNotReachHere();
4820 }
4821 if (!query) {
4822 return nullptr; // never a branch
4823 } else { // always a branch
4824 Node* always_branch = control();
4825 if (region != nullptr)
4826 region->add_req(always_branch);
4827 set_control(top());
4828 return always_branch;
4829 }
4830 }
4831 unsigned int value = 0;
4832 BoolTest::mask btest = BoolTest::illegal;
4833 switch(kind) {
4834 case RefArray:
4835 case NonRefArray: {
4836 value = Klass::_lh_array_tag_ref_value;
4837 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4838 btest = (kind == RefArray) ? BoolTest::eq : BoolTest::ne;
4839 break;
4840 }
4841 case TypeArray: {
4842 value = Klass::_lh_array_tag_type_value;
4843 layout_val = _gvn.transform(new RShiftINode(layout_val, intcon(Klass::_lh_array_tag_shift)));
4844 btest = BoolTest::eq;
4845 break;
4846 }
4847 case AnyArray: value = Klass::_lh_neutral_value; btest = BoolTest::lt; break;
4848 case NonArray: value = Klass::_lh_neutral_value; btest = BoolTest::gt; break;
4849 default:
4850 ShouldNotReachHere();
4851 }
4852 // Now test the correct condition.
4853 jint nval = (jint)value;
4854 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4855 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4856 Node* ctrl = generate_fair_guard(bol, region);
4857 Node* is_array_ctrl = kind == NonArray ? control() : ctrl;
4858 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4859 // Keep track of the fact that 'obj' is an array to prevent
4860 // array specific accesses from floating above the guard.
4861 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4862 }
4863 return ctrl;
4864 }
4865
4866 // public static native Object[] ValueClass::newNullRestrictedAtomicArray(Class<?> componentType, int length, Object initVal);
4867 // public static native Object[] ValueClass::newNullRestrictedNonAtomicArray(Class<?> componentType, int length, Object initVal);
4868 // public static native Object[] ValueClass::newNullableAtomicArray(Class<?> componentType, int length);
4869 bool LibraryCallKit::inline_newArray(bool null_free, bool atomic) {
4870 assert(null_free || atomic, "nullable implies atomic");
4871 Node* componentType = argument(0);
4872 Node* length = argument(1);
4873 Node* init_val = null_free ? argument(2) : nullptr;
4874
4875 const TypeInstPtr* tp = _gvn.type(componentType)->isa_instptr();
4876 if (tp != nullptr) {
4877 ciInstanceKlass* ik = tp->instance_klass();
4878 if (ik == C->env()->Class_klass()) {
4879 ciType* t = tp->java_mirror_type();
4880 if (t != nullptr && t->is_inlinetype()) {
4881
4882 ciArrayKlass* array_klass = ciArrayKlass::make(t, null_free, atomic, true);
4883 assert(array_klass->is_elem_null_free() == null_free, "inconsistency");
4884 assert(array_klass->is_elem_atomic() == atomic, "inconsistency");
4885
4886 // TOOD 8350865 ZGC needs card marks on initializing oop stores
4887 if (UseZGC && null_free && !array_klass->is_flat_array_klass()) {
4888 return false;
4889 }
4890
4891 if (array_klass->is_loaded() && array_klass->element_klass()->as_inline_klass()->is_initialized()) {
4892 const TypeAryKlassPtr* array_klass_type = TypeAryKlassPtr::make(array_klass, Type::trust_interfaces, true);
4893 if (null_free) {
4894 if (init_val->is_InlineType()) {
4895 if (array_klass_type->is_flat() && init_val->as_InlineType()->is_all_zero(&gvn(), /* flat */ true)) {
4896 // Zeroing is enough because the init value is the all-zero value
4897 init_val = nullptr;
4898 } else {
4899 init_val = init_val->as_InlineType()->buffer(this);
4900 }
4901 }
4902 // TODO 8350865 Should we add a check of the init_val type (maybe in debug only + halt)?
4903 }
4904 Node* obj = new_array(makecon(array_klass_type), length, 0, nullptr, false, init_val);
4905 const TypeAryPtr* arytype = gvn().type(obj)->is_aryptr();
4906 assert(arytype->is_null_free() == null_free, "inconsistency");
4907 assert(arytype->is_not_null_free() == !null_free, "inconsistency");
4908 assert(arytype->is_atomic() == atomic, "inconsistency");
4909 set_result(obj);
4910 return true;
4911 }
4912 }
4913 }
4914 }
4915 return false;
4916 }
4917
4918 // public static native boolean ValueClass::isFlatArray(Object array);
4919 // public static native boolean ValueClass::isNullRestrictedArray(Object array);
4920 // public static native boolean ValueClass::isAtomicArray(Object array);
4921 bool LibraryCallKit::inline_getArrayProperties(ArrayPropertiesCheck check) {
4922 Node* array = argument(0);
4923
4924 Node* bol;
4925 switch(check) {
4926 case IsFlat:
4927 // TODO 8350865 Use the object version here instead of loading the klass
4928 // The problem is that PhaseMacroExpand::expand_flatarraycheck_node can only handle some IR shapes and will fail, for example, if the bol is directly wired to a ReturnNode
4929 bol = flat_array_test(load_object_klass(array));
4930 break;
4931 case IsNullRestricted:
4932 bol = null_free_array_test(array);
4933 break;
4934 case IsAtomic:
4935 // TODO 8350865 Implement this. It's a bit more complicated, see conditions in JVM_IsAtomicArray
4936 // Enable TestIntrinsics::test87/88 once this is implemented
4937 // bol = null_free_atomic_array_test
4938 return false;
4939 default:
4940 ShouldNotReachHere();
4941 }
4942
4943 Node* res = gvn().transform(new CMoveINode(bol, intcon(0), intcon(1), TypeInt::BOOL));
4944 set_result(res);
4945 return true;
4946 }
4947
4948 // Load the default refined array klass from an ObjArrayKlass. This relies on the first entry in the
4949 // '_next_refined_array_klass' linked list being the default (see ObjArrayKlass::klass_with_properties).
4950 Node* LibraryCallKit::load_default_refined_array_klass(Node* klass_node, bool type_array_guard) {
4951 RegionNode* region = new RegionNode(2);
4952 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT_OR_NULL);
4953
4954 if (type_array_guard) {
4955 generate_typeArray_guard(klass_node, region);
4956 if (region->req() == 3) {
4957 phi->add_req(klass_node);
4958 }
4959 }
4960 Node* adr_refined_klass = basic_plus_adr(klass_node, in_bytes(ObjArrayKlass::next_refined_array_klass_offset()));
4961 Node* refined_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), adr_refined_klass, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4962
4963 // Can be null if not initialized yet, just deopt
4964 Node* null_ctl = top();
4965 refined_klass = null_check_oop(refined_klass, &null_ctl, /* never_see_null= */ true);
4966
4967 region->init_req(1, control());
4968 phi->init_req(1, refined_klass);
4969
4970 set_control(_gvn.transform(region));
4971 return _gvn.transform(phi);
4972 }
4973
4974 // Load the non-refined array klass from an ObjArrayKlass.
4975 Node* LibraryCallKit::load_non_refined_array_klass(Node* klass_node) {
4976 const TypeAryKlassPtr* ary_klass_ptr = _gvn.type(klass_node)->isa_aryklassptr();
4977 if (ary_klass_ptr != nullptr && ary_klass_ptr->klass_is_exact()) {
4978 return _gvn.makecon(ary_klass_ptr->cast_to_refined_array_klass_ptr(false));
4979 }
4980
4981 RegionNode* region = new RegionNode(2);
4982 Node* phi = new PhiNode(region, TypeInstKlassPtr::OBJECT);
4983
4984 generate_typeArray_guard(klass_node, region);
4985 if (region->req() == 3) {
4986 phi->add_req(klass_node);
4987 }
4988 Node* super_adr = basic_plus_adr(klass_node, in_bytes(Klass::super_offset()));
4989 Node* super_klass = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), super_adr, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT));
4990
4991 region->init_req(1, control());
4992 phi->init_req(1, super_klass);
4993
4994 set_control(_gvn.transform(region));
4995 return _gvn.transform(phi);
4996 }
4997
4998 //-----------------------inline_native_newArray--------------------------
4999 // private static native Object java.lang.reflect.Array.newArray(Class<?> componentType, int length);
5000 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
5001 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
5002 Node* mirror;
5003 Node* count_val;
5004 if (uninitialized) {
5005 null_check_receiver();
5006 mirror = argument(1);
5007 count_val = argument(2);
5008 } else {
5009 mirror = argument(0);
5010 count_val = argument(1);
5011 }
5012
5013 mirror = null_check(mirror);
5014 // If mirror or obj is dead, only null-path is taken.
5015 if (stopped()) return true;
5016
5017 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
5018 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5019 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5020 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5021 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5022
5023 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
5024 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
5025 result_reg, _slow_path);
5026 Node* normal_ctl = control();
5027 Node* no_array_ctl = result_reg->in(_slow_path);
5028
5029 // Generate code for the slow case. We make a call to newArray().
5030 set_control(no_array_ctl);
5031 if (!stopped()) {
5032 // Either the input type is void.class, or else the
5033 // array klass has not yet been cached. Either the
5034 // ensuing call will throw an exception, or else it
5035 // will cache the array klass for next time.
5036 PreserveJVMState pjvms(this);
5037 CallJavaNode* slow_call = nullptr;
5038 if (uninitialized) {
5039 // Generate optimized virtual call (holder class 'Unsafe' is final)
5040 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
5041 } else {
5042 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
5043 }
5044 Node* slow_result = set_results_for_java_call(slow_call);
5045 // this->control() comes from set_results_for_java_call
5046 result_reg->set_req(_slow_path, control());
5047 result_val->set_req(_slow_path, slow_result);
5048 result_io ->set_req(_slow_path, i_o());
5049 result_mem->set_req(_slow_path, reset_memory());
5050 }
5051
5052 set_control(normal_ctl);
5053 if (!stopped()) {
5054 // Normal case: The array type has been cached in the java.lang.Class.
5055 // The following call works fine even if the array type is polymorphic.
5056 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5057
5058 klass_node = load_default_refined_array_klass(klass_node);
5059
5060 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
5061 result_reg->init_req(_normal_path, control());
5062 result_val->init_req(_normal_path, obj);
5063 result_io ->init_req(_normal_path, i_o());
5064 result_mem->init_req(_normal_path, reset_memory());
5065
5066 if (uninitialized) {
5067 // Mark the allocation so that zeroing is skipped
5068 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
5069 alloc->maybe_set_complete(&_gvn);
5070 }
5071 }
5072
5073 // Return the combined state.
5074 set_i_o( _gvn.transform(result_io) );
5075 set_all_memory( _gvn.transform(result_mem));
5076
5077 C->set_has_split_ifs(true); // Has chance for split-if optimization
5078 set_result(result_reg, result_val);
5079 return true;
5080 }
5081
5082 //----------------------inline_native_getLength--------------------------
5083 // public static native int java.lang.reflect.Array.getLength(Object array);
5084 bool LibraryCallKit::inline_native_getLength() {
5085 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5086
5087 Node* array = null_check(argument(0));
5088 // If array is dead, only null-path is taken.
5089 if (stopped()) return true;
5090
5091 // Deoptimize if it is a non-array.
5092 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
5093
5094 if (non_array != nullptr) {
5095 PreserveJVMState pjvms(this);
5096 set_control(non_array);
5097 uncommon_trap(Deoptimization::Reason_intrinsic,
5098 Deoptimization::Action_maybe_recompile);
5099 }
5100
5101 // If control is dead, only non-array-path is taken.
5102 if (stopped()) return true;
5103
5104 // The works fine even if the array type is polymorphic.
5105 // It could be a dynamic mix of int[], boolean[], Object[], etc.
5106 Node* result = load_array_length(array);
5107
5108 C->set_has_split_ifs(true); // Has chance for split-if optimization
5109 set_result(result);
5110 return true;
5111 }
5112
5113 //------------------------inline_array_copyOf----------------------------
5114 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
5115 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
5116 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
5117 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
5118
5119 // Get the arguments.
5120 Node* original = argument(0);
5121 Node* start = is_copyOfRange? argument(1): intcon(0);
5122 Node* end = is_copyOfRange? argument(2): argument(1);
5123 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
5124
5125 Node* newcopy = nullptr;
5126
5127 // Set the original stack and the reexecute bit for the interpreter to reexecute
5128 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
5129 { PreserveReexecuteState preexecs(this);
5130 jvms()->set_should_reexecute(true);
5131
5132 array_type_mirror = null_check(array_type_mirror);
5133 original = null_check(original);
5134
5135 // Check if a null path was taken unconditionally.
5136 if (stopped()) return true;
5137
5138 Node* orig_length = load_array_length(original);
5139
5140 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
5141 klass_node = null_check(klass_node);
5142
5143 RegionNode* bailout = new RegionNode(1);
5144 record_for_igvn(bailout);
5145
5146 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
5147 // Bail out if that is so.
5148 // Inline type array may have object field that would require a
5149 // write barrier. Conservatively, go to slow path.
5150 // TODO 8251971: Optimize for the case when flat src/dst are later found
5151 // to not contain oops (i.e., move this check to the macro expansion phase).
5152 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5153 const TypeAryPtr* orig_t = _gvn.type(original)->isa_aryptr();
5154 const TypeKlassPtr* tklass = _gvn.type(klass_node)->is_klassptr();
5155 bool exclude_flat = UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, false, false, BarrierSetC2::Parsing) &&
5156 // Can src array be flat and contain oops?
5157 (orig_t == nullptr || (!orig_t->is_not_flat() && (!orig_t->is_flat() || orig_t->elem()->inline_klass()->contains_oops()))) &&
5158 // Can dest array be flat and contain oops?
5159 tklass->can_be_inline_array() && (!tklass->is_flat() || tklass->is_aryklassptr()->elem()->is_instklassptr()->instance_klass()->as_inline_klass()->contains_oops());
5160 Node* not_objArray = exclude_flat ? generate_non_refArray_guard(klass_node, bailout) : generate_typeArray_guard(klass_node, bailout);
5161
5162 Node* refined_klass_node = load_default_refined_array_klass(klass_node, /* type_array_guard= */ false);
5163
5164 if (not_objArray != nullptr) {
5165 // Improve the klass node's type from the new optimistic assumption:
5166 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
5167 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, Type::Offset(0));
5168 Node* cast = new CastPPNode(control(), refined_klass_node, akls);
5169 refined_klass_node = _gvn.transform(cast);
5170 }
5171
5172 // Bail out if either start or end is negative.
5173 generate_negative_guard(start, bailout, &start);
5174 generate_negative_guard(end, bailout, &end);
5175
5176 Node* length = end;
5177 if (_gvn.type(start) != TypeInt::ZERO) {
5178 length = _gvn.transform(new SubINode(end, start));
5179 }
5180
5181 // Bail out if length is negative (i.e., if start > end).
5182 // Without this the new_array would throw
5183 // NegativeArraySizeException but IllegalArgumentException is what
5184 // should be thrown
5185 generate_negative_guard(length, bailout, &length);
5186
5187 // Handle inline type arrays
5188 bool can_validate = !too_many_traps(Deoptimization::Reason_class_check);
5189 if (!stopped()) {
5190 // TODO 8251971
5191 if (!orig_t->is_null_free()) {
5192 // Not statically known to be null free, add a check
5193 generate_fair_guard(null_free_array_test(original), bailout);
5194 }
5195 orig_t = _gvn.type(original)->isa_aryptr();
5196 if (orig_t != nullptr && orig_t->is_flat()) {
5197 // Src is flat, check that dest is flat as well
5198 if (exclude_flat) {
5199 // Dest can't be flat, bail out
5200 bailout->add_req(control());
5201 set_control(top());
5202 } else {
5203 generate_fair_guard(flat_array_test(refined_klass_node, /* flat = */ false), bailout);
5204 }
5205 // TODO 8350865 This is not correct anymore. Write tests and fix logic similar to arraycopy.
5206 } else if (UseArrayFlattening && (orig_t == nullptr || !orig_t->is_not_flat()) &&
5207 // If dest is flat, src must be flat as well (guaranteed by src <: dest check if validated).
5208 ((!tklass->is_flat() && tklass->can_be_inline_array()) || !can_validate)) {
5209 // Src might be flat and dest might not be flat. Go to the slow path if src is flat.
5210 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat.
5211 generate_fair_guard(flat_array_test(load_object_klass(original)), bailout);
5212 if (orig_t != nullptr) {
5213 orig_t = orig_t->cast_to_not_flat();
5214 original = _gvn.transform(new CheckCastPPNode(control(), original, orig_t));
5215 }
5216 }
5217 if (!can_validate) {
5218 // No validation. The subtype check emitted at macro expansion time will not go to the slow
5219 // path but call checkcast_arraycopy which can not handle flat/null-free inline type arrays.
5220 // TODO 8251971: Optimize for the case when src/dest are later found to be both flat/null-free.
5221 generate_fair_guard(flat_array_test(refined_klass_node), bailout);
5222 generate_fair_guard(null_free_array_test(original), bailout);
5223 }
5224 }
5225
5226 // Bail out if start is larger than the original length
5227 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
5228 generate_negative_guard(orig_tail, bailout, &orig_tail);
5229
5230 if (bailout->req() > 1) {
5231 PreserveJVMState pjvms(this);
5232 set_control(_gvn.transform(bailout));
5233 uncommon_trap(Deoptimization::Reason_intrinsic,
5234 Deoptimization::Action_maybe_recompile);
5235 }
5236
5237 if (!stopped()) {
5238 // How many elements will we copy from the original?
5239 // The answer is MinI(orig_tail, length).
5240 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
5241
5242 // Generate a direct call to the right arraycopy function(s).
5243 // We know the copy is disjoint but we might not know if the
5244 // oop stores need checking.
5245 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
5246 // This will fail a store-check if x contains any non-nulls.
5247
5248 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
5249 // loads/stores but it is legal only if we're sure the
5250 // Arrays.copyOf would succeed. So we need all input arguments
5251 // to the copyOf to be validated, including that the copy to the
5252 // new array won't trigger an ArrayStoreException. That subtype
5253 // check can be optimized if we know something on the type of
5254 // the input array from type speculation.
5255 if (_gvn.type(klass_node)->singleton()) {
5256 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
5257 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
5258
5259 int test = C->static_subtype_check(superk, subk);
5260 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
5261 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
5262 if (t_original->speculative_type() != nullptr) {
5263 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
5264 }
5265 }
5266 }
5267
5268 bool validated = false;
5269 // Reason_class_check rather than Reason_intrinsic because we
5270 // want to intrinsify even if this traps.
5271 if (can_validate) {
5272 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
5273
5274 if (not_subtype_ctrl != top()) {
5275 PreserveJVMState pjvms(this);
5276 set_control(not_subtype_ctrl);
5277 uncommon_trap(Deoptimization::Reason_class_check,
5278 Deoptimization::Action_make_not_entrant);
5279 assert(stopped(), "Should be stopped");
5280 }
5281 validated = true;
5282 }
5283
5284 if (!stopped()) {
5285 newcopy = new_array(refined_klass_node, length, 0); // no arguments to push
5286
5287 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
5288 load_object_klass(original), klass_node);
5289 if (!is_copyOfRange) {
5290 ac->set_copyof(validated);
5291 } else {
5292 ac->set_copyofrange(validated);
5293 }
5294 Node* n = _gvn.transform(ac);
5295 if (n == ac) {
5296 ac->connect_outputs(this);
5297 } else {
5298 assert(validated, "shouldn't transform if all arguments not validated");
5299 set_all_memory(n);
5300 }
5301 }
5302 }
5303 } // original reexecute is set back here
5304
5305 C->set_has_split_ifs(true); // Has chance for split-if optimization
5306 if (!stopped()) {
5307 set_result(newcopy);
5308 }
5309 return true;
5310 }
5311
5312
5313 //----------------------generate_virtual_guard---------------------------
5314 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
5315 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
5316 RegionNode* slow_region) {
5317 ciMethod* method = callee();
5318 int vtable_index = method->vtable_index();
5319 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5320 "bad index %d", vtable_index);
5321 // Get the Method* out of the appropriate vtable entry.
5322 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
5323 vtable_index*vtableEntry::size_in_bytes() +
5324 in_bytes(vtableEntry::method_offset());
5325 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
5326 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
5327
5328 // Compare the target method with the expected method (e.g., Object.hashCode).
5329 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
5330
5331 Node* native_call = makecon(native_call_addr);
5332 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
5333 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
5334
5335 return generate_slow_guard(test_native, slow_region);
5336 }
5337
5338 //-----------------------generate_method_call----------------------------
5339 // Use generate_method_call to make a slow-call to the real
5340 // method if the fast path fails. An alternative would be to
5341 // use a stub like OptoRuntime::slow_arraycopy_Java.
5342 // This only works for expanding the current library call,
5343 // not another intrinsic. (E.g., don't use this for making an
5344 // arraycopy call inside of the copyOf intrinsic.)
5345 CallJavaNode*
5346 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
5347 // When compiling the intrinsic method itself, do not use this technique.
5348 guarantee(callee() != C->method(), "cannot make slow-call to self");
5349
5350 ciMethod* method = callee();
5351 // ensure the JVMS we have will be correct for this call
5352 guarantee(method_id == method->intrinsic_id(), "must match");
5353
5354 const TypeFunc* tf = TypeFunc::make(method);
5355 if (res_not_null) {
5356 assert(tf->return_type() == T_OBJECT, "");
5357 const TypeTuple* range = tf->range_cc();
5358 const Type** fields = TypeTuple::fields(range->cnt());
5359 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
5360 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
5361 tf = TypeFunc::make(tf->domain_cc(), new_range);
5362 }
5363 CallJavaNode* slow_call;
5364 if (is_static) {
5365 assert(!is_virtual, "");
5366 slow_call = new CallStaticJavaNode(C, tf,
5367 SharedRuntime::get_resolve_static_call_stub(), method);
5368 } else if (is_virtual) {
5369 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5370 int vtable_index = Method::invalid_vtable_index;
5371 if (UseInlineCaches) {
5372 // Suppress the vtable call
5373 } else {
5374 // hashCode and clone are not a miranda methods,
5375 // so the vtable index is fixed.
5376 // No need to use the linkResolver to get it.
5377 vtable_index = method->vtable_index();
5378 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
5379 "bad index %d", vtable_index);
5380 }
5381 slow_call = new CallDynamicJavaNode(tf,
5382 SharedRuntime::get_resolve_virtual_call_stub(),
5383 method, vtable_index);
5384 } else { // neither virtual nor static: opt_virtual
5385 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
5386 slow_call = new CallStaticJavaNode(C, tf,
5387 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
5388 slow_call->set_optimized_virtual(true);
5389 }
5390 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
5391 // To be able to issue a direct call (optimized virtual or virtual)
5392 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
5393 // about the method being invoked should be attached to the call site to
5394 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
5395 slow_call->set_override_symbolic_info(true);
5396 }
5397 set_arguments_for_java_call(slow_call);
5398 set_edges_for_java_call(slow_call);
5399 return slow_call;
5400 }
5401
5402
5403 /**
5404 * Build special case code for calls to hashCode on an object. This call may
5405 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
5406 * slightly different code.
5407 */
5408 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
5409 assert(is_static == callee()->is_static(), "correct intrinsic selection");
5410 assert(!(is_virtual && is_static), "either virtual, special, or static");
5411
5412 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
5413
5414 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5415 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
5416 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
5417 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5418 Node* obj = argument(0);
5419
5420 // Don't intrinsify hashcode on inline types for now.
5421 // The "is locked" runtime check also subsumes the inline type check (as inline types cannot be locked) and goes to the slow path.
5422 if (gvn().type(obj)->is_inlinetypeptr()) {
5423 return false;
5424 }
5425
5426 if (!is_static) {
5427 // Check for hashing null object
5428 obj = null_check_receiver();
5429 if (stopped()) return true; // unconditionally null
5430 result_reg->init_req(_null_path, top());
5431 result_val->init_req(_null_path, top());
5432 } else {
5433 // Do a null check, and return zero if null.
5434 // System.identityHashCode(null) == 0
5435 Node* null_ctl = top();
5436 obj = null_check_oop(obj, &null_ctl);
5437 result_reg->init_req(_null_path, null_ctl);
5438 result_val->init_req(_null_path, _gvn.intcon(0));
5439 }
5440
5441 // Unconditionally null? Then return right away.
5442 if (stopped()) {
5443 set_control( result_reg->in(_null_path));
5444 if (!stopped())
5445 set_result(result_val->in(_null_path));
5446 return true;
5447 }
5448
5449 // We only go to the fast case code if we pass a number of guards. The
5450 // paths which do not pass are accumulated in the slow_region.
5451 RegionNode* slow_region = new RegionNode(1);
5452 record_for_igvn(slow_region);
5453
5454 // If this is a virtual call, we generate a funny guard. We pull out
5455 // the vtable entry corresponding to hashCode() from the target object.
5456 // If the target method which we are calling happens to be the native
5457 // Object hashCode() method, we pass the guard. We do not need this
5458 // guard for non-virtual calls -- the caller is known to be the native
5459 // Object hashCode().
5460 if (is_virtual) {
5461 // After null check, get the object's klass.
5462 Node* obj_klass = load_object_klass(obj);
5463 generate_virtual_guard(obj_klass, slow_region);
5464 }
5465
5466 // Get the header out of the object, use LoadMarkNode when available
5467 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
5468 // The control of the load must be null. Otherwise, the load can move before
5469 // the null check after castPP removal.
5470 Node* no_ctrl = nullptr;
5471 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
5472
5473 if (!UseObjectMonitorTable) {
5474 // Test the header to see if it is safe to read w.r.t. locking.
5475 // We cannot use the inline type mask as this may check bits that are overriden
5476 // by an object monitor's pointer when inflating locking.
5477 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
5478 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
5479 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
5480 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
5481 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
5482
5483 generate_slow_guard(test_monitor, slow_region);
5484 }
5485
5486 // Get the hash value and check to see that it has been properly assigned.
5487 // We depend on hash_mask being at most 32 bits and avoid the use of
5488 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
5489 // vm: see markWord.hpp.
5490 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
5491 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
5492 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
5493 // This hack lets the hash bits live anywhere in the mark object now, as long
5494 // as the shift drops the relevant bits into the low 32 bits. Note that
5495 // Java spec says that HashCode is an int so there's no point in capturing
5496 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
5497 hshifted_header = ConvX2I(hshifted_header);
5498 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
5499
5500 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
5501 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
5502 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
5503
5504 generate_slow_guard(test_assigned, slow_region);
5505
5506 Node* init_mem = reset_memory();
5507 // fill in the rest of the null path:
5508 result_io ->init_req(_null_path, i_o());
5509 result_mem->init_req(_null_path, init_mem);
5510
5511 result_val->init_req(_fast_path, hash_val);
5512 result_reg->init_req(_fast_path, control());
5513 result_io ->init_req(_fast_path, i_o());
5514 result_mem->init_req(_fast_path, init_mem);
5515
5516 // Generate code for the slow case. We make a call to hashCode().
5517 set_control(_gvn.transform(slow_region));
5518 if (!stopped()) {
5519 // No need for PreserveJVMState, because we're using up the present state.
5520 set_all_memory(init_mem);
5521 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
5522 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
5523 Node* slow_result = set_results_for_java_call(slow_call);
5524 // this->control() comes from set_results_for_java_call
5525 result_reg->init_req(_slow_path, control());
5526 result_val->init_req(_slow_path, slow_result);
5527 result_io ->set_req(_slow_path, i_o());
5528 result_mem ->set_req(_slow_path, reset_memory());
5529 }
5530
5531 // Return the combined state.
5532 set_i_o( _gvn.transform(result_io) );
5533 set_all_memory( _gvn.transform(result_mem));
5534
5535 set_result(result_reg, result_val);
5536 return true;
5537 }
5538
5539 //---------------------------inline_native_getClass----------------------------
5540 // public final native Class<?> java.lang.Object.getClass();
5541 //
5542 // Build special case code for calls to getClass on an object.
5543 bool LibraryCallKit::inline_native_getClass() {
5544 Node* obj = argument(0);
5545 if (obj->is_InlineType()) {
5546 const Type* t = _gvn.type(obj);
5547 if (t->maybe_null()) {
5548 null_check(obj);
5549 }
5550 set_result(makecon(TypeInstPtr::make(t->inline_klass()->java_mirror())));
5551 return true;
5552 }
5553 obj = null_check_receiver();
5554 if (stopped()) return true;
5555 set_result(load_mirror_from_klass(load_object_klass(obj)));
5556 return true;
5557 }
5558
5559 //-----------------inline_native_Reflection_getCallerClass---------------------
5560 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
5561 //
5562 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
5563 //
5564 // NOTE: This code must perform the same logic as JVM_GetCallerClass
5565 // in that it must skip particular security frames and checks for
5566 // caller sensitive methods.
5567 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
5568 #ifndef PRODUCT
5569 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5570 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
5571 }
5572 #endif
5573
5574 if (!jvms()->has_method()) {
5575 #ifndef PRODUCT
5576 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5577 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
5578 }
5579 #endif
5580 return false;
5581 }
5582
5583 // Walk back up the JVM state to find the caller at the required
5584 // depth.
5585 JVMState* caller_jvms = jvms();
5586
5587 // Cf. JVM_GetCallerClass
5588 // NOTE: Start the loop at depth 1 because the current JVM state does
5589 // not include the Reflection.getCallerClass() frame.
5590 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5591 ciMethod* m = caller_jvms->method();
5592 switch (n) {
5593 case 0:
5594 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5595 break;
5596 case 1:
5597 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5598 if (!m->caller_sensitive()) {
5599 #ifndef PRODUCT
5600 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5601 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5602 }
5603 #endif
5604 return false; // bail-out; let JVM_GetCallerClass do the work
5605 }
5606 break;
5607 default:
5608 if (!m->is_ignored_by_security_stack_walk()) {
5609 // We have reached the desired frame; return the holder class.
5610 // Acquire method holder as java.lang.Class and push as constant.
5611 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5612 ciInstance* caller_mirror = caller_klass->java_mirror();
5613 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5614
5615 #ifndef PRODUCT
5616 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5617 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());
5618 tty->print_cr(" JVM state at this point:");
5619 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5620 ciMethod* m = jvms()->of_depth(i)->method();
5621 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5622 }
5623 }
5624 #endif
5625 return true;
5626 }
5627 break;
5628 }
5629 }
5630
5631 #ifndef PRODUCT
5632 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5633 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5634 tty->print_cr(" JVM state at this point:");
5635 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5636 ciMethod* m = jvms()->of_depth(i)->method();
5637 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5638 }
5639 }
5640 #endif
5641
5642 return false; // bail-out; let JVM_GetCallerClass do the work
5643 }
5644
5645 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5646 Node* arg = argument(0);
5647 Node* result = nullptr;
5648
5649 switch (id) {
5650 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5651 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5652 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5653 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5654 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5655 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5656
5657 case vmIntrinsics::_doubleToLongBits: {
5658 // two paths (plus control) merge in a wood
5659 RegionNode *r = new RegionNode(3);
5660 Node *phi = new PhiNode(r, TypeLong::LONG);
5661
5662 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5663 // Build the boolean node
5664 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5665
5666 // Branch either way.
5667 // NaN case is less traveled, which makes all the difference.
5668 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5669 Node *opt_isnan = _gvn.transform(ifisnan);
5670 assert( opt_isnan->is_If(), "Expect an IfNode");
5671 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5672 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5673
5674 set_control(iftrue);
5675
5676 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5677 Node *slow_result = longcon(nan_bits); // return NaN
5678 phi->init_req(1, _gvn.transform( slow_result ));
5679 r->init_req(1, iftrue);
5680
5681 // Else fall through
5682 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5683 set_control(iffalse);
5684
5685 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5686 r->init_req(2, iffalse);
5687
5688 // Post merge
5689 set_control(_gvn.transform(r));
5690 record_for_igvn(r);
5691
5692 C->set_has_split_ifs(true); // Has chance for split-if optimization
5693 result = phi;
5694 assert(result->bottom_type()->isa_long(), "must be");
5695 break;
5696 }
5697
5698 case vmIntrinsics::_floatToIntBits: {
5699 // two paths (plus control) merge in a wood
5700 RegionNode *r = new RegionNode(3);
5701 Node *phi = new PhiNode(r, TypeInt::INT);
5702
5703 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5704 // Build the boolean node
5705 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5706
5707 // Branch either way.
5708 // NaN case is less traveled, which makes all the difference.
5709 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5710 Node *opt_isnan = _gvn.transform(ifisnan);
5711 assert( opt_isnan->is_If(), "Expect an IfNode");
5712 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5713 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5714
5715 set_control(iftrue);
5716
5717 static const jint nan_bits = 0x7fc00000;
5718 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5719 phi->init_req(1, _gvn.transform( slow_result ));
5720 r->init_req(1, iftrue);
5721
5722 // Else fall through
5723 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5724 set_control(iffalse);
5725
5726 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5727 r->init_req(2, iffalse);
5728
5729 // Post merge
5730 set_control(_gvn.transform(r));
5731 record_for_igvn(r);
5732
5733 C->set_has_split_ifs(true); // Has chance for split-if optimization
5734 result = phi;
5735 assert(result->bottom_type()->isa_int(), "must be");
5736 break;
5737 }
5738
5739 default:
5740 fatal_unexpected_iid(id);
5741 break;
5742 }
5743 set_result(_gvn.transform(result));
5744 return true;
5745 }
5746
5747 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5748 Node* arg = argument(0);
5749 Node* result = nullptr;
5750
5751 switch (id) {
5752 case vmIntrinsics::_floatIsInfinite:
5753 result = new IsInfiniteFNode(arg);
5754 break;
5755 case vmIntrinsics::_floatIsFinite:
5756 result = new IsFiniteFNode(arg);
5757 break;
5758 case vmIntrinsics::_doubleIsInfinite:
5759 result = new IsInfiniteDNode(arg);
5760 break;
5761 case vmIntrinsics::_doubleIsFinite:
5762 result = new IsFiniteDNode(arg);
5763 break;
5764 default:
5765 fatal_unexpected_iid(id);
5766 break;
5767 }
5768 set_result(_gvn.transform(result));
5769 return true;
5770 }
5771
5772 //----------------------inline_unsafe_copyMemory-------------------------
5773 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5774
5775 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5776 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5777 const Type* base_t = gvn.type(base);
5778
5779 bool in_native = (base_t == TypePtr::NULL_PTR);
5780 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5781 bool is_mixed = !in_heap && !in_native;
5782
5783 if (is_mixed) {
5784 return true; // mixed accesses can touch both on-heap and off-heap memory
5785 }
5786 if (in_heap) {
5787 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5788 if (!is_prim_array) {
5789 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5790 // there's not enough type information available to determine proper memory slice for it.
5791 return true;
5792 }
5793 }
5794 return false;
5795 }
5796
5797 bool LibraryCallKit::inline_unsafe_copyMemory() {
5798 if (callee()->is_static()) return false; // caller must have the capability!
5799 null_check_receiver(); // null-check receiver
5800 if (stopped()) return true;
5801
5802 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5803
5804 Node* src_base = argument(1); // type: oop
5805 Node* src_off = ConvL2X(argument(2)); // type: long
5806 Node* dst_base = argument(4); // type: oop
5807 Node* dst_off = ConvL2X(argument(5)); // type: long
5808 Node* size = ConvL2X(argument(7)); // type: long
5809
5810 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5811 "fieldOffset must be byte-scaled");
5812
5813 Node* src_addr = make_unsafe_address(src_base, src_off);
5814 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5815
5816 Node* thread = _gvn.transform(new ThreadLocalNode());
5817 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5818 BasicType doing_unsafe_access_bt = T_BYTE;
5819 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5820
5821 // update volatile field
5822 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5823
5824 int flags = RC_LEAF | RC_NO_FP;
5825
5826 const TypePtr* dst_type = TypePtr::BOTTOM;
5827
5828 // Adjust memory effects of the runtime call based on input values.
5829 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5830 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5831 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5832
5833 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5834 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5835 flags |= RC_NARROW_MEM; // narrow in memory
5836 }
5837 }
5838
5839 // Call it. Note that the length argument is not scaled.
5840 make_runtime_call(flags,
5841 OptoRuntime::fast_arraycopy_Type(),
5842 StubRoutines::unsafe_arraycopy(),
5843 "unsafe_arraycopy",
5844 dst_type,
5845 src_addr, dst_addr, size XTOP);
5846
5847 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5848
5849 return true;
5850 }
5851
5852 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5853 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5854 bool LibraryCallKit::inline_unsafe_setMemory() {
5855 if (callee()->is_static()) return false; // caller must have the capability!
5856 null_check_receiver(); // null-check receiver
5857 if (stopped()) return true;
5858
5859 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5860
5861 Node* dst_base = argument(1); // type: oop
5862 Node* dst_off = ConvL2X(argument(2)); // type: long
5863 Node* size = ConvL2X(argument(4)); // type: long
5864 Node* byte = argument(6); // type: byte
5865
5866 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5867 "fieldOffset must be byte-scaled");
5868
5869 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5870
5871 Node* thread = _gvn.transform(new ThreadLocalNode());
5872 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5873 BasicType doing_unsafe_access_bt = T_BYTE;
5874 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5875
5876 // update volatile field
5877 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5878
5879 int flags = RC_LEAF | RC_NO_FP;
5880
5881 const TypePtr* dst_type = TypePtr::BOTTOM;
5882
5883 // Adjust memory effects of the runtime call based on input values.
5884 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5885 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5886
5887 flags |= RC_NARROW_MEM; // narrow in memory
5888 }
5889
5890 // Call it. Note that the length argument is not scaled.
5891 make_runtime_call(flags,
5892 OptoRuntime::unsafe_setmemory_Type(),
5893 StubRoutines::unsafe_setmemory(),
5894 "unsafe_setmemory",
5895 dst_type,
5896 dst_addr, size XTOP, byte);
5897
5898 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5899
5900 return true;
5901 }
5902
5903 #undef XTOP
5904
5905 //------------------------clone_coping-----------------------------------
5906 // Helper function for inline_native_clone.
5907 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5908 assert(obj_size != nullptr, "");
5909 Node* raw_obj = alloc_obj->in(1);
5910 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5911
5912 AllocateNode* alloc = nullptr;
5913 if (ReduceBulkZeroing &&
5914 // If we are implementing an array clone without knowing its source type
5915 // (can happen when compiling the array-guarded branch of a reflective
5916 // Object.clone() invocation), initialize the array within the allocation.
5917 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5918 // to a runtime clone call that assumes fully initialized source arrays.
5919 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5920 // We will be completely responsible for initializing this object -
5921 // mark Initialize node as complete.
5922 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5923 // The object was just allocated - there should be no any stores!
5924 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5925 // Mark as complete_with_arraycopy so that on AllocateNode
5926 // expansion, we know this AllocateNode is initialized by an array
5927 // copy and a StoreStore barrier exists after the array copy.
5928 alloc->initialization()->set_complete_with_arraycopy();
5929 }
5930
5931 Node* size = _gvn.transform(obj_size);
5932 access_clone(obj, alloc_obj, size, is_array);
5933
5934 // Do not let reads from the cloned object float above the arraycopy.
5935 if (alloc != nullptr) {
5936 // Do not let stores that initialize this object be reordered with
5937 // a subsequent store that would make this object accessible by
5938 // other threads.
5939 // Record what AllocateNode this StoreStore protects so that
5940 // escape analysis can go from the MemBarStoreStoreNode to the
5941 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5942 // based on the escape status of the AllocateNode.
5943 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5944 } else {
5945 insert_mem_bar(Op_MemBarCPUOrder);
5946 }
5947 }
5948
5949 //------------------------inline_native_clone----------------------------
5950 // protected native Object java.lang.Object.clone();
5951 //
5952 // Here are the simple edge cases:
5953 // null receiver => normal trap
5954 // virtual and clone was overridden => slow path to out-of-line clone
5955 // not cloneable or finalizer => slow path to out-of-line Object.clone
5956 //
5957 // The general case has two steps, allocation and copying.
5958 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5959 //
5960 // Copying also has two cases, oop arrays and everything else.
5961 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5962 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5963 //
5964 // These steps fold up nicely if and when the cloned object's klass
5965 // can be sharply typed as an object array, a type array, or an instance.
5966 //
5967 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5968 PhiNode* result_val;
5969
5970 // Set the reexecute bit for the interpreter to reexecute
5971 // the bytecode that invokes Object.clone if deoptimization happens.
5972 { PreserveReexecuteState preexecs(this);
5973 jvms()->set_should_reexecute(true);
5974
5975 Node* obj = argument(0);
5976 obj = null_check_receiver();
5977 if (stopped()) return true;
5978
5979 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5980 if (obj_type->is_inlinetypeptr()) {
5981 // If the object to clone is an inline type, we can simply return it (i.e. a nop) since inline types have
5982 // no identity.
5983 set_result(obj);
5984 return true;
5985 }
5986
5987 // If we are going to clone an instance, we need its exact type to
5988 // know the number and types of fields to convert the clone to
5989 // loads/stores. Maybe a speculative type can help us.
5990 if (!obj_type->klass_is_exact() &&
5991 obj_type->speculative_type() != nullptr &&
5992 obj_type->speculative_type()->is_instance_klass() &&
5993 !obj_type->speculative_type()->is_inlinetype()) {
5994 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5995 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5996 !spec_ik->has_injected_fields()) {
5997 if (!obj_type->isa_instptr() ||
5998 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5999 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
6000 }
6001 }
6002 }
6003
6004 // Conservatively insert a memory barrier on all memory slices.
6005 // Do not let writes into the original float below the clone.
6006 insert_mem_bar(Op_MemBarCPUOrder);
6007
6008 // paths into result_reg:
6009 enum {
6010 _slow_path = 1, // out-of-line call to clone method (virtual or not)
6011 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
6012 _array_path, // plain array allocation, plus arrayof_long_arraycopy
6013 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
6014 PATH_LIMIT
6015 };
6016 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
6017 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
6018 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
6019 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
6020 record_for_igvn(result_reg);
6021
6022 Node* obj_klass = load_object_klass(obj);
6023 // We only go to the fast case code if we pass a number of guards.
6024 // The paths which do not pass are accumulated in the slow_region.
6025 RegionNode* slow_region = new RegionNode(1);
6026 record_for_igvn(slow_region);
6027
6028 Node* array_obj = obj;
6029 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
6030 if (array_ctl != nullptr) {
6031 // It's an array.
6032 PreserveJVMState pjvms(this);
6033 set_control(array_ctl);
6034
6035 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6036 const TypeAryPtr* ary_ptr = obj_type->isa_aryptr();
6037 if (UseArrayFlattening && bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Expansion) &&
6038 obj_type->can_be_inline_array() &&
6039 (ary_ptr == nullptr || (!ary_ptr->is_not_flat() && (!ary_ptr->is_flat() || ary_ptr->elem()->inline_klass()->contains_oops())))) {
6040 // Flat inline type array may have object field that would require a
6041 // write barrier. Conservatively, go to slow path.
6042 generate_fair_guard(flat_array_test(obj_klass), slow_region);
6043 }
6044
6045 if (!stopped()) {
6046 Node* obj_length = load_array_length(array_obj);
6047 Node* array_size = nullptr; // Size of the array without object alignment padding.
6048 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
6049
6050 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
6051 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
6052 // If it is an oop array, it requires very special treatment,
6053 // because gc barriers are required when accessing the array.
6054 Node* is_obja = generate_refArray_guard(obj_klass, (RegionNode*)nullptr);
6055 if (is_obja != nullptr) {
6056 PreserveJVMState pjvms2(this);
6057 set_control(is_obja);
6058 // Generate a direct call to the right arraycopy function(s).
6059 // Clones are always tightly coupled.
6060 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
6061 ac->set_clone_oop_array();
6062 Node* n = _gvn.transform(ac);
6063 assert(n == ac, "cannot disappear");
6064 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
6065
6066 result_reg->init_req(_objArray_path, control());
6067 result_val->init_req(_objArray_path, alloc_obj);
6068 result_i_o ->set_req(_objArray_path, i_o());
6069 result_mem ->set_req(_objArray_path, reset_memory());
6070 }
6071 }
6072 // Otherwise, there are no barriers to worry about.
6073 // (We can dispense with card marks if we know the allocation
6074 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
6075 // causes the non-eden paths to take compensating steps to
6076 // simulate a fresh allocation, so that no further
6077 // card marks are required in compiled code to initialize
6078 // the object.)
6079
6080 if (!stopped()) {
6081 copy_to_clone(obj, alloc_obj, array_size, true);
6082
6083 // Present the results of the copy.
6084 result_reg->init_req(_array_path, control());
6085 result_val->init_req(_array_path, alloc_obj);
6086 result_i_o ->set_req(_array_path, i_o());
6087 result_mem ->set_req(_array_path, reset_memory());
6088 }
6089 }
6090 }
6091
6092 if (!stopped()) {
6093 // It's an instance (we did array above). Make the slow-path tests.
6094 // If this is a virtual call, we generate a funny guard. We grab
6095 // the vtable entry corresponding to clone() from the target object.
6096 // If the target method which we are calling happens to be the
6097 // Object clone() method, we pass the guard. We do not need this
6098 // guard for non-virtual calls; the caller is known to be the native
6099 // Object clone().
6100 if (is_virtual) {
6101 generate_virtual_guard(obj_klass, slow_region);
6102 }
6103
6104 // The object must be easily cloneable and must not have a finalizer.
6105 // Both of these conditions may be checked in a single test.
6106 // We could optimize the test further, but we don't care.
6107 generate_misc_flags_guard(obj_klass,
6108 // Test both conditions:
6109 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
6110 // Must be cloneable but not finalizer:
6111 KlassFlags::_misc_is_cloneable_fast,
6112 slow_region);
6113 }
6114
6115 if (!stopped()) {
6116 // It's an instance, and it passed the slow-path tests.
6117 PreserveJVMState pjvms(this);
6118 Node* obj_size = nullptr; // Total object size, including object alignment padding.
6119 // Need to deoptimize on exception from allocation since Object.clone intrinsic
6120 // is reexecuted if deoptimization occurs and there could be problems when merging
6121 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
6122 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
6123
6124 copy_to_clone(obj, alloc_obj, obj_size, false);
6125
6126 // Present the results of the slow call.
6127 result_reg->init_req(_instance_path, control());
6128 result_val->init_req(_instance_path, alloc_obj);
6129 result_i_o ->set_req(_instance_path, i_o());
6130 result_mem ->set_req(_instance_path, reset_memory());
6131 }
6132
6133 // Generate code for the slow case. We make a call to clone().
6134 set_control(_gvn.transform(slow_region));
6135 if (!stopped()) {
6136 PreserveJVMState pjvms(this);
6137 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
6138 // We need to deoptimize on exception (see comment above)
6139 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
6140 // this->control() comes from set_results_for_java_call
6141 result_reg->init_req(_slow_path, control());
6142 result_val->init_req(_slow_path, slow_result);
6143 result_i_o ->set_req(_slow_path, i_o());
6144 result_mem ->set_req(_slow_path, reset_memory());
6145 }
6146
6147 // Return the combined state.
6148 set_control( _gvn.transform(result_reg));
6149 set_i_o( _gvn.transform(result_i_o));
6150 set_all_memory( _gvn.transform(result_mem));
6151 } // original reexecute is set back here
6152
6153 set_result(_gvn.transform(result_val));
6154 return true;
6155 }
6156
6157 // If we have a tightly coupled allocation, the arraycopy may take care
6158 // of the array initialization. If one of the guards we insert between
6159 // the allocation and the arraycopy causes a deoptimization, an
6160 // uninitialized array will escape the compiled method. To prevent that
6161 // we set the JVM state for uncommon traps between the allocation and
6162 // the arraycopy to the state before the allocation so, in case of
6163 // deoptimization, we'll reexecute the allocation and the
6164 // initialization.
6165 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
6166 if (alloc != nullptr) {
6167 ciMethod* trap_method = alloc->jvms()->method();
6168 int trap_bci = alloc->jvms()->bci();
6169
6170 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6171 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
6172 // Make sure there's no store between the allocation and the
6173 // arraycopy otherwise visible side effects could be rexecuted
6174 // in case of deoptimization and cause incorrect execution.
6175 bool no_interfering_store = true;
6176 Node* mem = alloc->in(TypeFunc::Memory);
6177 if (mem->is_MergeMem()) {
6178 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
6179 Node* n = mms.memory();
6180 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6181 assert(n->is_Store(), "what else?");
6182 no_interfering_store = false;
6183 break;
6184 }
6185 }
6186 } else {
6187 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
6188 Node* n = mms.memory();
6189 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
6190 assert(n->is_Store(), "what else?");
6191 no_interfering_store = false;
6192 break;
6193 }
6194 }
6195 }
6196
6197 if (no_interfering_store) {
6198 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6199
6200 JVMState* saved_jvms = jvms();
6201 saved_reexecute_sp = _reexecute_sp;
6202
6203 set_jvms(sfpt->jvms());
6204 _reexecute_sp = jvms()->sp();
6205
6206 return saved_jvms;
6207 }
6208 }
6209 }
6210 return nullptr;
6211 }
6212
6213 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
6214 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
6215 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
6216 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
6217 uint size = alloc->req();
6218 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
6219 old_jvms->set_map(sfpt);
6220 for (uint i = 0; i < size; i++) {
6221 sfpt->init_req(i, alloc->in(i));
6222 }
6223 int adjustment = 1;
6224 const TypeAryKlassPtr* ary_klass_ptr = alloc->in(AllocateNode::KlassNode)->bottom_type()->is_aryklassptr();
6225 if (ary_klass_ptr->is_null_free()) {
6226 // A null-free, tightly coupled array allocation can only come from LibraryCallKit::inline_newArray which
6227 // also requires the componentType and initVal on stack for re-execution.
6228 // Re-create and push the componentType.
6229 ciArrayKlass* klass = ary_klass_ptr->exact_klass()->as_array_klass();
6230 ciInstance* instance = klass->component_mirror_instance();
6231 const TypeInstPtr* t_instance = TypeInstPtr::make(instance);
6232 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), makecon(t_instance));
6233 adjustment++;
6234 }
6235 // re-push array length for deoptimization
6236 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment - 1, alloc->in(AllocateNode::ALength));
6237 if (ary_klass_ptr->is_null_free()) {
6238 // Re-create and push the initVal.
6239 Node* init_val = alloc->in(AllocateNode::InitValue);
6240 if (init_val == nullptr) {
6241 init_val = InlineTypeNode::make_all_zero(_gvn, ary_klass_ptr->elem()->is_instklassptr()->instance_klass()->as_inline_klass());
6242 } else if (UseCompressedOops) {
6243 init_val = _gvn.transform(new DecodeNNode(init_val, init_val->bottom_type()->make_ptr()));
6244 }
6245 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp() + adjustment, init_val);
6246 adjustment++;
6247 }
6248 old_jvms->set_sp(old_jvms->sp() + adjustment);
6249 old_jvms->set_monoff(old_jvms->monoff() + adjustment);
6250 old_jvms->set_scloff(old_jvms->scloff() + adjustment);
6251 old_jvms->set_endoff(old_jvms->endoff() + adjustment);
6252 old_jvms->set_should_reexecute(true);
6253
6254 sfpt->set_i_o(map()->i_o());
6255 sfpt->set_memory(map()->memory());
6256 sfpt->set_control(map()->control());
6257 return sfpt;
6258 }
6259
6260 // In case of a deoptimization, we restart execution at the
6261 // allocation, allocating a new array. We would leave an uninitialized
6262 // array in the heap that GCs wouldn't expect. Move the allocation
6263 // after the traps so we don't allocate the array if we
6264 // deoptimize. This is possible because tightly_coupled_allocation()
6265 // guarantees there's no observer of the allocated array at this point
6266 // and the control flow is simple enough.
6267 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
6268 int saved_reexecute_sp, uint new_idx) {
6269 if (saved_jvms_before_guards != nullptr && !stopped()) {
6270 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
6271
6272 assert(alloc != nullptr, "only with a tightly coupled allocation");
6273 // restore JVM state to the state at the arraycopy
6274 saved_jvms_before_guards->map()->set_control(map()->control());
6275 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
6276 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
6277 // If we've improved the types of some nodes (null check) while
6278 // emitting the guards, propagate them to the current state
6279 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
6280 set_jvms(saved_jvms_before_guards);
6281 _reexecute_sp = saved_reexecute_sp;
6282
6283 // Remove the allocation from above the guards
6284 CallProjections* callprojs = alloc->extract_projections(true);
6285 InitializeNode* init = alloc->initialization();
6286 Node* alloc_mem = alloc->in(TypeFunc::Memory);
6287 C->gvn_replace_by(callprojs->fallthrough_ioproj, alloc->in(TypeFunc::I_O));
6288 init->replace_mem_projs_by(alloc_mem, C);
6289
6290 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
6291 // the allocation (i.e. is only valid if the allocation succeeds):
6292 // 1) replace CastIINode with AllocateArrayNode's length here
6293 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
6294 //
6295 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
6296 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
6297 Node* init_control = init->proj_out(TypeFunc::Control);
6298 Node* alloc_length = alloc->Ideal_length();
6299 #ifdef ASSERT
6300 Node* prev_cast = nullptr;
6301 #endif
6302 for (uint i = 0; i < init_control->outcnt(); i++) {
6303 Node* init_out = init_control->raw_out(i);
6304 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
6305 #ifdef ASSERT
6306 if (prev_cast == nullptr) {
6307 prev_cast = init_out;
6308 } else {
6309 if (prev_cast->cmp(*init_out) == false) {
6310 prev_cast->dump();
6311 init_out->dump();
6312 assert(false, "not equal CastIINode");
6313 }
6314 }
6315 #endif
6316 C->gvn_replace_by(init_out, alloc_length);
6317 }
6318 }
6319 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
6320
6321 // move the allocation here (after the guards)
6322 _gvn.hash_delete(alloc);
6323 alloc->set_req(TypeFunc::Control, control());
6324 alloc->set_req(TypeFunc::I_O, i_o());
6325 Node *mem = reset_memory();
6326 set_all_memory(mem);
6327 alloc->set_req(TypeFunc::Memory, mem);
6328 set_control(init->proj_out_or_null(TypeFunc::Control));
6329 set_i_o(callprojs->fallthrough_ioproj);
6330
6331 // Update memory as done in GraphKit::set_output_for_allocation()
6332 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
6333 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
6334 if (ary_type->isa_aryptr() && length_type != nullptr) {
6335 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
6336 }
6337 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
6338 int elemidx = C->get_alias_index(telemref);
6339 // Need to properly move every memory projection for the Initialize
6340 #ifdef ASSERT
6341 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
6342 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
6343 #endif
6344 auto move_proj = [&](ProjNode* proj) {
6345 int alias_idx = C->get_alias_index(proj->adr_type());
6346 assert(alias_idx == Compile::AliasIdxRaw ||
6347 alias_idx == elemidx ||
6348 alias_idx == mark_idx ||
6349 alias_idx == klass_idx, "should be raw memory or array element type");
6350 set_memory(proj, alias_idx);
6351 };
6352 init->for_each_proj(move_proj, TypeFunc::Memory);
6353
6354 Node* allocx = _gvn.transform(alloc);
6355 assert(allocx == alloc, "where has the allocation gone?");
6356 assert(dest->is_CheckCastPP(), "not an allocation result?");
6357
6358 _gvn.hash_delete(dest);
6359 dest->set_req(0, control());
6360 Node* destx = _gvn.transform(dest);
6361 assert(destx == dest, "where has the allocation result gone?");
6362
6363 array_ideal_length(alloc, ary_type, true);
6364 }
6365 }
6366
6367 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
6368 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
6369 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
6370 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
6371 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
6372 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
6373 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
6374 JVMState* saved_jvms_before_guards) {
6375 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
6376 // There is at least one unrelated uncommon trap which needs to be replaced.
6377 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
6378
6379 JVMState* saved_jvms = jvms();
6380 const int saved_reexecute_sp = _reexecute_sp;
6381 set_jvms(sfpt->jvms());
6382 _reexecute_sp = jvms()->sp();
6383
6384 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
6385
6386 // Restore state
6387 set_jvms(saved_jvms);
6388 _reexecute_sp = saved_reexecute_sp;
6389 }
6390 }
6391
6392 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
6393 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
6394 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
6395 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
6396 while (if_proj->is_IfProj()) {
6397 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
6398 if (uncommon_trap != nullptr) {
6399 create_new_uncommon_trap(uncommon_trap);
6400 }
6401 assert(if_proj->in(0)->is_If(), "must be If");
6402 if_proj = if_proj->in(0)->in(0);
6403 }
6404 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
6405 "must have reached control projection of init node");
6406 }
6407
6408 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
6409 const int trap_request = uncommon_trap_call->uncommon_trap_request();
6410 assert(trap_request != 0, "no valid UCT trap request");
6411 PreserveJVMState pjvms(this);
6412 set_control(uncommon_trap_call->in(0));
6413 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
6414 Deoptimization::trap_request_action(trap_request));
6415 assert(stopped(), "Should be stopped");
6416 _gvn.hash_delete(uncommon_trap_call);
6417 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
6418 }
6419
6420 // Common checks for array sorting intrinsics arguments.
6421 // Returns `true` if checks passed.
6422 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
6423 // check address of the class
6424 if (elementType == nullptr || elementType->is_top()) {
6425 return false; // dead path
6426 }
6427 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
6428 if (elem_klass == nullptr) {
6429 return false; // dead path
6430 }
6431 // java_mirror_type() returns non-null for compile-time Class constants only
6432 ciType* elem_type = elem_klass->java_mirror_type();
6433 if (elem_type == nullptr) {
6434 return false;
6435 }
6436 bt = elem_type->basic_type();
6437 // Disable the intrinsic if the CPU does not support SIMD sort
6438 if (!Matcher::supports_simd_sort(bt)) {
6439 return false;
6440 }
6441 // check address of the array
6442 if (obj == nullptr || obj->is_top()) {
6443 return false; // dead path
6444 }
6445 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
6446 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
6447 return false; // failed input validation
6448 }
6449 return true;
6450 }
6451
6452 //------------------------------inline_array_partition-----------------------
6453 bool LibraryCallKit::inline_array_partition() {
6454 address stubAddr = StubRoutines::select_array_partition_function();
6455 if (stubAddr == nullptr) {
6456 return false; // Intrinsic's stub is not implemented on this platform
6457 }
6458 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
6459
6460 // no receiver because it is a static method
6461 Node* elementType = argument(0);
6462 Node* obj = argument(1);
6463 Node* offset = argument(2); // long
6464 Node* fromIndex = argument(4);
6465 Node* toIndex = argument(5);
6466 Node* indexPivot1 = argument(6);
6467 Node* indexPivot2 = argument(7);
6468 // PartitionOperation: argument(8) is ignored
6469
6470 Node* pivotIndices = nullptr;
6471 BasicType bt = T_ILLEGAL;
6472
6473 if (!check_array_sort_arguments(elementType, obj, bt)) {
6474 return false;
6475 }
6476 null_check(obj);
6477 // If obj is dead, only null-path is taken.
6478 if (stopped()) {
6479 return true;
6480 }
6481 // Set the original stack and the reexecute bit for the interpreter to reexecute
6482 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
6483 { PreserveReexecuteState preexecs(this);
6484 jvms()->set_should_reexecute(true);
6485
6486 Node* obj_adr = make_unsafe_address(obj, offset);
6487
6488 // create the pivotIndices array of type int and size = 2
6489 Node* size = intcon(2);
6490 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
6491 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
6492 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
6493 guarantee(alloc != nullptr, "created above");
6494 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
6495
6496 // pass the basic type enum to the stub
6497 Node* elemType = intcon(bt);
6498
6499 // Call the stub
6500 const char *stubName = "array_partition_stub";
6501 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
6502 stubAddr, stubName, TypePtr::BOTTOM,
6503 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
6504 indexPivot1, indexPivot2);
6505
6506 } // original reexecute is set back here
6507
6508 if (!stopped()) {
6509 set_result(pivotIndices);
6510 }
6511
6512 return true;
6513 }
6514
6515
6516 //------------------------------inline_array_sort-----------------------
6517 bool LibraryCallKit::inline_array_sort() {
6518 address stubAddr = StubRoutines::select_arraysort_function();
6519 if (stubAddr == nullptr) {
6520 return false; // Intrinsic's stub is not implemented on this platform
6521 }
6522 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
6523
6524 // no receiver because it is a static method
6525 Node* elementType = argument(0);
6526 Node* obj = argument(1);
6527 Node* offset = argument(2); // long
6528 Node* fromIndex = argument(4);
6529 Node* toIndex = argument(5);
6530 // SortOperation: argument(6) is ignored
6531
6532 BasicType bt = T_ILLEGAL;
6533
6534 if (!check_array_sort_arguments(elementType, obj, bt)) {
6535 return false;
6536 }
6537 null_check(obj);
6538 // If obj is dead, only null-path is taken.
6539 if (stopped()) {
6540 return true;
6541 }
6542 Node* obj_adr = make_unsafe_address(obj, offset);
6543
6544 // pass the basic type enum to the stub
6545 Node* elemType = intcon(bt);
6546
6547 // Call the stub.
6548 const char *stubName = "arraysort_stub";
6549 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
6550 stubAddr, stubName, TypePtr::BOTTOM,
6551 obj_adr, elemType, fromIndex, toIndex);
6552
6553 return true;
6554 }
6555
6556
6557 //------------------------------inline_arraycopy-----------------------
6558 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
6559 // Object dest, int destPos,
6560 // int length);
6561 bool LibraryCallKit::inline_arraycopy() {
6562 // Get the arguments.
6563 Node* src = argument(0); // type: oop
6564 Node* src_offset = argument(1); // type: int
6565 Node* dest = argument(2); // type: oop
6566 Node* dest_offset = argument(3); // type: int
6567 Node* length = argument(4); // type: int
6568
6569 uint new_idx = C->unique();
6570
6571 // Check for allocation before we add nodes that would confuse
6572 // tightly_coupled_allocation()
6573 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
6574
6575 int saved_reexecute_sp = -1;
6576 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
6577 // See arraycopy_restore_alloc_state() comment
6578 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
6579 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
6580 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
6581 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
6582
6583 // The following tests must be performed
6584 // (1) src and dest are arrays.
6585 // (2) src and dest arrays must have elements of the same BasicType
6586 // (3) src and dest must not be null.
6587 // (4) src_offset must not be negative.
6588 // (5) dest_offset must not be negative.
6589 // (6) length must not be negative.
6590 // (7) src_offset + length must not exceed length of src.
6591 // (8) dest_offset + length must not exceed length of dest.
6592 // (9) each element of an oop array must be assignable
6593
6594 // (3) src and dest must not be null.
6595 // always do this here because we need the JVM state for uncommon traps
6596 Node* null_ctl = top();
6597 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
6598 assert(null_ctl->is_top(), "no null control here");
6599 dest = null_check(dest, T_ARRAY);
6600
6601 if (!can_emit_guards) {
6602 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
6603 // guards but the arraycopy node could still take advantage of a
6604 // tightly allocated allocation. tightly_coupled_allocation() is
6605 // called again to make sure it takes the null check above into
6606 // account: the null check is mandatory and if it caused an
6607 // uncommon trap to be emitted then the allocation can't be
6608 // considered tightly coupled in this context.
6609 alloc = tightly_coupled_allocation(dest);
6610 }
6611
6612 bool validated = false;
6613
6614 const Type* src_type = _gvn.type(src);
6615 const Type* dest_type = _gvn.type(dest);
6616 const TypeAryPtr* top_src = src_type->isa_aryptr();
6617 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
6618
6619 // Do we have the type of src?
6620 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6621 // Do we have the type of dest?
6622 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6623 // Is the type for src from speculation?
6624 bool src_spec = false;
6625 // Is the type for dest from speculation?
6626 bool dest_spec = false;
6627
6628 if ((!has_src || !has_dest) && can_emit_guards) {
6629 // We don't have sufficient type information, let's see if
6630 // speculative types can help. We need to have types for both src
6631 // and dest so that it pays off.
6632
6633 // Do we already have or could we have type information for src
6634 bool could_have_src = has_src;
6635 // Do we already have or could we have type information for dest
6636 bool could_have_dest = has_dest;
6637
6638 ciKlass* src_k = nullptr;
6639 if (!has_src) {
6640 src_k = src_type->speculative_type_not_null();
6641 if (src_k != nullptr && src_k->is_array_klass()) {
6642 could_have_src = true;
6643 }
6644 }
6645
6646 ciKlass* dest_k = nullptr;
6647 if (!has_dest) {
6648 dest_k = dest_type->speculative_type_not_null();
6649 if (dest_k != nullptr && dest_k->is_array_klass()) {
6650 could_have_dest = true;
6651 }
6652 }
6653
6654 if (could_have_src && could_have_dest) {
6655 // This is going to pay off so emit the required guards
6656 if (!has_src) {
6657 src = maybe_cast_profiled_obj(src, src_k, true);
6658 src_type = _gvn.type(src);
6659 top_src = src_type->isa_aryptr();
6660 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6661 src_spec = true;
6662 }
6663 if (!has_dest) {
6664 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6665 dest_type = _gvn.type(dest);
6666 top_dest = dest_type->isa_aryptr();
6667 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6668 dest_spec = true;
6669 }
6670 }
6671 }
6672
6673 if (has_src && has_dest && can_emit_guards) {
6674 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6675 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6676 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6677 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6678
6679 if (src_elem == dest_elem && top_src->is_flat() == top_dest->is_flat() && src_elem == T_OBJECT) {
6680 // If both arrays are object arrays then having the exact types
6681 // for both will remove the need for a subtype check at runtime
6682 // before the call and may make it possible to pick a faster copy
6683 // routine (without a subtype check on every element)
6684 // Do we have the exact type of src?
6685 bool could_have_src = src_spec;
6686 // Do we have the exact type of dest?
6687 bool could_have_dest = dest_spec;
6688 ciKlass* src_k = nullptr;
6689 ciKlass* dest_k = nullptr;
6690 if (!src_spec) {
6691 src_k = src_type->speculative_type_not_null();
6692 if (src_k != nullptr && src_k->is_array_klass()) {
6693 could_have_src = true;
6694 }
6695 }
6696 if (!dest_spec) {
6697 dest_k = dest_type->speculative_type_not_null();
6698 if (dest_k != nullptr && dest_k->is_array_klass()) {
6699 could_have_dest = true;
6700 }
6701 }
6702 if (could_have_src && could_have_dest) {
6703 // If we can have both exact types, emit the missing guards
6704 if (could_have_src && !src_spec) {
6705 src = maybe_cast_profiled_obj(src, src_k, true);
6706 src_type = _gvn.type(src);
6707 top_src = src_type->isa_aryptr();
6708 }
6709 if (could_have_dest && !dest_spec) {
6710 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6711 dest_type = _gvn.type(dest);
6712 top_dest = dest_type->isa_aryptr();
6713 }
6714 }
6715 }
6716 }
6717
6718 ciMethod* trap_method = method();
6719 int trap_bci = bci();
6720 if (saved_jvms_before_guards != nullptr) {
6721 trap_method = alloc->jvms()->method();
6722 trap_bci = alloc->jvms()->bci();
6723 }
6724
6725 bool negative_length_guard_generated = false;
6726
6727 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6728 can_emit_guards && !src->is_top() && !dest->is_top()) {
6729 // validate arguments: enables transformation the ArrayCopyNode
6730 validated = true;
6731
6732 RegionNode* slow_region = new RegionNode(1);
6733 record_for_igvn(slow_region);
6734
6735 // (1) src and dest are arrays.
6736 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6737 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6738
6739 // (2) src and dest arrays must have elements of the same BasicType
6740 // done at macro expansion or at Ideal transformation time
6741
6742 // (4) src_offset must not be negative.
6743 generate_negative_guard(src_offset, slow_region);
6744
6745 // (5) dest_offset must not be negative.
6746 generate_negative_guard(dest_offset, slow_region);
6747
6748 // (7) src_offset + length must not exceed length of src.
6749 generate_limit_guard(src_offset, length,
6750 load_array_length(src),
6751 slow_region);
6752
6753 // (8) dest_offset + length must not exceed length of dest.
6754 generate_limit_guard(dest_offset, length,
6755 load_array_length(dest),
6756 slow_region);
6757
6758 // (6) length must not be negative.
6759 // This is also checked in generate_arraycopy() during macro expansion, but
6760 // we also have to check it here for the case where the ArrayCopyNode will
6761 // be eliminated by Escape Analysis.
6762 if (EliminateAllocations) {
6763 generate_negative_guard(length, slow_region);
6764 negative_length_guard_generated = true;
6765 }
6766
6767 // (9) each element of an oop array must be assignable
6768 Node* dest_klass = load_object_klass(dest);
6769 Node* refined_dest_klass = dest_klass;
6770 if (src != dest) {
6771 dest_klass = load_non_refined_array_klass(refined_dest_klass);
6772 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6773 slow_region->add_req(not_subtype_ctrl);
6774 }
6775
6776 // TODO 8350865 Improve this. What about atomicity? Make sure this is always folded for type arrays.
6777 // If destination is null-restricted, source must be null-restricted as well: src_null_restricted || !dst_null_restricted
6778 Node* src_klass = load_object_klass(src);
6779 Node* adr_prop_src = basic_plus_adr(src_klass, in_bytes(ArrayKlass::properties_offset()));
6780 Node* prop_src = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_src, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6781 Node* adr_prop_dest = basic_plus_adr(refined_dest_klass, in_bytes(ArrayKlass::properties_offset()));
6782 Node* prop_dest = _gvn.transform(LoadNode::make(_gvn, control(), immutable_memory(), adr_prop_dest, TypeRawPtr::BOTTOM, TypeInt::INT, T_INT, MemNode::unordered));
6783
6784 prop_dest = _gvn.transform(new XorINode(prop_dest, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6785 prop_src = _gvn.transform(new OrINode(prop_dest, prop_src));
6786 prop_src = _gvn.transform(new AndINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6787
6788 Node* chk = _gvn.transform(new CmpINode(prop_src, intcon(ArrayKlass::ArrayProperties::NULL_RESTRICTED)));
6789 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::ne));
6790 generate_fair_guard(tst, slow_region);
6791
6792 // TODO 8350865 This is too strong
6793 generate_fair_guard(flat_array_test(src), slow_region);
6794 generate_fair_guard(flat_array_test(dest), slow_region);
6795
6796 {
6797 PreserveJVMState pjvms(this);
6798 set_control(_gvn.transform(slow_region));
6799 uncommon_trap(Deoptimization::Reason_intrinsic,
6800 Deoptimization::Action_make_not_entrant);
6801 assert(stopped(), "Should be stopped");
6802 }
6803
6804 const TypeKlassPtr* dest_klass_t = _gvn.type(refined_dest_klass)->is_klassptr();
6805 const Type* toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6806 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6807 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6808 }
6809
6810 if (stopped()) {
6811 return true;
6812 }
6813
6814 Node* dest_klass = load_object_klass(dest);
6815 dest_klass = load_non_refined_array_klass(dest_klass);
6816
6817 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6818 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6819 // so the compiler has a chance to eliminate them: during macro expansion,
6820 // we have to set their control (CastPP nodes are eliminated).
6821 load_object_klass(src), dest_klass,
6822 load_array_length(src), load_array_length(dest));
6823
6824 ac->set_arraycopy(validated);
6825
6826 Node* n = _gvn.transform(ac);
6827 if (n == ac) {
6828 ac->connect_outputs(this);
6829 } else {
6830 assert(validated, "shouldn't transform if all arguments not validated");
6831 set_all_memory(n);
6832 }
6833 clear_upper_avx();
6834
6835
6836 return true;
6837 }
6838
6839
6840 // Helper function which determines if an arraycopy immediately follows
6841 // an allocation, with no intervening tests or other escapes for the object.
6842 AllocateArrayNode*
6843 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6844 if (stopped()) return nullptr; // no fast path
6845 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6846
6847 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6848 if (alloc == nullptr) return nullptr;
6849
6850 Node* rawmem = memory(Compile::AliasIdxRaw);
6851 // Is the allocation's memory state untouched?
6852 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6853 // Bail out if there have been raw-memory effects since the allocation.
6854 // (Example: There might have been a call or safepoint.)
6855 return nullptr;
6856 }
6857 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6858 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6859 return nullptr;
6860 }
6861
6862 // There must be no unexpected observers of this allocation.
6863 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6864 Node* obs = ptr->fast_out(i);
6865 if (obs != this->map()) {
6866 return nullptr;
6867 }
6868 }
6869
6870 // This arraycopy must unconditionally follow the allocation of the ptr.
6871 Node* alloc_ctl = ptr->in(0);
6872 Node* ctl = control();
6873 while (ctl != alloc_ctl) {
6874 // There may be guards which feed into the slow_region.
6875 // Any other control flow means that we might not get a chance
6876 // to finish initializing the allocated object.
6877 // Various low-level checks bottom out in uncommon traps. These
6878 // are considered safe since we've already checked above that
6879 // there is no unexpected observer of this allocation.
6880 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6881 assert(ctl->in(0)->is_If(), "must be If");
6882 ctl = ctl->in(0)->in(0);
6883 } else {
6884 return nullptr;
6885 }
6886 }
6887
6888 // If we get this far, we have an allocation which immediately
6889 // precedes the arraycopy, and we can take over zeroing the new object.
6890 // The arraycopy will finish the initialization, and provide
6891 // a new control state to which we will anchor the destination pointer.
6892
6893 return alloc;
6894 }
6895
6896 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6897 if (node->is_IfProj()) {
6898 Node* other_proj = node->as_IfProj()->other_if_proj();
6899 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6900 Node* obs = other_proj->fast_out(j);
6901 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6902 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6903 return obs->as_CallStaticJava();
6904 }
6905 }
6906 }
6907 return nullptr;
6908 }
6909
6910 //-------------inline_encodeISOArray-----------------------------------
6911 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6912 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6913 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6914 // encode char[] to byte[] in ISO_8859_1 or ASCII
6915 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6916 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6917 // no receiver since it is static method
6918 Node *src = argument(0);
6919 Node *src_offset = argument(1);
6920 Node *dst = argument(2);
6921 Node *dst_offset = argument(3);
6922 Node *length = argument(4);
6923
6924 // Cast source & target arrays to not-null
6925 if (VerifyIntrinsicChecks) {
6926 src = must_be_not_null(src, true);
6927 dst = must_be_not_null(dst, true);
6928 if (stopped()) {
6929 return true;
6930 }
6931 }
6932
6933 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6934 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6935 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6936 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6937 // failed array check
6938 return false;
6939 }
6940
6941 // Figure out the size and type of the elements we will be copying.
6942 BasicType src_elem = src_type->elem()->array_element_basic_type();
6943 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6944 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6945 return false;
6946 }
6947
6948 // Check source & target bounds
6949 if (VerifyIntrinsicChecks) {
6950 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, true);
6951 generate_string_range_check(dst, dst_offset, length, false, true);
6952 if (stopped()) {
6953 return true;
6954 }
6955 }
6956
6957 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6958 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6959 // 'src_start' points to src array + scaled offset
6960 // 'dst_start' points to dst array + scaled offset
6961
6962 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6963 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6964 enc = _gvn.transform(enc);
6965 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6966 set_memory(res_mem, mtype);
6967 set_result(enc);
6968 clear_upper_avx();
6969
6970 return true;
6971 }
6972
6973 //-------------inline_multiplyToLen-----------------------------------
6974 bool LibraryCallKit::inline_multiplyToLen() {
6975 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6976
6977 address stubAddr = StubRoutines::multiplyToLen();
6978 if (stubAddr == nullptr) {
6979 return false; // Intrinsic's stub is not implemented on this platform
6980 }
6981 const char* stubName = "multiplyToLen";
6982
6983 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6984
6985 // no receiver because it is a static method
6986 Node* x = argument(0);
6987 Node* xlen = argument(1);
6988 Node* y = argument(2);
6989 Node* ylen = argument(3);
6990 Node* z = argument(4);
6991
6992 x = must_be_not_null(x, true);
6993 y = must_be_not_null(y, true);
6994
6995 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6996 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6997 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6998 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6999 // failed array check
7000 return false;
7001 }
7002
7003 BasicType x_elem = x_type->elem()->array_element_basic_type();
7004 BasicType y_elem = y_type->elem()->array_element_basic_type();
7005 if (x_elem != T_INT || y_elem != T_INT) {
7006 return false;
7007 }
7008
7009 Node* x_start = array_element_address(x, intcon(0), x_elem);
7010 Node* y_start = array_element_address(y, intcon(0), y_elem);
7011 // 'x_start' points to x array + scaled xlen
7012 // 'y_start' points to y array + scaled ylen
7013
7014 Node* z_start = array_element_address(z, intcon(0), T_INT);
7015
7016 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7017 OptoRuntime::multiplyToLen_Type(),
7018 stubAddr, stubName, TypePtr::BOTTOM,
7019 x_start, xlen, y_start, ylen, z_start);
7020
7021 C->set_has_split_ifs(true); // Has chance for split-if optimization
7022 set_result(z);
7023 return true;
7024 }
7025
7026 //-------------inline_squareToLen------------------------------------
7027 bool LibraryCallKit::inline_squareToLen() {
7028 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
7029
7030 address stubAddr = StubRoutines::squareToLen();
7031 if (stubAddr == nullptr) {
7032 return false; // Intrinsic's stub is not implemented on this platform
7033 }
7034 const char* stubName = "squareToLen";
7035
7036 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
7037
7038 Node* x = argument(0);
7039 Node* len = argument(1);
7040 Node* z = argument(2);
7041 Node* zlen = argument(3);
7042
7043 x = must_be_not_null(x, true);
7044 z = must_be_not_null(z, true);
7045
7046 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
7047 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
7048 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
7049 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
7050 // failed array check
7051 return false;
7052 }
7053
7054 BasicType x_elem = x_type->elem()->array_element_basic_type();
7055 BasicType z_elem = z_type->elem()->array_element_basic_type();
7056 if (x_elem != T_INT || z_elem != T_INT) {
7057 return false;
7058 }
7059
7060
7061 Node* x_start = array_element_address(x, intcon(0), x_elem);
7062 Node* z_start = array_element_address(z, intcon(0), z_elem);
7063
7064 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7065 OptoRuntime::squareToLen_Type(),
7066 stubAddr, stubName, TypePtr::BOTTOM,
7067 x_start, len, z_start, zlen);
7068
7069 set_result(z);
7070 return true;
7071 }
7072
7073 //-------------inline_mulAdd------------------------------------------
7074 bool LibraryCallKit::inline_mulAdd() {
7075 assert(UseMulAddIntrinsic, "not implemented on this platform");
7076
7077 address stubAddr = StubRoutines::mulAdd();
7078 if (stubAddr == nullptr) {
7079 return false; // Intrinsic's stub is not implemented on this platform
7080 }
7081 const char* stubName = "mulAdd";
7082
7083 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
7084
7085 Node* out = argument(0);
7086 Node* in = argument(1);
7087 Node* offset = argument(2);
7088 Node* len = argument(3);
7089 Node* k = argument(4);
7090
7091 in = must_be_not_null(in, true);
7092 out = must_be_not_null(out, true);
7093
7094 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
7095 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
7096 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
7097 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
7098 // failed array check
7099 return false;
7100 }
7101
7102 BasicType out_elem = out_type->elem()->array_element_basic_type();
7103 BasicType in_elem = in_type->elem()->array_element_basic_type();
7104 if (out_elem != T_INT || in_elem != T_INT) {
7105 return false;
7106 }
7107
7108 Node* outlen = load_array_length(out);
7109 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
7110 Node* out_start = array_element_address(out, intcon(0), out_elem);
7111 Node* in_start = array_element_address(in, intcon(0), in_elem);
7112
7113 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
7114 OptoRuntime::mulAdd_Type(),
7115 stubAddr, stubName, TypePtr::BOTTOM,
7116 out_start,in_start, new_offset, len, k);
7117 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7118 set_result(result);
7119 return true;
7120 }
7121
7122 //-------------inline_montgomeryMultiply-----------------------------------
7123 bool LibraryCallKit::inline_montgomeryMultiply() {
7124 address stubAddr = StubRoutines::montgomeryMultiply();
7125 if (stubAddr == nullptr) {
7126 return false; // Intrinsic's stub is not implemented on this platform
7127 }
7128
7129 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
7130 const char* stubName = "montgomery_multiply";
7131
7132 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
7133
7134 Node* a = argument(0);
7135 Node* b = argument(1);
7136 Node* n = argument(2);
7137 Node* len = argument(3);
7138 Node* inv = argument(4);
7139 Node* m = argument(6);
7140
7141 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7142 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
7143 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7144 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7145 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7146 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
7147 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7148 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7149 // failed array check
7150 return false;
7151 }
7152
7153 BasicType a_elem = a_type->elem()->array_element_basic_type();
7154 BasicType b_elem = b_type->elem()->array_element_basic_type();
7155 BasicType n_elem = n_type->elem()->array_element_basic_type();
7156 BasicType m_elem = m_type->elem()->array_element_basic_type();
7157 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7158 return false;
7159 }
7160
7161 // Make the call
7162 {
7163 Node* a_start = array_element_address(a, intcon(0), a_elem);
7164 Node* b_start = array_element_address(b, intcon(0), b_elem);
7165 Node* n_start = array_element_address(n, intcon(0), n_elem);
7166 Node* m_start = array_element_address(m, intcon(0), m_elem);
7167
7168 Node* call = make_runtime_call(RC_LEAF,
7169 OptoRuntime::montgomeryMultiply_Type(),
7170 stubAddr, stubName, TypePtr::BOTTOM,
7171 a_start, b_start, n_start, len, inv, top(),
7172 m_start);
7173 set_result(m);
7174 }
7175
7176 return true;
7177 }
7178
7179 bool LibraryCallKit::inline_montgomerySquare() {
7180 address stubAddr = StubRoutines::montgomerySquare();
7181 if (stubAddr == nullptr) {
7182 return false; // Intrinsic's stub is not implemented on this platform
7183 }
7184
7185 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
7186 const char* stubName = "montgomery_square";
7187
7188 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
7189
7190 Node* a = argument(0);
7191 Node* n = argument(1);
7192 Node* len = argument(2);
7193 Node* inv = argument(3);
7194 Node* m = argument(5);
7195
7196 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
7197 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
7198 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
7199 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
7200 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
7201 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
7202 // failed array check
7203 return false;
7204 }
7205
7206 BasicType a_elem = a_type->elem()->array_element_basic_type();
7207 BasicType n_elem = n_type->elem()->array_element_basic_type();
7208 BasicType m_elem = m_type->elem()->array_element_basic_type();
7209 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
7210 return false;
7211 }
7212
7213 // Make the call
7214 {
7215 Node* a_start = array_element_address(a, intcon(0), a_elem);
7216 Node* n_start = array_element_address(n, intcon(0), n_elem);
7217 Node* m_start = array_element_address(m, intcon(0), m_elem);
7218
7219 Node* call = make_runtime_call(RC_LEAF,
7220 OptoRuntime::montgomerySquare_Type(),
7221 stubAddr, stubName, TypePtr::BOTTOM,
7222 a_start, n_start, len, inv, top(),
7223 m_start);
7224 set_result(m);
7225 }
7226
7227 return true;
7228 }
7229
7230 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
7231 address stubAddr = nullptr;
7232 const char* stubName = nullptr;
7233
7234 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
7235 if (stubAddr == nullptr) {
7236 return false; // Intrinsic's stub is not implemented on this platform
7237 }
7238
7239 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
7240
7241 assert(callee()->signature()->size() == 5, "expected 5 arguments");
7242
7243 Node* newArr = argument(0);
7244 Node* oldArr = argument(1);
7245 Node* newIdx = argument(2);
7246 Node* shiftCount = argument(3);
7247 Node* numIter = argument(4);
7248
7249 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
7250 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
7251 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
7252 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
7253 return false;
7254 }
7255
7256 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
7257 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
7258 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
7259 return false;
7260 }
7261
7262 // Make the call
7263 {
7264 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
7265 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
7266
7267 Node* call = make_runtime_call(RC_LEAF,
7268 OptoRuntime::bigIntegerShift_Type(),
7269 stubAddr,
7270 stubName,
7271 TypePtr::BOTTOM,
7272 newArr_start,
7273 oldArr_start,
7274 newIdx,
7275 shiftCount,
7276 numIter);
7277 }
7278
7279 return true;
7280 }
7281
7282 //-------------inline_vectorizedMismatch------------------------------
7283 bool LibraryCallKit::inline_vectorizedMismatch() {
7284 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
7285
7286 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
7287 Node* obja = argument(0); // Object
7288 Node* aoffset = argument(1); // long
7289 Node* objb = argument(3); // Object
7290 Node* boffset = argument(4); // long
7291 Node* length = argument(6); // int
7292 Node* scale = argument(7); // int
7293
7294 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
7295 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
7296 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
7297 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
7298 scale == top()) {
7299 return false; // failed input validation
7300 }
7301
7302 Node* obja_adr = make_unsafe_address(obja, aoffset);
7303 Node* objb_adr = make_unsafe_address(objb, boffset);
7304
7305 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
7306 //
7307 // inline_limit = ArrayOperationPartialInlineSize / element_size;
7308 // if (length <= inline_limit) {
7309 // inline_path:
7310 // vmask = VectorMaskGen length
7311 // vload1 = LoadVectorMasked obja, vmask
7312 // vload2 = LoadVectorMasked objb, vmask
7313 // result1 = VectorCmpMasked vload1, vload2, vmask
7314 // } else {
7315 // call_stub_path:
7316 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
7317 // }
7318 // exit_block:
7319 // return Phi(result1, result2);
7320 //
7321 enum { inline_path = 1, // input is small enough to process it all at once
7322 stub_path = 2, // input is too large; call into the VM
7323 PATH_LIMIT = 3
7324 };
7325
7326 Node* exit_block = new RegionNode(PATH_LIMIT);
7327 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
7328 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
7329
7330 Node* call_stub_path = control();
7331
7332 BasicType elem_bt = T_ILLEGAL;
7333
7334 const TypeInt* scale_t = _gvn.type(scale)->is_int();
7335 if (scale_t->is_con()) {
7336 switch (scale_t->get_con()) {
7337 case 0: elem_bt = T_BYTE; break;
7338 case 1: elem_bt = T_SHORT; break;
7339 case 2: elem_bt = T_INT; break;
7340 case 3: elem_bt = T_LONG; break;
7341
7342 default: elem_bt = T_ILLEGAL; break; // not supported
7343 }
7344 }
7345
7346 int inline_limit = 0;
7347 bool do_partial_inline = false;
7348
7349 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
7350 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
7351 do_partial_inline = inline_limit >= 16;
7352 }
7353
7354 if (do_partial_inline) {
7355 assert(elem_bt != T_ILLEGAL, "sanity");
7356
7357 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
7358 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
7359 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
7360
7361 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
7362 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
7363 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
7364
7365 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
7366
7367 if (!stopped()) {
7368 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
7369
7370 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
7371 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
7372 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
7373 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
7374
7375 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
7376 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
7377 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
7378 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
7379
7380 exit_block->init_req(inline_path, control());
7381 memory_phi->init_req(inline_path, map()->memory());
7382 result_phi->init_req(inline_path, result);
7383
7384 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
7385 clear_upper_avx();
7386 }
7387 }
7388 }
7389
7390 if (call_stub_path != nullptr) {
7391 set_control(call_stub_path);
7392
7393 Node* call = make_runtime_call(RC_LEAF,
7394 OptoRuntime::vectorizedMismatch_Type(),
7395 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
7396 obja_adr, objb_adr, length, scale);
7397
7398 exit_block->init_req(stub_path, control());
7399 memory_phi->init_req(stub_path, map()->memory());
7400 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
7401 }
7402
7403 exit_block = _gvn.transform(exit_block);
7404 memory_phi = _gvn.transform(memory_phi);
7405 result_phi = _gvn.transform(result_phi);
7406
7407 record_for_igvn(exit_block);
7408 record_for_igvn(memory_phi);
7409 record_for_igvn(result_phi);
7410
7411 set_control(exit_block);
7412 set_all_memory(memory_phi);
7413 set_result(result_phi);
7414
7415 return true;
7416 }
7417
7418 //------------------------------inline_vectorizedHashcode----------------------------
7419 bool LibraryCallKit::inline_vectorizedHashCode() {
7420 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
7421
7422 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
7423 Node* array = argument(0);
7424 Node* offset = argument(1);
7425 Node* length = argument(2);
7426 Node* initialValue = argument(3);
7427 Node* basic_type = argument(4);
7428
7429 if (basic_type == top()) {
7430 return false; // failed input validation
7431 }
7432
7433 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
7434 if (!basic_type_t->is_con()) {
7435 return false; // Only intrinsify if mode argument is constant
7436 }
7437
7438 array = must_be_not_null(array, true);
7439
7440 BasicType bt = (BasicType)basic_type_t->get_con();
7441
7442 // Resolve address of first element
7443 Node* array_start = array_element_address(array, offset, bt);
7444
7445 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
7446 array_start, length, initialValue, basic_type)));
7447 clear_upper_avx();
7448
7449 return true;
7450 }
7451
7452 /**
7453 * Calculate CRC32 for byte.
7454 * int java.util.zip.CRC32.update(int crc, int b)
7455 */
7456 bool LibraryCallKit::inline_updateCRC32() {
7457 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7458 assert(callee()->signature()->size() == 2, "update has 2 parameters");
7459 // no receiver since it is static method
7460 Node* crc = argument(0); // type: int
7461 Node* b = argument(1); // type: int
7462
7463 /*
7464 * int c = ~ crc;
7465 * b = timesXtoThe32[(b ^ c) & 0xFF];
7466 * b = b ^ (c >>> 8);
7467 * crc = ~b;
7468 */
7469
7470 Node* M1 = intcon(-1);
7471 crc = _gvn.transform(new XorINode(crc, M1));
7472 Node* result = _gvn.transform(new XorINode(crc, b));
7473 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
7474
7475 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
7476 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
7477 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
7478 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
7479
7480 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
7481 result = _gvn.transform(new XorINode(crc, result));
7482 result = _gvn.transform(new XorINode(result, M1));
7483 set_result(result);
7484 return true;
7485 }
7486
7487 /**
7488 * Calculate CRC32 for byte[] array.
7489 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
7490 */
7491 bool LibraryCallKit::inline_updateBytesCRC32() {
7492 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7493 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7494 // no receiver since it is static method
7495 Node* crc = argument(0); // type: int
7496 Node* src = argument(1); // type: oop
7497 Node* offset = argument(2); // type: int
7498 Node* length = argument(3); // type: int
7499
7500 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7501 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7502 // failed array check
7503 return false;
7504 }
7505
7506 // Figure out the size and type of the elements we will be copying.
7507 BasicType src_elem = src_type->elem()->array_element_basic_type();
7508 if (src_elem != T_BYTE) {
7509 return false;
7510 }
7511
7512 // 'src_start' points to src array + scaled offset
7513 src = must_be_not_null(src, true);
7514 Node* src_start = array_element_address(src, offset, src_elem);
7515
7516 // We assume that range check is done by caller.
7517 // TODO: generate range check (offset+length < src.length) in debug VM.
7518
7519 // Call the stub.
7520 address stubAddr = StubRoutines::updateBytesCRC32();
7521 const char *stubName = "updateBytesCRC32";
7522
7523 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7524 stubAddr, stubName, TypePtr::BOTTOM,
7525 crc, src_start, length);
7526 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7527 set_result(result);
7528 return true;
7529 }
7530
7531 /**
7532 * Calculate CRC32 for ByteBuffer.
7533 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
7534 */
7535 bool LibraryCallKit::inline_updateByteBufferCRC32() {
7536 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
7537 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7538 // no receiver since it is static method
7539 Node* crc = argument(0); // type: int
7540 Node* src = argument(1); // type: long
7541 Node* offset = argument(3); // type: int
7542 Node* length = argument(4); // type: int
7543
7544 src = ConvL2X(src); // adjust Java long to machine word
7545 Node* base = _gvn.transform(new CastX2PNode(src));
7546 offset = ConvI2X(offset);
7547
7548 // 'src_start' points to src array + scaled offset
7549 Node* src_start = basic_plus_adr(top(), base, offset);
7550
7551 // Call the stub.
7552 address stubAddr = StubRoutines::updateBytesCRC32();
7553 const char *stubName = "updateBytesCRC32";
7554
7555 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
7556 stubAddr, stubName, TypePtr::BOTTOM,
7557 crc, src_start, length);
7558 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7559 set_result(result);
7560 return true;
7561 }
7562
7563 //------------------------------get_table_from_crc32c_class-----------------------
7564 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
7565 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
7566 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
7567
7568 return table;
7569 }
7570
7571 //------------------------------inline_updateBytesCRC32C-----------------------
7572 //
7573 // Calculate CRC32C for byte[] array.
7574 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
7575 //
7576 bool LibraryCallKit::inline_updateBytesCRC32C() {
7577 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7578 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7579 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7580 // no receiver since it is a 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* end = argument(3); // type: int
7585
7586 Node* length = _gvn.transform(new SubINode(end, offset));
7587
7588 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7589 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7590 // failed array check
7591 return false;
7592 }
7593
7594 // Figure out the size and type of the elements we will be copying.
7595 BasicType src_elem = src_type->elem()->array_element_basic_type();
7596 if (src_elem != T_BYTE) {
7597 return false;
7598 }
7599
7600 // 'src_start' points to src array + scaled offset
7601 src = must_be_not_null(src, true);
7602 Node* src_start = array_element_address(src, offset, src_elem);
7603
7604 // static final int[] byteTable in class CRC32C
7605 Node* table = get_table_from_crc32c_class(callee()->holder());
7606 table = must_be_not_null(table, true);
7607 Node* table_start = array_element_address(table, intcon(0), T_INT);
7608
7609 // We assume that range check is done by caller.
7610 // TODO: generate range check (offset+length < src.length) in debug VM.
7611
7612 // Call the stub.
7613 address stubAddr = StubRoutines::updateBytesCRC32C();
7614 const char *stubName = "updateBytesCRC32C";
7615
7616 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7617 stubAddr, stubName, TypePtr::BOTTOM,
7618 crc, src_start, length, table_start);
7619 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7620 set_result(result);
7621 return true;
7622 }
7623
7624 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
7625 //
7626 // Calculate CRC32C for DirectByteBuffer.
7627 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
7628 //
7629 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
7630 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
7631 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
7632 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
7633 // no receiver since it is a static method
7634 Node* crc = argument(0); // type: int
7635 Node* src = argument(1); // type: long
7636 Node* offset = argument(3); // type: int
7637 Node* end = argument(4); // type: int
7638
7639 Node* length = _gvn.transform(new SubINode(end, offset));
7640
7641 src = ConvL2X(src); // adjust Java long to machine word
7642 Node* base = _gvn.transform(new CastX2PNode(src));
7643 offset = ConvI2X(offset);
7644
7645 // 'src_start' points to src array + scaled offset
7646 Node* src_start = basic_plus_adr(top(), base, offset);
7647
7648 // static final int[] byteTable in class CRC32C
7649 Node* table = get_table_from_crc32c_class(callee()->holder());
7650 table = must_be_not_null(table, true);
7651 Node* table_start = array_element_address(table, intcon(0), T_INT);
7652
7653 // Call the stub.
7654 address stubAddr = StubRoutines::updateBytesCRC32C();
7655 const char *stubName = "updateBytesCRC32C";
7656
7657 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
7658 stubAddr, stubName, TypePtr::BOTTOM,
7659 crc, src_start, length, table_start);
7660 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7661 set_result(result);
7662 return true;
7663 }
7664
7665 //------------------------------inline_updateBytesAdler32----------------------
7666 //
7667 // Calculate Adler32 checksum for byte[] array.
7668 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
7669 //
7670 bool LibraryCallKit::inline_updateBytesAdler32() {
7671 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7672 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
7673 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7674 // no receiver since it is static method
7675 Node* crc = argument(0); // type: int
7676 Node* src = argument(1); // type: oop
7677 Node* offset = argument(2); // type: int
7678 Node* length = argument(3); // type: int
7679
7680 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7681 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
7682 // failed array check
7683 return false;
7684 }
7685
7686 // Figure out the size and type of the elements we will be copying.
7687 BasicType src_elem = src_type->elem()->array_element_basic_type();
7688 if (src_elem != T_BYTE) {
7689 return false;
7690 }
7691
7692 // 'src_start' points to src array + scaled offset
7693 Node* src_start = array_element_address(src, offset, src_elem);
7694
7695 // We assume that range check is done by caller.
7696 // TODO: generate range check (offset+length < src.length) in debug VM.
7697
7698 // Call the stub.
7699 address stubAddr = StubRoutines::updateBytesAdler32();
7700 const char *stubName = "updateBytesAdler32";
7701
7702 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7703 stubAddr, stubName, TypePtr::BOTTOM,
7704 crc, src_start, length);
7705 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7706 set_result(result);
7707 return true;
7708 }
7709
7710 //------------------------------inline_updateByteBufferAdler32---------------
7711 //
7712 // Calculate Adler32 checksum for DirectByteBuffer.
7713 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7714 //
7715 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7716 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7717 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7718 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7719 // no receiver since it is static method
7720 Node* crc = argument(0); // type: int
7721 Node* src = argument(1); // type: long
7722 Node* offset = argument(3); // type: int
7723 Node* length = argument(4); // type: int
7724
7725 src = ConvL2X(src); // adjust Java long to machine word
7726 Node* base = _gvn.transform(new CastX2PNode(src));
7727 offset = ConvI2X(offset);
7728
7729 // 'src_start' points to src array + scaled offset
7730 Node* src_start = basic_plus_adr(top(), base, offset);
7731
7732 // Call the stub.
7733 address stubAddr = StubRoutines::updateBytesAdler32();
7734 const char *stubName = "updateBytesAdler32";
7735
7736 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7737 stubAddr, stubName, TypePtr::BOTTOM,
7738 crc, src_start, length);
7739
7740 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7741 set_result(result);
7742 return true;
7743 }
7744
7745 //----------------------------inline_reference_get0----------------------------
7746 // public T java.lang.ref.Reference.get();
7747 bool LibraryCallKit::inline_reference_get0() {
7748 const int referent_offset = java_lang_ref_Reference::referent_offset();
7749
7750 // Get the argument:
7751 Node* reference_obj = null_check_receiver();
7752 if (stopped()) return true;
7753
7754 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7755 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7756 decorators, /*is_static*/ false, nullptr);
7757 if (result == nullptr) return false;
7758
7759 // Add memory barrier to prevent commoning reads from this field
7760 // across safepoint since GC can change its value.
7761 insert_mem_bar(Op_MemBarCPUOrder);
7762
7763 set_result(result);
7764 return true;
7765 }
7766
7767 //----------------------------inline_reference_refersTo0----------------------------
7768 // bool java.lang.ref.Reference.refersTo0();
7769 // bool java.lang.ref.PhantomReference.refersTo0();
7770 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7771 // Get arguments:
7772 Node* reference_obj = null_check_receiver();
7773 Node* other_obj = argument(1);
7774 if (stopped()) return true;
7775
7776 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7777 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7778 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7779 decorators, /*is_static*/ false, nullptr);
7780 if (referent == nullptr) return false;
7781
7782 // Add memory barrier to prevent commoning reads from this field
7783 // across safepoint since GC can change its value.
7784 insert_mem_bar(Op_MemBarCPUOrder);
7785
7786 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7787 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7788 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7789
7790 RegionNode* region = new RegionNode(3);
7791 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7792
7793 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7794 region->init_req(1, if_true);
7795 phi->init_req(1, intcon(1));
7796
7797 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7798 region->init_req(2, if_false);
7799 phi->init_req(2, intcon(0));
7800
7801 set_control(_gvn.transform(region));
7802 record_for_igvn(region);
7803 set_result(_gvn.transform(phi));
7804 return true;
7805 }
7806
7807 //----------------------------inline_reference_clear0----------------------------
7808 // void java.lang.ref.Reference.clear0();
7809 // void java.lang.ref.PhantomReference.clear0();
7810 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7811 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7812
7813 // Get arguments
7814 Node* reference_obj = null_check_receiver();
7815 if (stopped()) return true;
7816
7817 // Common access parameters
7818 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7819 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7820 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7821 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7822 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7823
7824 Node* referent = access_load_at(reference_obj,
7825 referent_field_addr,
7826 referent_field_addr_type,
7827 val_type,
7828 T_OBJECT,
7829 decorators);
7830
7831 IdealKit ideal(this);
7832 #define __ ideal.
7833 __ if_then(referent, BoolTest::ne, null());
7834 sync_kit(ideal);
7835 access_store_at(reference_obj,
7836 referent_field_addr,
7837 referent_field_addr_type,
7838 null(),
7839 val_type,
7840 T_OBJECT,
7841 decorators);
7842 __ sync_kit(this);
7843 __ end_if();
7844 final_sync(ideal);
7845 #undef __
7846
7847 return true;
7848 }
7849
7850 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7851 DecoratorSet decorators, bool is_static,
7852 ciInstanceKlass* fromKls) {
7853 if (fromKls == nullptr) {
7854 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7855 assert(tinst != nullptr, "obj is null");
7856 assert(tinst->is_loaded(), "obj is not loaded");
7857 fromKls = tinst->instance_klass();
7858 } else {
7859 assert(is_static, "only for static field access");
7860 }
7861 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7862 ciSymbol::make(fieldTypeString),
7863 is_static);
7864
7865 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7866 if (field == nullptr) return (Node *) nullptr;
7867
7868 if (is_static) {
7869 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7870 fromObj = makecon(tip);
7871 }
7872
7873 // Next code copied from Parse::do_get_xxx():
7874
7875 // Compute address and memory type.
7876 int offset = field->offset_in_bytes();
7877 bool is_vol = field->is_volatile();
7878 ciType* field_klass = field->type();
7879 assert(field_klass->is_loaded(), "should be loaded");
7880 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7881 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7882 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7883 "slice of address and input slice don't match");
7884 BasicType bt = field->layout_type();
7885
7886 // Build the resultant type of the load
7887 const Type *type;
7888 if (bt == T_OBJECT) {
7889 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7890 } else {
7891 type = Type::get_const_basic_type(bt);
7892 }
7893
7894 if (is_vol) {
7895 decorators |= MO_SEQ_CST;
7896 }
7897
7898 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7899 }
7900
7901 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7902 bool is_exact /* true */, bool is_static /* false */,
7903 ciInstanceKlass * fromKls /* nullptr */) {
7904 if (fromKls == nullptr) {
7905 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7906 assert(tinst != nullptr, "obj is null");
7907 assert(tinst->is_loaded(), "obj is not loaded");
7908 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7909 fromKls = tinst->instance_klass();
7910 }
7911 else {
7912 assert(is_static, "only for static field access");
7913 }
7914 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7915 ciSymbol::make(fieldTypeString),
7916 is_static);
7917
7918 assert(field != nullptr, "undefined field");
7919 assert(!field->is_volatile(), "not defined for volatile fields");
7920
7921 if (is_static) {
7922 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7923 fromObj = makecon(tip);
7924 }
7925
7926 // Next code copied from Parse::do_get_xxx():
7927
7928 // Compute address and memory type.
7929 int offset = field->offset_in_bytes();
7930 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7931
7932 return adr;
7933 }
7934
7935 //------------------------------inline_aescrypt_Block-----------------------
7936 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7937 address stubAddr = nullptr;
7938 const char *stubName;
7939 assert(UseAES, "need AES instruction support");
7940
7941 switch(id) {
7942 case vmIntrinsics::_aescrypt_encryptBlock:
7943 stubAddr = StubRoutines::aescrypt_encryptBlock();
7944 stubName = "aescrypt_encryptBlock";
7945 break;
7946 case vmIntrinsics::_aescrypt_decryptBlock:
7947 stubAddr = StubRoutines::aescrypt_decryptBlock();
7948 stubName = "aescrypt_decryptBlock";
7949 break;
7950 default:
7951 break;
7952 }
7953 if (stubAddr == nullptr) return false;
7954
7955 Node* aescrypt_object = argument(0);
7956 Node* src = argument(1);
7957 Node* src_offset = argument(2);
7958 Node* dest = argument(3);
7959 Node* dest_offset = argument(4);
7960
7961 src = must_be_not_null(src, true);
7962 dest = must_be_not_null(dest, true);
7963
7964 // (1) src and dest are arrays.
7965 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7966 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7967 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7968 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7969
7970 // for the quick and dirty code we will skip all the checks.
7971 // we are just trying to get the call to be generated.
7972 Node* src_start = src;
7973 Node* dest_start = dest;
7974 if (src_offset != nullptr || dest_offset != nullptr) {
7975 assert(src_offset != nullptr && dest_offset != nullptr, "");
7976 src_start = array_element_address(src, src_offset, T_BYTE);
7977 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7978 }
7979
7980 // now need to get the start of its expanded key array
7981 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7982 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7983 if (k_start == nullptr) return false;
7984
7985 // Call the stub.
7986 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7987 stubAddr, stubName, TypePtr::BOTTOM,
7988 src_start, dest_start, k_start);
7989
7990 return true;
7991 }
7992
7993 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7994 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7995 address stubAddr = nullptr;
7996 const char *stubName = nullptr;
7997
7998 assert(UseAES, "need AES instruction support");
7999
8000 switch(id) {
8001 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
8002 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
8003 stubName = "cipherBlockChaining_encryptAESCrypt";
8004 break;
8005 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
8006 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
8007 stubName = "cipherBlockChaining_decryptAESCrypt";
8008 break;
8009 default:
8010 break;
8011 }
8012 if (stubAddr == nullptr) return false;
8013
8014 Node* cipherBlockChaining_object = argument(0);
8015 Node* src = argument(1);
8016 Node* src_offset = argument(2);
8017 Node* len = argument(3);
8018 Node* dest = argument(4);
8019 Node* dest_offset = argument(5);
8020
8021 src = must_be_not_null(src, false);
8022 dest = must_be_not_null(dest, false);
8023
8024 // (1) src and dest are arrays.
8025 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8026 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8027 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8028 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8029
8030 // checks are the responsibility of the caller
8031 Node* src_start = src;
8032 Node* dest_start = dest;
8033 if (src_offset != nullptr || dest_offset != nullptr) {
8034 assert(src_offset != nullptr && dest_offset != nullptr, "");
8035 src_start = array_element_address(src, src_offset, T_BYTE);
8036 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8037 }
8038
8039 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8040 // (because of the predicated logic executed earlier).
8041 // so we cast it here safely.
8042 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8043
8044 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8045 if (embeddedCipherObj == nullptr) return false;
8046
8047 // cast it to what we know it will be at runtime
8048 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
8049 assert(tinst != nullptr, "CBC obj is null");
8050 assert(tinst->is_loaded(), "CBC obj is not loaded");
8051 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8052 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8053
8054 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8055 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8056 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8057 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8058 aescrypt_object = _gvn.transform(aescrypt_object);
8059
8060 // we need to get the start of the aescrypt_object's expanded key array
8061 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8062 if (k_start == nullptr) return false;
8063
8064 // similarly, get the start address of the r vector
8065 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
8066 if (objRvec == nullptr) return false;
8067 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
8068
8069 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8070 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8071 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
8072 stubAddr, stubName, TypePtr::BOTTOM,
8073 src_start, dest_start, k_start, r_start, len);
8074
8075 // return cipher length (int)
8076 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
8077 set_result(retvalue);
8078 return true;
8079 }
8080
8081 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
8082 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
8083 address stubAddr = nullptr;
8084 const char *stubName = nullptr;
8085
8086 assert(UseAES, "need AES instruction support");
8087
8088 switch (id) {
8089 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
8090 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
8091 stubName = "electronicCodeBook_encryptAESCrypt";
8092 break;
8093 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
8094 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
8095 stubName = "electronicCodeBook_decryptAESCrypt";
8096 break;
8097 default:
8098 break;
8099 }
8100
8101 if (stubAddr == nullptr) return false;
8102
8103 Node* electronicCodeBook_object = argument(0);
8104 Node* src = argument(1);
8105 Node* src_offset = argument(2);
8106 Node* len = argument(3);
8107 Node* dest = argument(4);
8108 Node* dest_offset = argument(5);
8109
8110 // (1) src and dest are arrays.
8111 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8112 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8113 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8114 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8115
8116 // checks are the responsibility of the caller
8117 Node* src_start = src;
8118 Node* dest_start = dest;
8119 if (src_offset != nullptr || dest_offset != nullptr) {
8120 assert(src_offset != nullptr && dest_offset != nullptr, "");
8121 src_start = array_element_address(src, src_offset, T_BYTE);
8122 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8123 }
8124
8125 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8126 // (because of the predicated logic executed earlier).
8127 // so we cast it here safely.
8128 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8129
8130 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8131 if (embeddedCipherObj == nullptr) return false;
8132
8133 // cast it to what we know it will be at runtime
8134 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
8135 assert(tinst != nullptr, "ECB obj is null");
8136 assert(tinst->is_loaded(), "ECB obj is not loaded");
8137 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8138 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8139
8140 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8141 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8142 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8143 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8144 aescrypt_object = _gvn.transform(aescrypt_object);
8145
8146 // we need to get the start of the aescrypt_object's expanded key array
8147 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8148 if (k_start == nullptr) return false;
8149
8150 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8151 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
8152 OptoRuntime::electronicCodeBook_aescrypt_Type(),
8153 stubAddr, stubName, TypePtr::BOTTOM,
8154 src_start, dest_start, k_start, len);
8155
8156 // return cipher length (int)
8157 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
8158 set_result(retvalue);
8159 return true;
8160 }
8161
8162 //------------------------------inline_counterMode_AESCrypt-----------------------
8163 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
8164 assert(UseAES, "need AES instruction support");
8165 if (!UseAESCTRIntrinsics) return false;
8166
8167 address stubAddr = nullptr;
8168 const char *stubName = nullptr;
8169 if (id == vmIntrinsics::_counterMode_AESCrypt) {
8170 stubAddr = StubRoutines::counterMode_AESCrypt();
8171 stubName = "counterMode_AESCrypt";
8172 }
8173 if (stubAddr == nullptr) return false;
8174
8175 Node* counterMode_object = argument(0);
8176 Node* src = argument(1);
8177 Node* src_offset = argument(2);
8178 Node* len = argument(3);
8179 Node* dest = argument(4);
8180 Node* dest_offset = argument(5);
8181
8182 // (1) src and dest are arrays.
8183 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8184 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
8185 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
8186 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
8187
8188 // checks are the responsibility of the caller
8189 Node* src_start = src;
8190 Node* dest_start = dest;
8191 if (src_offset != nullptr || dest_offset != nullptr) {
8192 assert(src_offset != nullptr && dest_offset != nullptr, "");
8193 src_start = array_element_address(src, src_offset, T_BYTE);
8194 dest_start = array_element_address(dest, dest_offset, T_BYTE);
8195 }
8196
8197 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8198 // (because of the predicated logic executed earlier).
8199 // so we cast it here safely.
8200 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8201 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8202 if (embeddedCipherObj == nullptr) return false;
8203 // cast it to what we know it will be at runtime
8204 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
8205 assert(tinst != nullptr, "CTR obj is null");
8206 assert(tinst->is_loaded(), "CTR obj is not loaded");
8207 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8208 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8209 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8210 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8211 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8212 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8213 aescrypt_object = _gvn.transform(aescrypt_object);
8214 // we need to get the start of the aescrypt_object's expanded key array
8215 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8216 if (k_start == nullptr) return false;
8217 // similarly, get the start address of the r vector
8218 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
8219 if (obj_counter == nullptr) return false;
8220 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
8221
8222 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
8223 if (saved_encCounter == nullptr) return false;
8224 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
8225 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
8226
8227 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
8228 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8229 OptoRuntime::counterMode_aescrypt_Type(),
8230 stubAddr, stubName, TypePtr::BOTTOM,
8231 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
8232
8233 // return cipher length (int)
8234 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
8235 set_result(retvalue);
8236 return true;
8237 }
8238
8239 //------------------------------get_key_start_from_aescrypt_object-----------------------
8240 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
8241 #if defined(PPC64) || defined(S390) || defined(RISCV64)
8242 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
8243 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
8244 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
8245 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
8246 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
8247 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8248 if (objSessionK == nullptr) {
8249 return (Node *) nullptr;
8250 }
8251 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
8252 #else
8253 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
8254 #endif // PPC64
8255 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
8256 if (objAESCryptKey == nullptr) return (Node *) nullptr;
8257
8258 // now have the array, need to get the start address of the K array
8259 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
8260 return k_start;
8261 }
8262
8263 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
8264 // Return node representing slow path of predicate check.
8265 // the pseudo code we want to emulate with this predicate is:
8266 // for encryption:
8267 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8268 // for decryption:
8269 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8270 // note cipher==plain is more conservative than the original java code but that's OK
8271 //
8272 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
8273 // The receiver was checked for null already.
8274 Node* objCBC = argument(0);
8275
8276 Node* src = argument(1);
8277 Node* dest = argument(4);
8278
8279 // Load embeddedCipher field of CipherBlockChaining object.
8280 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8281
8282 // get AESCrypt klass for instanceOf check
8283 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8284 // will have same classloader as CipherBlockChaining object
8285 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
8286 assert(tinst != nullptr, "CBCobj is null");
8287 assert(tinst->is_loaded(), "CBCobj is not loaded");
8288
8289 // we want to do an instanceof comparison against the AESCrypt class
8290 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8291 if (!klass_AESCrypt->is_loaded()) {
8292 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8293 Node* ctrl = control();
8294 set_control(top()); // no regular fast path
8295 return ctrl;
8296 }
8297
8298 src = must_be_not_null(src, true);
8299 dest = must_be_not_null(dest, true);
8300
8301 // Resolve oops to stable for CmpP below.
8302 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8303
8304 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8305 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8306 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8307
8308 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8309
8310 // for encryption, we are done
8311 if (!decrypting)
8312 return instof_false; // even if it is null
8313
8314 // for decryption, we need to add a further check to avoid
8315 // taking the intrinsic path when cipher and plain are the same
8316 // see the original java code for why.
8317 RegionNode* region = new RegionNode(3);
8318 region->init_req(1, instof_false);
8319
8320 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8321 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8322 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8323 region->init_req(2, src_dest_conjoint);
8324
8325 record_for_igvn(region);
8326 return _gvn.transform(region);
8327 }
8328
8329 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
8330 // Return node representing slow path of predicate check.
8331 // the pseudo code we want to emulate with this predicate is:
8332 // for encryption:
8333 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8334 // for decryption:
8335 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8336 // note cipher==plain is more conservative than the original java code but that's OK
8337 //
8338 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
8339 // The receiver was checked for null already.
8340 Node* objECB = argument(0);
8341
8342 // Load embeddedCipher field of ElectronicCodeBook object.
8343 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8344
8345 // get AESCrypt klass for instanceOf check
8346 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8347 // will have same classloader as ElectronicCodeBook object
8348 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
8349 assert(tinst != nullptr, "ECBobj is null");
8350 assert(tinst->is_loaded(), "ECBobj is not loaded");
8351
8352 // we want to do an instanceof comparison against the AESCrypt class
8353 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8354 if (!klass_AESCrypt->is_loaded()) {
8355 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8356 Node* ctrl = control();
8357 set_control(top()); // no regular fast path
8358 return ctrl;
8359 }
8360 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8361
8362 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8363 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8364 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8365
8366 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8367
8368 // for encryption, we are done
8369 if (!decrypting)
8370 return instof_false; // even if it is null
8371
8372 // for decryption, we need to add a further check to avoid
8373 // taking the intrinsic path when cipher and plain are the same
8374 // see the original java code for why.
8375 RegionNode* region = new RegionNode(3);
8376 region->init_req(1, instof_false);
8377 Node* src = argument(1);
8378 Node* dest = argument(4);
8379 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
8380 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
8381 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
8382 region->init_req(2, src_dest_conjoint);
8383
8384 record_for_igvn(region);
8385 return _gvn.transform(region);
8386 }
8387
8388 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
8389 // Return node representing slow path of predicate check.
8390 // the pseudo code we want to emulate with this predicate is:
8391 // for encryption:
8392 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8393 // for decryption:
8394 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8395 // note cipher==plain is more conservative than the original java code but that's OK
8396 //
8397
8398 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
8399 // The receiver was checked for null already.
8400 Node* objCTR = argument(0);
8401
8402 // Load embeddedCipher field of CipherBlockChaining object.
8403 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8404
8405 // get AESCrypt klass for instanceOf check
8406 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8407 // will have same classloader as CipherBlockChaining object
8408 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
8409 assert(tinst != nullptr, "CTRobj is null");
8410 assert(tinst->is_loaded(), "CTRobj is not loaded");
8411
8412 // we want to do an instanceof comparison against the AESCrypt class
8413 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8414 if (!klass_AESCrypt->is_loaded()) {
8415 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8416 Node* ctrl = control();
8417 set_control(top()); // no regular fast path
8418 return ctrl;
8419 }
8420
8421 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8422 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8423 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8424 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8425 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8426
8427 return instof_false; // even if it is null
8428 }
8429
8430 //------------------------------inline_ghash_processBlocks
8431 bool LibraryCallKit::inline_ghash_processBlocks() {
8432 address stubAddr;
8433 const char *stubName;
8434 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
8435
8436 stubAddr = StubRoutines::ghash_processBlocks();
8437 stubName = "ghash_processBlocks";
8438
8439 Node* data = argument(0);
8440 Node* offset = argument(1);
8441 Node* len = argument(2);
8442 Node* state = argument(3);
8443 Node* subkeyH = argument(4);
8444
8445 state = must_be_not_null(state, true);
8446 subkeyH = must_be_not_null(subkeyH, true);
8447 data = must_be_not_null(data, true);
8448
8449 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8450 assert(state_start, "state is null");
8451 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
8452 assert(subkeyH_start, "subkeyH is null");
8453 Node* data_start = array_element_address(data, offset, T_BYTE);
8454 assert(data_start, "data is null");
8455
8456 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
8457 OptoRuntime::ghash_processBlocks_Type(),
8458 stubAddr, stubName, TypePtr::BOTTOM,
8459 state_start, subkeyH_start, data_start, len);
8460 return true;
8461 }
8462
8463 //------------------------------inline_chacha20Block
8464 bool LibraryCallKit::inline_chacha20Block() {
8465 address stubAddr;
8466 const char *stubName;
8467 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
8468
8469 stubAddr = StubRoutines::chacha20Block();
8470 stubName = "chacha20Block";
8471
8472 Node* state = argument(0);
8473 Node* result = argument(1);
8474
8475 state = must_be_not_null(state, true);
8476 result = must_be_not_null(result, true);
8477
8478 Node* state_start = array_element_address(state, intcon(0), T_INT);
8479 assert(state_start, "state is null");
8480 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
8481 assert(result_start, "result is null");
8482
8483 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
8484 OptoRuntime::chacha20Block_Type(),
8485 stubAddr, stubName, TypePtr::BOTTOM,
8486 state_start, result_start);
8487 // return key stream length (int)
8488 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
8489 set_result(retvalue);
8490 return true;
8491 }
8492
8493 //------------------------------inline_kyberNtt
8494 bool LibraryCallKit::inline_kyberNtt() {
8495 address stubAddr;
8496 const char *stubName;
8497 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8498 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
8499
8500 stubAddr = StubRoutines::kyberNtt();
8501 stubName = "kyberNtt";
8502 if (!stubAddr) return false;
8503
8504 Node* coeffs = argument(0);
8505 Node* ntt_zetas = argument(1);
8506
8507 coeffs = must_be_not_null(coeffs, true);
8508 ntt_zetas = must_be_not_null(ntt_zetas, true);
8509
8510 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8511 assert(coeffs_start, "coeffs is null");
8512 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
8513 assert(ntt_zetas_start, "ntt_zetas is null");
8514 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8515 OptoRuntime::kyberNtt_Type(),
8516 stubAddr, stubName, TypePtr::BOTTOM,
8517 coeffs_start, ntt_zetas_start);
8518 // return an int
8519 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
8520 set_result(retvalue);
8521 return true;
8522 }
8523
8524 //------------------------------inline_kyberInverseNtt
8525 bool LibraryCallKit::inline_kyberInverseNtt() {
8526 address stubAddr;
8527 const char *stubName;
8528 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8529 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
8530
8531 stubAddr = StubRoutines::kyberInverseNtt();
8532 stubName = "kyberInverseNtt";
8533 if (!stubAddr) return false;
8534
8535 Node* coeffs = argument(0);
8536 Node* zetas = argument(1);
8537
8538 coeffs = must_be_not_null(coeffs, true);
8539 zetas = must_be_not_null(zetas, true);
8540
8541 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8542 assert(coeffs_start, "coeffs is null");
8543 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8544 assert(zetas_start, "inverseNtt_zetas is null");
8545 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8546 OptoRuntime::kyberInverseNtt_Type(),
8547 stubAddr, stubName, TypePtr::BOTTOM,
8548 coeffs_start, zetas_start);
8549
8550 // return an int
8551 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
8552 set_result(retvalue);
8553 return true;
8554 }
8555
8556 //------------------------------inline_kyberNttMult
8557 bool LibraryCallKit::inline_kyberNttMult() {
8558 address stubAddr;
8559 const char *stubName;
8560 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8561 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
8562
8563 stubAddr = StubRoutines::kyberNttMult();
8564 stubName = "kyberNttMult";
8565 if (!stubAddr) return false;
8566
8567 Node* result = argument(0);
8568 Node* ntta = argument(1);
8569 Node* nttb = argument(2);
8570 Node* zetas = argument(3);
8571
8572 result = must_be_not_null(result, true);
8573 ntta = must_be_not_null(ntta, true);
8574 nttb = must_be_not_null(nttb, true);
8575 zetas = must_be_not_null(zetas, true);
8576
8577 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8578 assert(result_start, "result is null");
8579 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
8580 assert(ntta_start, "ntta is null");
8581 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
8582 assert(nttb_start, "nttb is null");
8583 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
8584 assert(zetas_start, "nttMult_zetas is null");
8585 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8586 OptoRuntime::kyberNttMult_Type(),
8587 stubAddr, stubName, TypePtr::BOTTOM,
8588 result_start, ntta_start, nttb_start,
8589 zetas_start);
8590
8591 // return an int
8592 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
8593 set_result(retvalue);
8594
8595 return true;
8596 }
8597
8598 //------------------------------inline_kyberAddPoly_2
8599 bool LibraryCallKit::inline_kyberAddPoly_2() {
8600 address stubAddr;
8601 const char *stubName;
8602 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8603 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
8604
8605 stubAddr = StubRoutines::kyberAddPoly_2();
8606 stubName = "kyberAddPoly_2";
8607 if (!stubAddr) return false;
8608
8609 Node* result = argument(0);
8610 Node* a = argument(1);
8611 Node* b = argument(2);
8612
8613 result = must_be_not_null(result, true);
8614 a = must_be_not_null(a, true);
8615 b = must_be_not_null(b, true);
8616
8617 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8618 assert(result_start, "result is null");
8619 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8620 assert(a_start, "a is null");
8621 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8622 assert(b_start, "b is null");
8623 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
8624 OptoRuntime::kyberAddPoly_2_Type(),
8625 stubAddr, stubName, TypePtr::BOTTOM,
8626 result_start, a_start, b_start);
8627 // return an int
8628 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
8629 set_result(retvalue);
8630 return true;
8631 }
8632
8633 //------------------------------inline_kyberAddPoly_3
8634 bool LibraryCallKit::inline_kyberAddPoly_3() {
8635 address stubAddr;
8636 const char *stubName;
8637 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8638 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
8639
8640 stubAddr = StubRoutines::kyberAddPoly_3();
8641 stubName = "kyberAddPoly_3";
8642 if (!stubAddr) return false;
8643
8644 Node* result = argument(0);
8645 Node* a = argument(1);
8646 Node* b = argument(2);
8647 Node* c = argument(3);
8648
8649 result = must_be_not_null(result, true);
8650 a = must_be_not_null(a, true);
8651 b = must_be_not_null(b, true);
8652 c = must_be_not_null(c, true);
8653
8654 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
8655 assert(result_start, "result is null");
8656 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
8657 assert(a_start, "a is null");
8658 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
8659 assert(b_start, "b is null");
8660 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
8661 assert(c_start, "c is null");
8662 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
8663 OptoRuntime::kyberAddPoly_3_Type(),
8664 stubAddr, stubName, TypePtr::BOTTOM,
8665 result_start, a_start, b_start, c_start);
8666 // return an int
8667 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
8668 set_result(retvalue);
8669 return true;
8670 }
8671
8672 //------------------------------inline_kyber12To16
8673 bool LibraryCallKit::inline_kyber12To16() {
8674 address stubAddr;
8675 const char *stubName;
8676 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8677 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
8678
8679 stubAddr = StubRoutines::kyber12To16();
8680 stubName = "kyber12To16";
8681 if (!stubAddr) return false;
8682
8683 Node* condensed = argument(0);
8684 Node* condensedOffs = argument(1);
8685 Node* parsed = argument(2);
8686 Node* parsedLength = argument(3);
8687
8688 condensed = must_be_not_null(condensed, true);
8689 parsed = must_be_not_null(parsed, true);
8690
8691 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8692 assert(condensed_start, "condensed is null");
8693 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8694 assert(parsed_start, "parsed is null");
8695 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8696 OptoRuntime::kyber12To16_Type(),
8697 stubAddr, stubName, TypePtr::BOTTOM,
8698 condensed_start, condensedOffs, parsed_start, parsedLength);
8699 // return an int
8700 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8701 set_result(retvalue);
8702 return true;
8703
8704 }
8705
8706 //------------------------------inline_kyberBarrettReduce
8707 bool LibraryCallKit::inline_kyberBarrettReduce() {
8708 address stubAddr;
8709 const char *stubName;
8710 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8711 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8712
8713 stubAddr = StubRoutines::kyberBarrettReduce();
8714 stubName = "kyberBarrettReduce";
8715 if (!stubAddr) return false;
8716
8717 Node* coeffs = argument(0);
8718
8719 coeffs = must_be_not_null(coeffs, true);
8720
8721 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8722 assert(coeffs_start, "coeffs is null");
8723 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8724 OptoRuntime::kyberBarrettReduce_Type(),
8725 stubAddr, stubName, TypePtr::BOTTOM,
8726 coeffs_start);
8727 // return an int
8728 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8729 set_result(retvalue);
8730 return true;
8731 }
8732
8733 //------------------------------inline_dilithiumAlmostNtt
8734 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8735 address stubAddr;
8736 const char *stubName;
8737 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8738 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8739
8740 stubAddr = StubRoutines::dilithiumAlmostNtt();
8741 stubName = "dilithiumAlmostNtt";
8742 if (!stubAddr) return false;
8743
8744 Node* coeffs = argument(0);
8745 Node* ntt_zetas = argument(1);
8746
8747 coeffs = must_be_not_null(coeffs, true);
8748 ntt_zetas = must_be_not_null(ntt_zetas, true);
8749
8750 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8751 assert(coeffs_start, "coeffs is null");
8752 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8753 assert(ntt_zetas_start, "ntt_zetas is null");
8754 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8755 OptoRuntime::dilithiumAlmostNtt_Type(),
8756 stubAddr, stubName, TypePtr::BOTTOM,
8757 coeffs_start, ntt_zetas_start);
8758 // return an int
8759 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8760 set_result(retvalue);
8761 return true;
8762 }
8763
8764 //------------------------------inline_dilithiumAlmostInverseNtt
8765 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8766 address stubAddr;
8767 const char *stubName;
8768 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8769 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8770
8771 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8772 stubName = "dilithiumAlmostInverseNtt";
8773 if (!stubAddr) return false;
8774
8775 Node* coeffs = argument(0);
8776 Node* zetas = argument(1);
8777
8778 coeffs = must_be_not_null(coeffs, true);
8779 zetas = must_be_not_null(zetas, true);
8780
8781 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8782 assert(coeffs_start, "coeffs is null");
8783 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8784 assert(zetas_start, "inverseNtt_zetas is null");
8785 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8786 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8787 stubAddr, stubName, TypePtr::BOTTOM,
8788 coeffs_start, zetas_start);
8789 // return an int
8790 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8791 set_result(retvalue);
8792 return true;
8793 }
8794
8795 //------------------------------inline_dilithiumNttMult
8796 bool LibraryCallKit::inline_dilithiumNttMult() {
8797 address stubAddr;
8798 const char *stubName;
8799 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8800 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8801
8802 stubAddr = StubRoutines::dilithiumNttMult();
8803 stubName = "dilithiumNttMult";
8804 if (!stubAddr) return false;
8805
8806 Node* result = argument(0);
8807 Node* ntta = argument(1);
8808 Node* nttb = argument(2);
8809 Node* zetas = argument(3);
8810
8811 result = must_be_not_null(result, true);
8812 ntta = must_be_not_null(ntta, true);
8813 nttb = must_be_not_null(nttb, true);
8814 zetas = must_be_not_null(zetas, true);
8815
8816 Node* result_start = array_element_address(result, intcon(0), T_INT);
8817 assert(result_start, "result is null");
8818 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8819 assert(ntta_start, "ntta is null");
8820 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8821 assert(nttb_start, "nttb is null");
8822 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8823 OptoRuntime::dilithiumNttMult_Type(),
8824 stubAddr, stubName, TypePtr::BOTTOM,
8825 result_start, ntta_start, nttb_start);
8826
8827 // return an int
8828 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8829 set_result(retvalue);
8830
8831 return true;
8832 }
8833
8834 //------------------------------inline_dilithiumMontMulByConstant
8835 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8836 address stubAddr;
8837 const char *stubName;
8838 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8839 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8840
8841 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8842 stubName = "dilithiumMontMulByConstant";
8843 if (!stubAddr) return false;
8844
8845 Node* coeffs = argument(0);
8846 Node* constant = argument(1);
8847
8848 coeffs = must_be_not_null(coeffs, true);
8849
8850 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8851 assert(coeffs_start, "coeffs is null");
8852 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8853 OptoRuntime::dilithiumMontMulByConstant_Type(),
8854 stubAddr, stubName, TypePtr::BOTTOM,
8855 coeffs_start, constant);
8856
8857 // return an int
8858 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8859 set_result(retvalue);
8860 return true;
8861 }
8862
8863
8864 //------------------------------inline_dilithiumDecomposePoly
8865 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8866 address stubAddr;
8867 const char *stubName;
8868 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8869 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8870
8871 stubAddr = StubRoutines::dilithiumDecomposePoly();
8872 stubName = "dilithiumDecomposePoly";
8873 if (!stubAddr) return false;
8874
8875 Node* input = argument(0);
8876 Node* lowPart = argument(1);
8877 Node* highPart = argument(2);
8878 Node* twoGamma2 = argument(3);
8879 Node* multiplier = argument(4);
8880
8881 input = must_be_not_null(input, true);
8882 lowPart = must_be_not_null(lowPart, true);
8883 highPart = must_be_not_null(highPart, true);
8884
8885 Node* input_start = array_element_address(input, intcon(0), T_INT);
8886 assert(input_start, "input is null");
8887 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8888 assert(lowPart_start, "lowPart is null");
8889 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8890 assert(highPart_start, "highPart is null");
8891
8892 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8893 OptoRuntime::dilithiumDecomposePoly_Type(),
8894 stubAddr, stubName, TypePtr::BOTTOM,
8895 input_start, lowPart_start, highPart_start,
8896 twoGamma2, multiplier);
8897
8898 // return an int
8899 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8900 set_result(retvalue);
8901 return true;
8902 }
8903
8904 bool LibraryCallKit::inline_base64_encodeBlock() {
8905 address stubAddr;
8906 const char *stubName;
8907 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8908 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8909 stubAddr = StubRoutines::base64_encodeBlock();
8910 stubName = "encodeBlock";
8911
8912 if (!stubAddr) return false;
8913 Node* base64obj = argument(0);
8914 Node* src = argument(1);
8915 Node* offset = argument(2);
8916 Node* len = argument(3);
8917 Node* dest = argument(4);
8918 Node* dp = argument(5);
8919 Node* isURL = argument(6);
8920
8921 src = must_be_not_null(src, true);
8922 dest = must_be_not_null(dest, true);
8923
8924 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8925 assert(src_start, "source array is null");
8926 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8927 assert(dest_start, "destination array is null");
8928
8929 Node* base64 = make_runtime_call(RC_LEAF,
8930 OptoRuntime::base64_encodeBlock_Type(),
8931 stubAddr, stubName, TypePtr::BOTTOM,
8932 src_start, offset, len, dest_start, dp, isURL);
8933 return true;
8934 }
8935
8936 bool LibraryCallKit::inline_base64_decodeBlock() {
8937 address stubAddr;
8938 const char *stubName;
8939 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8940 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8941 stubAddr = StubRoutines::base64_decodeBlock();
8942 stubName = "decodeBlock";
8943
8944 if (!stubAddr) return false;
8945 Node* base64obj = argument(0);
8946 Node* src = argument(1);
8947 Node* src_offset = argument(2);
8948 Node* len = argument(3);
8949 Node* dest = argument(4);
8950 Node* dest_offset = argument(5);
8951 Node* isURL = argument(6);
8952 Node* isMIME = argument(7);
8953
8954 src = must_be_not_null(src, true);
8955 dest = must_be_not_null(dest, true);
8956
8957 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8958 assert(src_start, "source array is null");
8959 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8960 assert(dest_start, "destination array is null");
8961
8962 Node* call = make_runtime_call(RC_LEAF,
8963 OptoRuntime::base64_decodeBlock_Type(),
8964 stubAddr, stubName, TypePtr::BOTTOM,
8965 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8966 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8967 set_result(result);
8968 return true;
8969 }
8970
8971 bool LibraryCallKit::inline_poly1305_processBlocks() {
8972 address stubAddr;
8973 const char *stubName;
8974 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8975 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8976 stubAddr = StubRoutines::poly1305_processBlocks();
8977 stubName = "poly1305_processBlocks";
8978
8979 if (!stubAddr) return false;
8980 null_check_receiver(); // null-check receiver
8981 if (stopped()) return true;
8982
8983 Node* input = argument(1);
8984 Node* input_offset = argument(2);
8985 Node* len = argument(3);
8986 Node* alimbs = argument(4);
8987 Node* rlimbs = argument(5);
8988
8989 input = must_be_not_null(input, true);
8990 alimbs = must_be_not_null(alimbs, true);
8991 rlimbs = must_be_not_null(rlimbs, true);
8992
8993 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8994 assert(input_start, "input array is null");
8995 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8996 assert(acc_start, "acc array is null");
8997 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8998 assert(r_start, "r array is null");
8999
9000 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9001 OptoRuntime::poly1305_processBlocks_Type(),
9002 stubAddr, stubName, TypePtr::BOTTOM,
9003 input_start, len, acc_start, r_start);
9004 return true;
9005 }
9006
9007 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
9008 address stubAddr;
9009 const char *stubName;
9010 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9011 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
9012 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
9013 stubName = "intpoly_montgomeryMult_P256";
9014
9015 if (!stubAddr) return false;
9016 null_check_receiver(); // null-check receiver
9017 if (stopped()) return true;
9018
9019 Node* a = argument(1);
9020 Node* b = argument(2);
9021 Node* r = argument(3);
9022
9023 a = must_be_not_null(a, true);
9024 b = must_be_not_null(b, true);
9025 r = must_be_not_null(r, true);
9026
9027 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9028 assert(a_start, "a array is null");
9029 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9030 assert(b_start, "b array is null");
9031 Node* r_start = array_element_address(r, intcon(0), T_LONG);
9032 assert(r_start, "r array is null");
9033
9034 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9035 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
9036 stubAddr, stubName, TypePtr::BOTTOM,
9037 a_start, b_start, r_start);
9038 return true;
9039 }
9040
9041 bool LibraryCallKit::inline_intpoly_assign() {
9042 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
9043 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
9044 const char *stubName = "intpoly_assign";
9045 address stubAddr = StubRoutines::intpoly_assign();
9046 if (!stubAddr) return false;
9047
9048 Node* set = argument(0);
9049 Node* a = argument(1);
9050 Node* b = argument(2);
9051 Node* arr_length = load_array_length(a);
9052
9053 a = must_be_not_null(a, true);
9054 b = must_be_not_null(b, true);
9055
9056 Node* a_start = array_element_address(a, intcon(0), T_LONG);
9057 assert(a_start, "a array is null");
9058 Node* b_start = array_element_address(b, intcon(0), T_LONG);
9059 assert(b_start, "b array is null");
9060
9061 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
9062 OptoRuntime::intpoly_assign_Type(),
9063 stubAddr, stubName, TypePtr::BOTTOM,
9064 set, a_start, b_start, arr_length);
9065 return true;
9066 }
9067
9068 //------------------------------inline_digestBase_implCompress-----------------------
9069 //
9070 // Calculate MD5 for single-block byte[] array.
9071 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
9072 //
9073 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
9074 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
9075 //
9076 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
9077 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
9078 //
9079 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
9080 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
9081 //
9082 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
9083 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
9084 //
9085 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
9086 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
9087
9088 Node* digestBase_obj = argument(0);
9089 Node* src = argument(1); // type oop
9090 Node* ofs = argument(2); // type int
9091
9092 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9093 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9094 // failed array check
9095 return false;
9096 }
9097 // Figure out the size and type of the elements we will be copying.
9098 BasicType src_elem = src_type->elem()->array_element_basic_type();
9099 if (src_elem != T_BYTE) {
9100 return false;
9101 }
9102 // 'src_start' points to src array + offset
9103 src = must_be_not_null(src, true);
9104 Node* src_start = array_element_address(src, ofs, src_elem);
9105 Node* state = nullptr;
9106 Node* block_size = nullptr;
9107 address stubAddr;
9108 const char *stubName;
9109
9110 switch(id) {
9111 case vmIntrinsics::_md5_implCompress:
9112 assert(UseMD5Intrinsics, "need MD5 instruction support");
9113 state = get_state_from_digest_object(digestBase_obj, T_INT);
9114 stubAddr = StubRoutines::md5_implCompress();
9115 stubName = "md5_implCompress";
9116 break;
9117 case vmIntrinsics::_sha_implCompress:
9118 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
9119 state = get_state_from_digest_object(digestBase_obj, T_INT);
9120 stubAddr = StubRoutines::sha1_implCompress();
9121 stubName = "sha1_implCompress";
9122 break;
9123 case vmIntrinsics::_sha2_implCompress:
9124 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
9125 state = get_state_from_digest_object(digestBase_obj, T_INT);
9126 stubAddr = StubRoutines::sha256_implCompress();
9127 stubName = "sha256_implCompress";
9128 break;
9129 case vmIntrinsics::_sha5_implCompress:
9130 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
9131 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9132 stubAddr = StubRoutines::sha512_implCompress();
9133 stubName = "sha512_implCompress";
9134 break;
9135 case vmIntrinsics::_sha3_implCompress:
9136 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
9137 state = get_state_from_digest_object(digestBase_obj, T_LONG);
9138 stubAddr = StubRoutines::sha3_implCompress();
9139 stubName = "sha3_implCompress";
9140 block_size = get_block_size_from_digest_object(digestBase_obj);
9141 if (block_size == nullptr) return false;
9142 break;
9143 default:
9144 fatal_unexpected_iid(id);
9145 return false;
9146 }
9147 if (state == nullptr) return false;
9148
9149 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
9150 if (stubAddr == nullptr) return false;
9151
9152 // Call the stub.
9153 Node* call;
9154 if (block_size == nullptr) {
9155 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
9156 stubAddr, stubName, TypePtr::BOTTOM,
9157 src_start, state);
9158 } else {
9159 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
9160 stubAddr, stubName, TypePtr::BOTTOM,
9161 src_start, state, block_size);
9162 }
9163
9164 return true;
9165 }
9166
9167 //------------------------------inline_double_keccak
9168 bool LibraryCallKit::inline_double_keccak() {
9169 address stubAddr;
9170 const char *stubName;
9171 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
9172 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
9173
9174 stubAddr = StubRoutines::double_keccak();
9175 stubName = "double_keccak";
9176 if (!stubAddr) return false;
9177
9178 Node* status0 = argument(0);
9179 Node* status1 = argument(1);
9180
9181 status0 = must_be_not_null(status0, true);
9182 status1 = must_be_not_null(status1, true);
9183
9184 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
9185 assert(status0_start, "status0 is null");
9186 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
9187 assert(status1_start, "status1 is null");
9188 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
9189 OptoRuntime::double_keccak_Type(),
9190 stubAddr, stubName, TypePtr::BOTTOM,
9191 status0_start, status1_start);
9192 // return an int
9193 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
9194 set_result(retvalue);
9195 return true;
9196 }
9197
9198
9199 //------------------------------inline_digestBase_implCompressMB-----------------------
9200 //
9201 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
9202 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
9203 //
9204 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
9205 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9206 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9207 assert((uint)predicate < 5, "sanity");
9208 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
9209
9210 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
9211 Node* src = argument(1); // byte[] array
9212 Node* ofs = argument(2); // type int
9213 Node* limit = argument(3); // type int
9214
9215 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
9216 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
9217 // failed array check
9218 return false;
9219 }
9220 // Figure out the size and type of the elements we will be copying.
9221 BasicType src_elem = src_type->elem()->array_element_basic_type();
9222 if (src_elem != T_BYTE) {
9223 return false;
9224 }
9225 // 'src_start' points to src array + offset
9226 src = must_be_not_null(src, false);
9227 Node* src_start = array_element_address(src, ofs, src_elem);
9228
9229 const char* klass_digestBase_name = nullptr;
9230 const char* stub_name = nullptr;
9231 address stub_addr = nullptr;
9232 BasicType elem_type = T_INT;
9233
9234 switch (predicate) {
9235 case 0:
9236 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
9237 klass_digestBase_name = "sun/security/provider/MD5";
9238 stub_name = "md5_implCompressMB";
9239 stub_addr = StubRoutines::md5_implCompressMB();
9240 }
9241 break;
9242 case 1:
9243 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
9244 klass_digestBase_name = "sun/security/provider/SHA";
9245 stub_name = "sha1_implCompressMB";
9246 stub_addr = StubRoutines::sha1_implCompressMB();
9247 }
9248 break;
9249 case 2:
9250 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
9251 klass_digestBase_name = "sun/security/provider/SHA2";
9252 stub_name = "sha256_implCompressMB";
9253 stub_addr = StubRoutines::sha256_implCompressMB();
9254 }
9255 break;
9256 case 3:
9257 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
9258 klass_digestBase_name = "sun/security/provider/SHA5";
9259 stub_name = "sha512_implCompressMB";
9260 stub_addr = StubRoutines::sha512_implCompressMB();
9261 elem_type = T_LONG;
9262 }
9263 break;
9264 case 4:
9265 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
9266 klass_digestBase_name = "sun/security/provider/SHA3";
9267 stub_name = "sha3_implCompressMB";
9268 stub_addr = StubRoutines::sha3_implCompressMB();
9269 elem_type = T_LONG;
9270 }
9271 break;
9272 default:
9273 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
9274 }
9275 if (klass_digestBase_name != nullptr) {
9276 assert(stub_addr != nullptr, "Stub is generated");
9277 if (stub_addr == nullptr) return false;
9278
9279 // get DigestBase klass to lookup for SHA klass
9280 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
9281 assert(tinst != nullptr, "digestBase_obj is not instance???");
9282 assert(tinst->is_loaded(), "DigestBase is not loaded");
9283
9284 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
9285 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
9286 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
9287 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
9288 }
9289 return false;
9290 }
9291
9292 //------------------------------inline_digestBase_implCompressMB-----------------------
9293 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
9294 BasicType elem_type, address stubAddr, const char *stubName,
9295 Node* src_start, Node* ofs, Node* limit) {
9296 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
9297 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
9298 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
9299 digest_obj = _gvn.transform(digest_obj);
9300
9301 Node* state = get_state_from_digest_object(digest_obj, elem_type);
9302 if (state == nullptr) return false;
9303
9304 Node* block_size = nullptr;
9305 if (strcmp("sha3_implCompressMB", stubName) == 0) {
9306 block_size = get_block_size_from_digest_object(digest_obj);
9307 if (block_size == nullptr) return false;
9308 }
9309
9310 // Call the stub.
9311 Node* call;
9312 if (block_size == nullptr) {
9313 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9314 OptoRuntime::digestBase_implCompressMB_Type(false),
9315 stubAddr, stubName, TypePtr::BOTTOM,
9316 src_start, state, ofs, limit);
9317 } else {
9318 call = make_runtime_call(RC_LEAF|RC_NO_FP,
9319 OptoRuntime::digestBase_implCompressMB_Type(true),
9320 stubAddr, stubName, TypePtr::BOTTOM,
9321 src_start, state, block_size, ofs, limit);
9322 }
9323
9324 // return ofs (int)
9325 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
9326 set_result(result);
9327
9328 return true;
9329 }
9330
9331 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
9332 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
9333 assert(UseAES, "need AES instruction support");
9334 address stubAddr = nullptr;
9335 const char *stubName = nullptr;
9336 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
9337 stubName = "galoisCounterMode_AESCrypt";
9338
9339 if (stubAddr == nullptr) return false;
9340
9341 Node* in = argument(0);
9342 Node* inOfs = argument(1);
9343 Node* len = argument(2);
9344 Node* ct = argument(3);
9345 Node* ctOfs = argument(4);
9346 Node* out = argument(5);
9347 Node* outOfs = argument(6);
9348 Node* gctr_object = argument(7);
9349 Node* ghash_object = argument(8);
9350
9351 // (1) in, ct and out are arrays.
9352 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
9353 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
9354 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
9355 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
9356 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
9357 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
9358
9359 // checks are the responsibility of the caller
9360 Node* in_start = in;
9361 Node* ct_start = ct;
9362 Node* out_start = out;
9363 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
9364 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
9365 in_start = array_element_address(in, inOfs, T_BYTE);
9366 ct_start = array_element_address(ct, ctOfs, T_BYTE);
9367 out_start = array_element_address(out, outOfs, T_BYTE);
9368 }
9369
9370 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
9371 // (because of the predicated logic executed earlier).
9372 // so we cast it here safely.
9373 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
9374 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9375 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
9376 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
9377 Node* state = load_field_from_object(ghash_object, "state", "[J");
9378
9379 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
9380 return false;
9381 }
9382 // cast it to what we know it will be at runtime
9383 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
9384 assert(tinst != nullptr, "GCTR obj is null");
9385 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9386 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9387 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
9388 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9389 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
9390 const TypeOopPtr* xtype = aklass->as_instance_type();
9391 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
9392 aescrypt_object = _gvn.transform(aescrypt_object);
9393 // we need to get the start of the aescrypt_object's expanded key array
9394 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
9395 if (k_start == nullptr) return false;
9396 // similarly, get the start address of the r vector
9397 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
9398 Node* state_start = array_element_address(state, intcon(0), T_LONG);
9399 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
9400
9401
9402 // Call the stub, passing params
9403 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
9404 OptoRuntime::galoisCounterMode_aescrypt_Type(),
9405 stubAddr, stubName, TypePtr::BOTTOM,
9406 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
9407
9408 // return cipher length (int)
9409 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
9410 set_result(retvalue);
9411
9412 return true;
9413 }
9414
9415 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
9416 // Return node representing slow path of predicate check.
9417 // the pseudo code we want to emulate with this predicate is:
9418 // for encryption:
9419 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
9420 // for decryption:
9421 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
9422 // note cipher==plain is more conservative than the original java code but that's OK
9423 //
9424
9425 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
9426 // The receiver was checked for null already.
9427 Node* objGCTR = argument(7);
9428 // Load embeddedCipher field of GCTR object.
9429 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
9430 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
9431
9432 // get AESCrypt klass for instanceOf check
9433 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
9434 // will have same classloader as CipherBlockChaining object
9435 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
9436 assert(tinst != nullptr, "GCTR obj is null");
9437 assert(tinst->is_loaded(), "GCTR obj is not loaded");
9438
9439 // we want to do an instanceof comparison against the AESCrypt class
9440 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
9441 if (!klass_AESCrypt->is_loaded()) {
9442 // if AESCrypt is not even loaded, we never take the intrinsic fast path
9443 Node* ctrl = control();
9444 set_control(top()); // no regular fast path
9445 return ctrl;
9446 }
9447
9448 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
9449 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
9450 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9451 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9452 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9453
9454 return instof_false; // even if it is null
9455 }
9456
9457 //------------------------------get_state_from_digest_object-----------------------
9458 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
9459 const char* state_type;
9460 switch (elem_type) {
9461 case T_BYTE: state_type = "[B"; break;
9462 case T_INT: state_type = "[I"; break;
9463 case T_LONG: state_type = "[J"; break;
9464 default: ShouldNotReachHere();
9465 }
9466 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
9467 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
9468 if (digest_state == nullptr) return (Node *) nullptr;
9469
9470 // now have the array, need to get the start address of the state array
9471 Node* state = array_element_address(digest_state, intcon(0), elem_type);
9472 return state;
9473 }
9474
9475 //------------------------------get_block_size_from_sha3_object----------------------------------
9476 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
9477 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
9478 assert (block_size != nullptr, "sanity");
9479 return block_size;
9480 }
9481
9482 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
9483 // Return node representing slow path of predicate check.
9484 // the pseudo code we want to emulate with this predicate is:
9485 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
9486 //
9487 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
9488 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
9489 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
9490 assert((uint)predicate < 5, "sanity");
9491
9492 // The receiver was checked for null already.
9493 Node* digestBaseObj = argument(0);
9494
9495 // get DigestBase klass for instanceOf check
9496 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
9497 assert(tinst != nullptr, "digestBaseObj is null");
9498 assert(tinst->is_loaded(), "DigestBase is not loaded");
9499
9500 const char* klass_name = nullptr;
9501 switch (predicate) {
9502 case 0:
9503 if (UseMD5Intrinsics) {
9504 // we want to do an instanceof comparison against the MD5 class
9505 klass_name = "sun/security/provider/MD5";
9506 }
9507 break;
9508 case 1:
9509 if (UseSHA1Intrinsics) {
9510 // we want to do an instanceof comparison against the SHA class
9511 klass_name = "sun/security/provider/SHA";
9512 }
9513 break;
9514 case 2:
9515 if (UseSHA256Intrinsics) {
9516 // we want to do an instanceof comparison against the SHA2 class
9517 klass_name = "sun/security/provider/SHA2";
9518 }
9519 break;
9520 case 3:
9521 if (UseSHA512Intrinsics) {
9522 // we want to do an instanceof comparison against the SHA5 class
9523 klass_name = "sun/security/provider/SHA5";
9524 }
9525 break;
9526 case 4:
9527 if (UseSHA3Intrinsics) {
9528 // we want to do an instanceof comparison against the SHA3 class
9529 klass_name = "sun/security/provider/SHA3";
9530 }
9531 break;
9532 default:
9533 fatal("unknown SHA intrinsic predicate: %d", predicate);
9534 }
9535
9536 ciKlass* klass = nullptr;
9537 if (klass_name != nullptr) {
9538 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
9539 }
9540 if ((klass == nullptr) || !klass->is_loaded()) {
9541 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
9542 Node* ctrl = control();
9543 set_control(top()); // no intrinsic path
9544 return ctrl;
9545 }
9546 ciInstanceKlass* instklass = klass->as_instance_klass();
9547
9548 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
9549 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
9550 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
9551 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
9552
9553 return instof_false; // even if it is null
9554 }
9555
9556 //-------------inline_fma-----------------------------------
9557 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
9558 Node *a = nullptr;
9559 Node *b = nullptr;
9560 Node *c = nullptr;
9561 Node* result = nullptr;
9562 switch (id) {
9563 case vmIntrinsics::_fmaD:
9564 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
9565 // no receiver since it is static method
9566 a = argument(0);
9567 b = argument(2);
9568 c = argument(4);
9569 result = _gvn.transform(new FmaDNode(a, b, c));
9570 break;
9571 case vmIntrinsics::_fmaF:
9572 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
9573 a = argument(0);
9574 b = argument(1);
9575 c = argument(2);
9576 result = _gvn.transform(new FmaFNode(a, b, c));
9577 break;
9578 default:
9579 fatal_unexpected_iid(id); break;
9580 }
9581 set_result(result);
9582 return true;
9583 }
9584
9585 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
9586 // argument(0) is receiver
9587 Node* codePoint = argument(1);
9588 Node* n = nullptr;
9589
9590 switch (id) {
9591 case vmIntrinsics::_isDigit :
9592 n = new DigitNode(control(), codePoint);
9593 break;
9594 case vmIntrinsics::_isLowerCase :
9595 n = new LowerCaseNode(control(), codePoint);
9596 break;
9597 case vmIntrinsics::_isUpperCase :
9598 n = new UpperCaseNode(control(), codePoint);
9599 break;
9600 case vmIntrinsics::_isWhitespace :
9601 n = new WhitespaceNode(control(), codePoint);
9602 break;
9603 default:
9604 fatal_unexpected_iid(id);
9605 }
9606
9607 set_result(_gvn.transform(n));
9608 return true;
9609 }
9610
9611 bool LibraryCallKit::inline_profileBoolean() {
9612 Node* counts = argument(1);
9613 const TypeAryPtr* ary = nullptr;
9614 ciArray* aobj = nullptr;
9615 if (counts->is_Con()
9616 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
9617 && (aobj = ary->const_oop()->as_array()) != nullptr
9618 && (aobj->length() == 2)) {
9619 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
9620 jint false_cnt = aobj->element_value(0).as_int();
9621 jint true_cnt = aobj->element_value(1).as_int();
9622
9623 if (C->log() != nullptr) {
9624 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
9625 false_cnt, true_cnt);
9626 }
9627
9628 if (false_cnt + true_cnt == 0) {
9629 // According to profile, never executed.
9630 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9631 Deoptimization::Action_reinterpret);
9632 return true;
9633 }
9634
9635 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
9636 // is a number of each value occurrences.
9637 Node* result = argument(0);
9638 if (false_cnt == 0 || true_cnt == 0) {
9639 // According to profile, one value has been never seen.
9640 int expected_val = (false_cnt == 0) ? 1 : 0;
9641
9642 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
9643 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
9644
9645 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
9646 Node* fast_path = _gvn.transform(new IfTrueNode(check));
9647 Node* slow_path = _gvn.transform(new IfFalseNode(check));
9648
9649 { // Slow path: uncommon trap for never seen value and then reexecute
9650 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
9651 // the value has been seen at least once.
9652 PreserveJVMState pjvms(this);
9653 PreserveReexecuteState preexecs(this);
9654 jvms()->set_should_reexecute(true);
9655
9656 set_control(slow_path);
9657 set_i_o(i_o());
9658
9659 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
9660 Deoptimization::Action_reinterpret);
9661 }
9662 // The guard for never seen value enables sharpening of the result and
9663 // returning a constant. It allows to eliminate branches on the same value
9664 // later on.
9665 set_control(fast_path);
9666 result = intcon(expected_val);
9667 }
9668 // Stop profiling.
9669 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
9670 // By replacing method body with profile data (represented as ProfileBooleanNode
9671 // on IR level) we effectively disable profiling.
9672 // It enables full speed execution once optimized code is generated.
9673 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
9674 C->record_for_igvn(profile);
9675 set_result(profile);
9676 return true;
9677 } else {
9678 // Continue profiling.
9679 // Profile data isn't available at the moment. So, execute method's bytecode version.
9680 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
9681 // is compiled and counters aren't available since corresponding MethodHandle
9682 // isn't a compile-time constant.
9683 return false;
9684 }
9685 }
9686
9687 bool LibraryCallKit::inline_isCompileConstant() {
9688 Node* n = argument(0);
9689 set_result(n->is_Con() ? intcon(1) : intcon(0));
9690 return true;
9691 }
9692
9693 //------------------------------- inline_getObjectSize --------------------------------------
9694 //
9695 // Calculate the runtime size of the object/array.
9696 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9697 //
9698 bool LibraryCallKit::inline_getObjectSize() {
9699 Node* obj = argument(3);
9700 Node* klass_node = load_object_klass(obj);
9701
9702 jint layout_con = Klass::_lh_neutral_value;
9703 Node* layout_val = get_layout_helper(klass_node, layout_con);
9704 int layout_is_con = (layout_val == nullptr);
9705
9706 if (layout_is_con) {
9707 // Layout helper is constant, can figure out things at compile time.
9708
9709 if (Klass::layout_helper_is_instance(layout_con)) {
9710 // Instance case: layout_con contains the size itself.
9711 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9712 set_result(size);
9713 } else {
9714 // Array case: size is round(header + element_size*arraylength).
9715 // Since arraylength is different for every array instance, we have to
9716 // compute the whole thing at runtime.
9717
9718 Node* arr_length = load_array_length(obj);
9719
9720 int round_mask = MinObjAlignmentInBytes - 1;
9721 int hsize = Klass::layout_helper_header_size(layout_con);
9722 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9723
9724 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9725 round_mask = 0; // strength-reduce it if it goes away completely
9726 }
9727 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9728 Node* header_size = intcon(hsize + round_mask);
9729
9730 Node* lengthx = ConvI2X(arr_length);
9731 Node* headerx = ConvI2X(header_size);
9732
9733 Node* abody = lengthx;
9734 if (eshift != 0) {
9735 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9736 }
9737 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9738 if (round_mask != 0) {
9739 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9740 }
9741 size = ConvX2L(size);
9742 set_result(size);
9743 }
9744 } else {
9745 // Layout helper is not constant, need to test for array-ness at runtime.
9746
9747 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9748 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9749 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9750 record_for_igvn(result_reg);
9751
9752 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9753 if (array_ctl != nullptr) {
9754 // Array case: size is round(header + element_size*arraylength).
9755 // Since arraylength is different for every array instance, we have to
9756 // compute the whole thing at runtime.
9757
9758 PreserveJVMState pjvms(this);
9759 set_control(array_ctl);
9760 Node* arr_length = load_array_length(obj);
9761
9762 int round_mask = MinObjAlignmentInBytes - 1;
9763 Node* mask = intcon(round_mask);
9764
9765 Node* hss = intcon(Klass::_lh_header_size_shift);
9766 Node* hsm = intcon(Klass::_lh_header_size_mask);
9767 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9768 header_size = _gvn.transform(new AndINode(header_size, hsm));
9769 header_size = _gvn.transform(new AddINode(header_size, mask));
9770
9771 // There is no need to mask or shift this value.
9772 // The semantics of LShiftINode include an implicit mask to 0x1F.
9773 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9774 Node* elem_shift = layout_val;
9775
9776 Node* lengthx = ConvI2X(arr_length);
9777 Node* headerx = ConvI2X(header_size);
9778
9779 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9780 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9781 if (round_mask != 0) {
9782 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9783 }
9784 size = ConvX2L(size);
9785
9786 result_reg->init_req(_array_path, control());
9787 result_val->init_req(_array_path, size);
9788 }
9789
9790 if (!stopped()) {
9791 // Instance case: the layout helper gives us instance size almost directly,
9792 // but we need to mask out the _lh_instance_slow_path_bit.
9793 Node* size = ConvI2X(layout_val);
9794 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9795 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9796 size = _gvn.transform(new AndXNode(size, mask));
9797 size = ConvX2L(size);
9798
9799 result_reg->init_req(_instance_path, control());
9800 result_val->init_req(_instance_path, size);
9801 }
9802
9803 set_result(result_reg, result_val);
9804 }
9805
9806 return true;
9807 }
9808
9809 //------------------------------- inline_blackhole --------------------------------------
9810 //
9811 // Make sure all arguments to this node are alive.
9812 // This matches methods that were requested to be blackholed through compile commands.
9813 //
9814 bool LibraryCallKit::inline_blackhole() {
9815 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9816 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9817 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9818
9819 // Blackhole node pinches only the control, not memory. This allows
9820 // the blackhole to be pinned in the loop that computes blackholed
9821 // values, but have no other side effects, like breaking the optimizations
9822 // across the blackhole.
9823
9824 Node* bh = _gvn.transform(new BlackholeNode(control()));
9825 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9826
9827 // Bind call arguments as blackhole arguments to keep them alive
9828 uint nargs = callee()->arg_size();
9829 for (uint i = 0; i < nargs; i++) {
9830 bh->add_req(argument(i));
9831 }
9832
9833 return true;
9834 }
9835
9836 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9837 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9838 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9839 return nullptr; // box klass is not Float16
9840 }
9841
9842 // Null check; get notnull casted pointer
9843 Node* null_ctl = top();
9844 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9845 // If not_null_box is dead, only null-path is taken
9846 if (stopped()) {
9847 set_control(null_ctl);
9848 return nullptr;
9849 }
9850 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9851 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9852 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9853 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9854 }
9855
9856 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9857 PreserveReexecuteState preexecs(this);
9858 jvms()->set_should_reexecute(true);
9859
9860 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9861 Node* klass_node = makecon(klass_type);
9862 Node* box = new_instance(klass_node);
9863
9864 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9865 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9866
9867 Node* field_store = _gvn.transform(access_store_at(box,
9868 value_field,
9869 value_adr_type,
9870 value,
9871 TypeInt::SHORT,
9872 T_SHORT,
9873 IN_HEAP));
9874 set_memory(field_store, value_adr_type);
9875 return box;
9876 }
9877
9878 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9879 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9880 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9881 return false;
9882 }
9883
9884 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9885 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9886 return false;
9887 }
9888
9889 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9890 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9891 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9892 ciSymbols::short_signature(),
9893 false);
9894 assert(field != nullptr, "");
9895
9896 // Transformed nodes
9897 Node* fld1 = nullptr;
9898 Node* fld2 = nullptr;
9899 Node* fld3 = nullptr;
9900 switch(num_args) {
9901 case 3:
9902 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9903 if (fld3 == nullptr) {
9904 return false;
9905 }
9906 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9907 // fall-through
9908 case 2:
9909 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9910 if (fld2 == nullptr) {
9911 return false;
9912 }
9913 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9914 // fall-through
9915 case 1:
9916 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9917 if (fld1 == nullptr) {
9918 return false;
9919 }
9920 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9921 break;
9922 default: fatal("Unsupported number of arguments %d", num_args);
9923 }
9924
9925 Node* result = nullptr;
9926 switch (id) {
9927 // Unary operations
9928 case vmIntrinsics::_sqrt_float16:
9929 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9930 break;
9931 // Ternary operations
9932 case vmIntrinsics::_fma_float16:
9933 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9934 break;
9935 default:
9936 fatal_unexpected_iid(id);
9937 break;
9938 }
9939 result = _gvn.transform(new ReinterpretHF2SNode(result));
9940 set_result(box_fp16_value(float16_box_type, field, result));
9941 return true;
9942 }
9943