1 /*
2 * Copyright (c) 1999, 2026, 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/ciSymbols.hpp"
27 #include "ci/ciUtilities.inline.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/rootnode.hpp"
51 #include "opto/runtime.hpp"
52 #include "opto/subnode.hpp"
53 #include "opto/vectornode.hpp"
54 #include "prims/jvmtiExport.hpp"
55 #include "prims/jvmtiThreadState.hpp"
56 #include "prims/unsafe.hpp"
57 #include "runtime/jniHandles.inline.hpp"
58 #include "runtime/mountUnmountDisabler.hpp"
59 #include "runtime/objectMonitor.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/stubRoutines.hpp"
62 #include "utilities/macros.hpp"
63 #include "utilities/powerOfTwo.hpp"
64
65 //---------------------------make_vm_intrinsic----------------------------
66 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
67 vmIntrinsicID id = m->intrinsic_id();
68 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
69
70 if (!m->is_loaded()) {
71 // Do not attempt to inline unloaded methods.
72 return nullptr;
73 }
74
75 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
76 bool is_available = false;
77
78 {
79 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
80 // the compiler must transition to '_thread_in_vm' state because both
81 // methods access VM-internal data.
82 VM_ENTRY_MARK;
83 methodHandle mh(THREAD, m->get_Method());
84 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
85 if (is_available && is_virtual) {
86 is_available = vmIntrinsics::does_virtual_dispatch(id);
87 }
88 }
89
90 if (is_available) {
91 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
92 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
93 return new LibraryIntrinsic(m, is_virtual,
94 vmIntrinsics::predicates_needed(id),
95 vmIntrinsics::does_virtual_dispatch(id),
96 id);
97 } else {
98 return nullptr;
99 }
100 }
101
102 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
103 LibraryCallKit kit(jvms, this);
104 Compile* C = kit.C;
105 int nodes = C->unique();
106 #ifndef PRODUCT
107 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
108 char buf[1000];
109 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
110 tty->print_cr("Intrinsic %s", str);
111 }
112 #endif
113 ciMethod* callee = kit.callee();
114 const int bci = kit.bci();
115 #ifdef ASSERT
116 Node* ctrl = kit.control();
117 #endif
118 // Try to inline the intrinsic.
119 if (callee->check_intrinsic_candidate() &&
120 kit.try_to_inline(_last_predicate)) {
121 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
122 : "(intrinsic)";
123 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
124 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
125 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
126 if (C->log()) {
127 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
128 vmIntrinsics::name_at(intrinsic_id()),
129 (is_virtual() ? " virtual='1'" : ""),
130 C->unique() - nodes);
131 }
132 // Push the result from the inlined method onto the stack.
133 kit.push_result();
134 return kit.transfer_exceptions_into_jvms();
135 }
136
137 // The intrinsic bailed out
138 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
139 assert(jvms->map() == kit.map(), "Out of sync JVM state");
140 if (jvms->has_method()) {
141 // Not a root compile.
142 const char* msg;
143 if (callee->intrinsic_candidate()) {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
145 } else {
146 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
147 : "failed to inline (intrinsic), method not annotated";
148 }
149 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
150 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
151 } else {
152 // Root compile
153 ResourceMark rm;
154 stringStream msg_stream;
155 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
156 vmIntrinsics::name_at(intrinsic_id()),
157 is_virtual() ? " (virtual)" : "", bci);
158 const char *msg = msg_stream.freeze();
159 log_debug(jit, inlining)("%s", msg);
160 if (C->print_intrinsics() || C->print_inlining()) {
161 tty->print("%s", msg);
162 }
163 }
164 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
165
166 return nullptr;
167 }
168
169 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
170 LibraryCallKit kit(jvms, this);
171 Compile* C = kit.C;
172 int nodes = C->unique();
173 _last_predicate = predicate;
174 #ifndef PRODUCT
175 assert(is_predicated() && predicate < predicates_count(), "sanity");
176 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
177 char buf[1000];
178 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
179 tty->print_cr("Predicate for intrinsic %s", str);
180 }
181 #endif
182 ciMethod* callee = kit.callee();
183 const int bci = kit.bci();
184
185 Node* slow_ctl = kit.try_to_predicate(predicate);
186 if (!kit.failing()) {
187 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
188 : "(intrinsic, predicate)";
189 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
190 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
191
192 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
193 if (C->log()) {
194 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
195 vmIntrinsics::name_at(intrinsic_id()),
196 (is_virtual() ? " virtual='1'" : ""),
197 C->unique() - nodes);
198 }
199 return slow_ctl; // Could be null if the check folds.
200 }
201
202 // The intrinsic bailed out
203 if (jvms->has_method()) {
204 // Not a root compile.
205 const char* msg = "failed to generate predicate for intrinsic";
206 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
207 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
208 } else {
209 // Root compile
210 ResourceMark rm;
211 stringStream msg_stream;
212 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
213 vmIntrinsics::name_at(intrinsic_id()),
214 is_virtual() ? " (virtual)" : "", bci);
215 const char *msg = msg_stream.freeze();
216 log_debug(jit, inlining)("%s", msg);
217 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
218 }
219 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
220 return nullptr;
221 }
222
223 bool LibraryCallKit::try_to_inline(int predicate) {
224 // Handle symbolic names for otherwise undistinguished boolean switches:
225 const bool is_store = true;
226 const bool is_compress = true;
227 const bool is_static = true;
228 const bool is_volatile = true;
229
230 if (!jvms()->has_method()) {
231 // Root JVMState has a null method.
232 assert(map()->memory()->Opcode() == Op_Parm, "");
233 // Insert the memory aliasing node
234 set_all_memory(reset_memory());
235 }
236 assert(merged_memory(), "");
237
238 switch (intrinsic_id()) {
239 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
240 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
241 case vmIntrinsics::_getClass: return inline_native_getClass();
242
243 case vmIntrinsics::_ceil:
244 case vmIntrinsics::_floor:
245 case vmIntrinsics::_rint:
246 case vmIntrinsics::_dsin:
247 case vmIntrinsics::_dcos:
248 case vmIntrinsics::_dtan:
249 case vmIntrinsics::_dsinh:
250 case vmIntrinsics::_dtanh:
251 case vmIntrinsics::_dcbrt:
252 case vmIntrinsics::_dabs:
253 case vmIntrinsics::_fabs:
254 case vmIntrinsics::_iabs:
255 case vmIntrinsics::_labs:
256 case vmIntrinsics::_datan2:
257 case vmIntrinsics::_dsqrt:
258 case vmIntrinsics::_dsqrt_strict:
259 case vmIntrinsics::_dexp:
260 case vmIntrinsics::_dlog:
261 case vmIntrinsics::_dlog10:
262 case vmIntrinsics::_dpow:
263 case vmIntrinsics::_dcopySign:
264 case vmIntrinsics::_fcopySign:
265 case vmIntrinsics::_dsignum:
266 case vmIntrinsics::_roundF:
267 case vmIntrinsics::_roundD:
268 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
269
270 case vmIntrinsics::_notify:
271 case vmIntrinsics::_notifyAll:
272 return inline_notify(intrinsic_id());
273
274 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
275 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
276 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
277 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
278 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
279 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
280 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
281 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
282 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
283 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
284 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
285 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
286 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
287 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
288
289 case vmIntrinsics::_arraycopy: return inline_arraycopy();
290
291 case vmIntrinsics::_arraySort: return inline_array_sort();
292 case vmIntrinsics::_arrayPartition: return inline_array_partition();
293
294 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
295 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
296 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
297 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
298
299 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
300 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
301 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
302 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
303 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
304 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
305 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
306 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
307
308 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
309
310 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
311
312 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
313 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
314 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
315 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
316
317 case vmIntrinsics::_compressStringC:
318 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
319 case vmIntrinsics::_inflateStringC:
320 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
321
322 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
323 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
324 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
325 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
326 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
327 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
328 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
329 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
330 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
331
332 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
333 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
334 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
335 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
336 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
337 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
338 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
339 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
340 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
341
342 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
343 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
344 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
345 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
346 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
347 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
348 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
349 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
350 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
351
352 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
353 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
354 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
355 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
356 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
357 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
358 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
359 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
360 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
361
362 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
363 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
364 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
365 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
366
367 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
368 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
369 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
370 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
371
372 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
373 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
374 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
375 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
376 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
377 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
378 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
379 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
380 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
381
382 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
383 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
384 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
385 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
386 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
387 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
388 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
389 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
390 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
391
392 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
393 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
394 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
395 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
396 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
397 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
398 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
399 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
400 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
401
402 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
403 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
404 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
405 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
406 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
407 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
408 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
409 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
410 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
411
412 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
413 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
414 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
415 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
416 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
417
418 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
434 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
435 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
436 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
437 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
438
439 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
440 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
441 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
442 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
443 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
444 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
445 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
446 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
447 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
448 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
449 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
450 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
451 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
452 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
453 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
454
455 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
456 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
457 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
458 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
459
460 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
461 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
462 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
463 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
464 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
465
466 case vmIntrinsics::_loadFence:
467 case vmIntrinsics::_storeFence:
468 case vmIntrinsics::_storeStoreFence:
469 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
470
471 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
472
473 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
474 case vmIntrinsics::_currentThread: return inline_native_currentThread();
475 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
476
477 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
478 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
479
480 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
481 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
482
483 case vmIntrinsics::_vthreadEndFirstTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_first_transition_Java()),
484 "endFirstTransition", true);
485 case vmIntrinsics::_vthreadStartFinalTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_final_transition_Java()),
486 "startFinalTransition", true);
487 case vmIntrinsics::_vthreadStartTransition: return inline_native_vthread_start_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_start_transition_Java()),
488 "startTransition", false);
489 case vmIntrinsics::_vthreadEndTransition: return inline_native_vthread_end_transition(CAST_FROM_FN_PTR(address, OptoRuntime::vthread_end_transition_Java()),
490 "endTransition", false);
491 #if INCLUDE_JVMTI
492 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
493 #endif
494
495 #ifdef JFR_HAVE_INTRINSICS
496 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
497 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
498 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
499 #endif
500 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
501 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
502 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
503 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
504 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
505 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
506 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
507 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
508 case vmIntrinsics::_getLength: return inline_native_getLength();
509 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
510 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
511 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
512 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
513 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
514 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
515 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
516
517 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
518 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
519
520 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
521
522 case vmIntrinsics::_isInstance:
523 case vmIntrinsics::_isHidden:
524 case vmIntrinsics::_getSuperclass: return inline_native_Class_query(intrinsic_id());
525
526 case vmIntrinsics::_floatToRawIntBits:
527 case vmIntrinsics::_floatToIntBits:
528 case vmIntrinsics::_intBitsToFloat:
529 case vmIntrinsics::_doubleToRawLongBits:
530 case vmIntrinsics::_doubleToLongBits:
531 case vmIntrinsics::_longBitsToDouble:
532 case vmIntrinsics::_floatToFloat16:
533 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
534 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
535 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
536 case vmIntrinsics::_floatIsFinite:
537 case vmIntrinsics::_floatIsInfinite:
538 case vmIntrinsics::_doubleIsFinite:
539 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
540
541 case vmIntrinsics::_numberOfLeadingZeros_i:
542 case vmIntrinsics::_numberOfLeadingZeros_l:
543 case vmIntrinsics::_numberOfTrailingZeros_i:
544 case vmIntrinsics::_numberOfTrailingZeros_l:
545 case vmIntrinsics::_bitCount_i:
546 case vmIntrinsics::_bitCount_l:
547 case vmIntrinsics::_reverse_i:
548 case vmIntrinsics::_reverse_l:
549 case vmIntrinsics::_reverseBytes_i:
550 case vmIntrinsics::_reverseBytes_l:
551 case vmIntrinsics::_reverseBytes_s:
552 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
553
554 case vmIntrinsics::_compress_i:
555 case vmIntrinsics::_compress_l:
556 case vmIntrinsics::_expand_i:
557 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
558
559 case vmIntrinsics::_compareUnsigned_i:
560 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
561
562 case vmIntrinsics::_divideUnsigned_i:
563 case vmIntrinsics::_divideUnsigned_l:
564 case vmIntrinsics::_remainderUnsigned_i:
565 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
566
567 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
568
569 case vmIntrinsics::_Reference_get0: return inline_reference_get0();
570 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
571 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
572 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
573 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
574
575 case vmIntrinsics::_Class_cast: return inline_Class_cast();
576
577 case vmIntrinsics::_aescrypt_encryptBlock:
578 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
579
580 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
581 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
582 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
583
584 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
585 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
586 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_counterMode_AESCrypt:
589 return inline_counterMode_AESCrypt(intrinsic_id());
590
591 case vmIntrinsics::_galoisCounterMode_AESCrypt:
592 return inline_galoisCounterMode_AESCrypt();
593
594 case vmIntrinsics::_md5_implCompress:
595 case vmIntrinsics::_sha_implCompress:
596 case vmIntrinsics::_sha2_implCompress:
597 case vmIntrinsics::_sha5_implCompress:
598 case vmIntrinsics::_sha3_implCompress:
599 return inline_digestBase_implCompress(intrinsic_id());
600 case vmIntrinsics::_double_keccak:
601 return inline_double_keccak();
602
603 case vmIntrinsics::_digestBase_implCompressMB:
604 return inline_digestBase_implCompressMB(predicate);
605
606 case vmIntrinsics::_multiplyToLen:
607 return inline_multiplyToLen();
608
609 case vmIntrinsics::_squareToLen:
610 return inline_squareToLen();
611
612 case vmIntrinsics::_mulAdd:
613 return inline_mulAdd();
614
615 case vmIntrinsics::_montgomeryMultiply:
616 return inline_montgomeryMultiply();
617 case vmIntrinsics::_montgomerySquare:
618 return inline_montgomerySquare();
619
620 case vmIntrinsics::_bigIntegerRightShiftWorker:
621 return inline_bigIntegerShift(true);
622 case vmIntrinsics::_bigIntegerLeftShiftWorker:
623 return inline_bigIntegerShift(false);
624
625 case vmIntrinsics::_vectorizedMismatch:
626 return inline_vectorizedMismatch();
627
628 case vmIntrinsics::_ghash_processBlocks:
629 return inline_ghash_processBlocks();
630 case vmIntrinsics::_chacha20Block:
631 return inline_chacha20Block();
632 case vmIntrinsics::_kyberNtt:
633 return inline_kyberNtt();
634 case vmIntrinsics::_kyberInverseNtt:
635 return inline_kyberInverseNtt();
636 case vmIntrinsics::_kyberNttMult:
637 return inline_kyberNttMult();
638 case vmIntrinsics::_kyberAddPoly_2:
639 return inline_kyberAddPoly_2();
640 case vmIntrinsics::_kyberAddPoly_3:
641 return inline_kyberAddPoly_3();
642 case vmIntrinsics::_kyber12To16:
643 return inline_kyber12To16();
644 case vmIntrinsics::_kyberBarrettReduce:
645 return inline_kyberBarrettReduce();
646 case vmIntrinsics::_dilithiumAlmostNtt:
647 return inline_dilithiumAlmostNtt();
648 case vmIntrinsics::_dilithiumAlmostInverseNtt:
649 return inline_dilithiumAlmostInverseNtt();
650 case vmIntrinsics::_dilithiumNttMult:
651 return inline_dilithiumNttMult();
652 case vmIntrinsics::_dilithiumMontMulByConstant:
653 return inline_dilithiumMontMulByConstant();
654 case vmIntrinsics::_dilithiumDecomposePoly:
655 return inline_dilithiumDecomposePoly();
656 case vmIntrinsics::_base64_encodeBlock:
657 return inline_base64_encodeBlock();
658 case vmIntrinsics::_base64_decodeBlock:
659 return inline_base64_decodeBlock();
660 case vmIntrinsics::_poly1305_processBlocks:
661 return inline_poly1305_processBlocks();
662 case vmIntrinsics::_intpoly_montgomeryMult_P256:
663 return inline_intpoly_montgomeryMult_P256();
664 case vmIntrinsics::_intpoly_assign:
665 return inline_intpoly_assign();
666 case vmIntrinsics::_encodeISOArray:
667 case vmIntrinsics::_encodeByteISOArray:
668 return inline_encodeISOArray(false);
669 case vmIntrinsics::_encodeAsciiArray:
670 return inline_encodeISOArray(true);
671
672 case vmIntrinsics::_updateCRC32:
673 return inline_updateCRC32();
674 case vmIntrinsics::_updateBytesCRC32:
675 return inline_updateBytesCRC32();
676 case vmIntrinsics::_updateByteBufferCRC32:
677 return inline_updateByteBufferCRC32();
678
679 case vmIntrinsics::_updateBytesCRC32C:
680 return inline_updateBytesCRC32C();
681 case vmIntrinsics::_updateDirectByteBufferCRC32C:
682 return inline_updateDirectByteBufferCRC32C();
683
684 case vmIntrinsics::_updateBytesAdler32:
685 return inline_updateBytesAdler32();
686 case vmIntrinsics::_updateByteBufferAdler32:
687 return inline_updateByteBufferAdler32();
688
689 case vmIntrinsics::_profileBoolean:
690 return inline_profileBoolean();
691 case vmIntrinsics::_isCompileConstant:
692 return inline_isCompileConstant();
693
694 case vmIntrinsics::_countPositives:
695 return inline_countPositives();
696
697 case vmIntrinsics::_fmaD:
698 case vmIntrinsics::_fmaF:
699 return inline_fma(intrinsic_id());
700
701 case vmIntrinsics::_isDigit:
702 case vmIntrinsics::_isLowerCase:
703 case vmIntrinsics::_isUpperCase:
704 case vmIntrinsics::_isWhitespace:
705 return inline_character_compare(intrinsic_id());
706
707 case vmIntrinsics::_min:
708 case vmIntrinsics::_max:
709 case vmIntrinsics::_min_strict:
710 case vmIntrinsics::_max_strict:
711 case vmIntrinsics::_minL:
712 case vmIntrinsics::_maxL:
713 case vmIntrinsics::_minF:
714 case vmIntrinsics::_maxF:
715 case vmIntrinsics::_minD:
716 case vmIntrinsics::_maxD:
717 case vmIntrinsics::_minF_strict:
718 case vmIntrinsics::_maxF_strict:
719 case vmIntrinsics::_minD_strict:
720 case vmIntrinsics::_maxD_strict:
721 return inline_min_max(intrinsic_id());
722
723 case vmIntrinsics::_VectorUnaryOp:
724 return inline_vector_nary_operation(1);
725 case vmIntrinsics::_VectorBinaryOp:
726 return inline_vector_nary_operation(2);
727 case vmIntrinsics::_VectorUnaryLibOp:
728 return inline_vector_call(1);
729 case vmIntrinsics::_VectorBinaryLibOp:
730 return inline_vector_call(2);
731 case vmIntrinsics::_VectorTernaryOp:
732 return inline_vector_nary_operation(3);
733 case vmIntrinsics::_VectorFromBitsCoerced:
734 return inline_vector_frombits_coerced();
735 case vmIntrinsics::_VectorMaskOp:
736 return inline_vector_mask_operation();
737 case vmIntrinsics::_VectorLoadOp:
738 return inline_vector_mem_operation(/*is_store=*/false);
739 case vmIntrinsics::_VectorLoadMaskedOp:
740 return inline_vector_mem_masked_operation(/*is_store*/false);
741 case vmIntrinsics::_VectorStoreOp:
742 return inline_vector_mem_operation(/*is_store=*/true);
743 case vmIntrinsics::_VectorStoreMaskedOp:
744 return inline_vector_mem_masked_operation(/*is_store=*/true);
745 case vmIntrinsics::_VectorGatherOp:
746 return inline_vector_gather_scatter(/*is_scatter*/ false);
747 case vmIntrinsics::_VectorScatterOp:
748 return inline_vector_gather_scatter(/*is_scatter*/ true);
749 case vmIntrinsics::_VectorReductionCoerced:
750 return inline_vector_reduction();
751 case vmIntrinsics::_VectorTest:
752 return inline_vector_test();
753 case vmIntrinsics::_VectorBlend:
754 return inline_vector_blend();
755 case vmIntrinsics::_VectorRearrange:
756 return inline_vector_rearrange();
757 case vmIntrinsics::_VectorSelectFrom:
758 return inline_vector_select_from();
759 case vmIntrinsics::_VectorCompare:
760 return inline_vector_compare();
761 case vmIntrinsics::_VectorBroadcastInt:
762 return inline_vector_broadcast_int();
763 case vmIntrinsics::_VectorConvert:
764 return inline_vector_convert();
765 case vmIntrinsics::_VectorInsert:
766 return inline_vector_insert();
767 case vmIntrinsics::_VectorExtract:
768 return inline_vector_extract();
769 case vmIntrinsics::_VectorCompressExpand:
770 return inline_vector_compress_expand();
771 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
772 return inline_vector_select_from_two_vectors();
773 case vmIntrinsics::_IndexVector:
774 return inline_index_vector();
775 case vmIntrinsics::_IndexPartiallyInUpperRange:
776 return inline_index_partially_in_upper_range();
777
778 case vmIntrinsics::_getObjectSize:
779 return inline_getObjectSize();
780
781 case vmIntrinsics::_blackhole:
782 return inline_blackhole();
783
784 default:
785 // If you get here, it may be that someone has added a new intrinsic
786 // to the list in vmIntrinsics.hpp without implementing it here.
787 #ifndef PRODUCT
788 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
789 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
790 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
791 }
792 #endif
793 return false;
794 }
795 }
796
797 Node* LibraryCallKit::try_to_predicate(int predicate) {
798 if (!jvms()->has_method()) {
799 // Root JVMState has a null method.
800 assert(map()->memory()->Opcode() == Op_Parm, "");
801 // Insert the memory aliasing node
802 set_all_memory(reset_memory());
803 }
804 assert(merged_memory(), "");
805
806 switch (intrinsic_id()) {
807 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
808 return inline_cipherBlockChaining_AESCrypt_predicate(false);
809 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
810 return inline_cipherBlockChaining_AESCrypt_predicate(true);
811 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
812 return inline_electronicCodeBook_AESCrypt_predicate(false);
813 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
814 return inline_electronicCodeBook_AESCrypt_predicate(true);
815 case vmIntrinsics::_counterMode_AESCrypt:
816 return inline_counterMode_AESCrypt_predicate();
817 case vmIntrinsics::_digestBase_implCompressMB:
818 return inline_digestBase_implCompressMB_predicate(predicate);
819 case vmIntrinsics::_galoisCounterMode_AESCrypt:
820 return inline_galoisCounterMode_AESCrypt_predicate();
821
822 default:
823 // If you get here, it may be that someone has added a new intrinsic
824 // to the list in vmIntrinsics.hpp without implementing it here.
825 #ifndef PRODUCT
826 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
827 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
828 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
829 }
830 #endif
831 Node* slow_ctl = control();
832 set_control(top()); // No fast path intrinsic
833 return slow_ctl;
834 }
835 }
836
837 //------------------------------set_result-------------------------------
838 // Helper function for finishing intrinsics.
839 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
840 record_for_igvn(region);
841 set_control(_gvn.transform(region));
842 set_result( _gvn.transform(value));
843 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
844 }
845
846 RegionNode* LibraryCallKit::create_bailout() {
847 RegionNode* bailout = new RegionNode(1);
848 record_for_igvn(bailout);
849 return bailout;
850 }
851
852 bool LibraryCallKit::check_bailout(RegionNode* bailout) {
853 if (bailout->req() > 1) {
854 bailout = _gvn.transform(bailout)->as_Region();
855 Node* frame = _gvn.transform(new ParmNode(C->start(), TypeFunc::FramePtr));
856 Node* halt = _gvn.transform(new HaltNode(bailout, frame, "unexpected guard failure in intrinsic"));
857 C->root()->add_req(halt);
858 }
859 return stopped();
860 }
861
862 //------------------------------generate_guard---------------------------
863 // Helper function for generating guarded fast-slow graph structures.
864 // The given 'test', if true, guards a slow path. If the test fails
865 // then a fast path can be taken. (We generally hope it fails.)
866 // In all cases, GraphKit::control() is updated to the fast path.
867 // The returned value represents the control for the slow path.
868 // The return value is never 'top'; it is either a valid control
869 // or null if it is obvious that the slow path can never be taken.
870 // Also, if region and the slow control are not null, the slow edge
871 // is appended to the region.
872 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
873 if (stopped()) {
874 // Already short circuited.
875 return nullptr;
876 }
877
878 // Build an if node and its projections.
879 // If test is true we take the slow path, which we assume is uncommon.
880 if (_gvn.type(test) == TypeInt::ZERO) {
881 // The slow branch is never taken. No need to build this guard.
882 return nullptr;
883 }
884
885 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
886
887 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
888 if (if_slow == top()) {
889 // The slow branch is never taken. No need to build this guard.
890 return nullptr;
891 }
892
893 if (region != nullptr)
894 region->add_req(if_slow);
895
896 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
897 set_control(if_fast);
898
899 return if_slow;
900 }
901
902 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
903 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
904 }
905 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
906 return generate_guard(test, region, PROB_FAIR);
907 }
908
909 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
910 Node** pos_index, bool with_opaque) {
911 if (stopped())
912 return nullptr; // already stopped
913 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
914 return nullptr; // index is already adequately typed
915 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
916 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
917 if (with_opaque) {
918 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
919 }
920 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
921 if (is_neg != nullptr && pos_index != nullptr) {
922 // Emulate effect of Parse::adjust_map_after_if.
923 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
924 (*pos_index) = _gvn.transform(ccast);
925 }
926 return is_neg;
927 }
928
929 // Make sure that 'position' is a valid limit index, in [0..length].
930 // There are two equivalent plans for checking this:
931 // A. (offset + copyLength) unsigned<= arrayLength
932 // B. offset <= (arrayLength - copyLength)
933 // We require that all of the values above, except for the sum and
934 // difference, are already known to be non-negative.
935 // Plan A is robust in the face of overflow, if offset and copyLength
936 // are both hugely positive.
937 //
938 // Plan B is less direct and intuitive, but it does not overflow at
939 // all, since the difference of two non-negatives is always
940 // representable. Whenever Java methods must perform the equivalent
941 // check they generally use Plan B instead of Plan A.
942 // For the moment we use Plan A.
943 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
944 Node* subseq_length,
945 Node* array_length,
946 RegionNode* region,
947 bool with_opaque) {
948 if (stopped())
949 return nullptr; // already stopped
950 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
951 if (zero_offset && subseq_length->eqv_uncast(array_length))
952 return nullptr; // common case of whole-array copy
953 Node* last = subseq_length;
954 if (!zero_offset) // last += offset
955 last = _gvn.transform(new AddINode(last, offset));
956 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
957 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
958 if (with_opaque) {
959 bol_lt = _gvn.transform(new OpaqueConstantBoolNode(C, bol_lt, false));
960 }
961 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
962 return is_over;
963 }
964
965 // Emit range checks for the given String.value byte array
966 void LibraryCallKit::generate_string_range_check(Node* array,
967 Node* offset,
968 Node* count,
969 bool char_count,
970 RegionNode* region) {
971 if (stopped()) {
972 return; // already stopped
973 }
974 if (char_count) {
975 // Convert char count to byte count
976 count = _gvn.transform(new LShiftINode(count, intcon(1)));
977 }
978 // Offset and count must not be negative
979 generate_negative_guard(offset, region, nullptr, true);
980 generate_negative_guard(count, region, nullptr, true);
981 // Offset + count must not exceed length of array
982 generate_limit_guard(offset, count, load_array_length(array), region, true);
983 }
984
985 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
986 bool is_immutable) {
987 ciKlass* thread_klass = env()->Thread_klass();
988 const Type* thread_type
989 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
990
991 Node* thread = _gvn.transform(new ThreadLocalNode());
992 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
993 tls_output = thread;
994
995 Node* thread_obj_handle
996 = (is_immutable
997 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
998 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
999 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
1000 thread_obj_handle = _gvn.transform(thread_obj_handle);
1001
1002 DecoratorSet decorators = IN_NATIVE;
1003 if (is_immutable) {
1004 decorators |= C2_IMMUTABLE_MEMORY;
1005 }
1006 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
1007 }
1008
1009 //--------------------------generate_current_thread--------------------
1010 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
1011 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
1012 /*is_immutable*/false);
1013 }
1014
1015 //--------------------------generate_virtual_thread--------------------
1016 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
1017 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
1018 !C->method()->changes_current_thread());
1019 }
1020
1021 //------------------------------make_string_method_node------------------------
1022 // Helper method for String intrinsic functions. This version is called with
1023 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1024 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1025 // containing the lengths of str1 and str2.
1026 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1027 Node* result = nullptr;
1028 switch (opcode) {
1029 case Op_StrIndexOf:
1030 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1031 str1_start, cnt1, str2_start, cnt2, ae);
1032 break;
1033 case Op_StrComp:
1034 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1035 str1_start, cnt1, str2_start, cnt2, ae);
1036 break;
1037 case Op_StrEquals:
1038 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1039 // Use the constant length if there is one because optimized match rule may exist.
1040 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1041 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1042 break;
1043 default:
1044 ShouldNotReachHere();
1045 return nullptr;
1046 }
1047
1048 // All these intrinsics have checks.
1049 C->set_has_split_ifs(true); // Has chance for split-if optimization
1050 clear_upper_avx();
1051
1052 return _gvn.transform(result);
1053 }
1054
1055 //------------------------------inline_string_compareTo------------------------
1056 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1057 Node* arg1 = argument(0);
1058 Node* arg2 = argument(1);
1059
1060 arg1 = must_be_not_null(arg1, true);
1061 arg2 = must_be_not_null(arg2, true);
1062
1063 // Get start addr and length of first argument
1064 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1065 Node* arg1_cnt = load_array_length(arg1);
1066
1067 // Get start addr and length of second argument
1068 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1069 Node* arg2_cnt = load_array_length(arg2);
1070
1071 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1072 set_result(result);
1073 return true;
1074 }
1075
1076 //------------------------------inline_string_equals------------------------
1077 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1078 Node* arg1 = argument(0);
1079 Node* arg2 = argument(1);
1080
1081 // paths (plus control) merge
1082 RegionNode* region = new RegionNode(3);
1083 Node* phi = new PhiNode(region, TypeInt::BOOL);
1084
1085 if (!stopped()) {
1086
1087 arg1 = must_be_not_null(arg1, true);
1088 arg2 = must_be_not_null(arg2, true);
1089
1090 // Get start addr and length of first argument
1091 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1092 Node* arg1_cnt = load_array_length(arg1);
1093
1094 // Get start addr and length of second argument
1095 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1096 Node* arg2_cnt = load_array_length(arg2);
1097
1098 // Check for arg1_cnt != arg2_cnt
1099 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1100 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1101 Node* if_ne = generate_slow_guard(bol, nullptr);
1102 if (if_ne != nullptr) {
1103 phi->init_req(2, intcon(0));
1104 region->init_req(2, if_ne);
1105 }
1106
1107 // Check for count == 0 is done by assembler code for StrEquals.
1108
1109 if (!stopped()) {
1110 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1111 phi->init_req(1, equals);
1112 region->init_req(1, control());
1113 }
1114 }
1115
1116 // post merge
1117 set_control(_gvn.transform(region));
1118 record_for_igvn(region);
1119
1120 set_result(_gvn.transform(phi));
1121 return true;
1122 }
1123
1124 //------------------------------inline_array_equals----------------------------
1125 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1126 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1127 Node* arg1 = argument(0);
1128 Node* arg2 = argument(1);
1129
1130 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1131 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1132 clear_upper_avx();
1133
1134 return true;
1135 }
1136
1137
1138 //------------------------------inline_countPositives------------------------------
1139 // int java.lang.StringCoding#countPositives0(byte[] ba, int off, int len)
1140 bool LibraryCallKit::inline_countPositives() {
1141 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1142 // no receiver since it is static method
1143 Node* ba = argument(0);
1144 Node* offset = argument(1);
1145 Node* len = argument(2);
1146
1147 ba = must_be_not_null(ba, true);
1148 RegionNode* bailout = create_bailout();
1149 generate_string_range_check(ba, offset, len, false, bailout);
1150 if (check_bailout(bailout)) {
1151 return true;
1152 }
1153
1154 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1155 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1156 set_result(_gvn.transform(result));
1157 clear_upper_avx();
1158 return true;
1159 }
1160
1161 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1162 Node* index = argument(0);
1163 Node* length = bt == T_INT ? argument(1) : argument(2);
1164 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1165 return false;
1166 }
1167
1168 // check that length is positive
1169 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1170 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1171
1172 {
1173 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1174 uncommon_trap(Deoptimization::Reason_intrinsic,
1175 Deoptimization::Action_make_not_entrant);
1176 }
1177
1178 if (stopped()) {
1179 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1180 return true;
1181 }
1182
1183 // length is now known positive, add a cast node to make this explicit
1184 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1185 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1186 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1187 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1188 casted_length = _gvn.transform(casted_length);
1189 replace_in_map(length, casted_length);
1190 length = casted_length;
1191
1192 // Use an unsigned comparison for the range check itself
1193 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1194 BoolTest::mask btest = BoolTest::lt;
1195 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1196 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1197 _gvn.set_type(rc, rc->Value(&_gvn));
1198 if (!rc_bool->is_Con()) {
1199 record_for_igvn(rc);
1200 }
1201 set_control(_gvn.transform(new IfTrueNode(rc)));
1202 {
1203 PreserveJVMState pjvms(this);
1204 set_control(_gvn.transform(new IfFalseNode(rc)));
1205 uncommon_trap(Deoptimization::Reason_range_check,
1206 Deoptimization::Action_make_not_entrant);
1207 }
1208
1209 if (stopped()) {
1210 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1211 return true;
1212 }
1213
1214 // index is now known to be >= 0 and < length, cast it
1215 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1216 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1217 ConstraintCastNode::DependencyType::FloatingNarrowing, bt);
1218 result = _gvn.transform(result);
1219 set_result(result);
1220 replace_in_map(index, result);
1221 return true;
1222 }
1223
1224 //------------------------------inline_string_indexOf------------------------
1225 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1226 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1227 return false;
1228 }
1229 Node* src = argument(0);
1230 Node* tgt = argument(1);
1231
1232 // Make the merge point
1233 RegionNode* result_rgn = new RegionNode(4);
1234 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1235
1236 src = must_be_not_null(src, true);
1237 tgt = must_be_not_null(tgt, true);
1238
1239 // Get start addr and length of source string
1240 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1241 Node* src_count = load_array_length(src);
1242
1243 // Get start addr and length of substring
1244 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1245 Node* tgt_count = load_array_length(tgt);
1246
1247 Node* result = nullptr;
1248 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1249
1250 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1251 // Divide src size by 2 if String is UTF16 encoded
1252 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1253 }
1254 if (ae == StrIntrinsicNode::UU) {
1255 // Divide substring size by 2 if String is UTF16 encoded
1256 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1257 }
1258
1259 if (call_opt_stub) {
1260 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1261 StubRoutines::_string_indexof_array[ae],
1262 "stringIndexOf", TypePtr::BOTTOM, src_start,
1263 src_count, tgt_start, tgt_count);
1264 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1265 } else {
1266 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1267 result_rgn, result_phi, ae);
1268 }
1269 if (result != nullptr) {
1270 result_phi->init_req(3, result);
1271 result_rgn->init_req(3, control());
1272 }
1273 set_control(_gvn.transform(result_rgn));
1274 record_for_igvn(result_rgn);
1275 set_result(_gvn.transform(result_phi));
1276
1277 return true;
1278 }
1279
1280 //-----------------------------inline_string_indexOfI-----------------------
1281 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1282 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1283 return false;
1284 }
1285
1286 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1287 Node* src = argument(0); // byte[]
1288 Node* src_count = argument(1); // char count
1289 Node* tgt = argument(2); // byte[]
1290 Node* tgt_count = argument(3); // char count
1291 Node* from_index = argument(4); // char index
1292
1293 src = must_be_not_null(src, true);
1294 tgt = must_be_not_null(tgt, true);
1295
1296 // Multiply byte array index by 2 if String is UTF16 encoded
1297 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1298 src_count = _gvn.transform(new SubINode(src_count, from_index));
1299 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1300 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1301
1302 // Range checks
1303 RegionNode* bailout = create_bailout();
1304 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL, bailout);
1305 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU, bailout);
1306 if (check_bailout(bailout)) {
1307 return true;
1308 }
1309
1310 RegionNode* region = new RegionNode(5);
1311 Node* phi = new PhiNode(region, TypeInt::INT);
1312 Node* result = nullptr;
1313
1314 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1315
1316 if (call_opt_stub) {
1317 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1318 StubRoutines::_string_indexof_array[ae],
1319 "stringIndexOf", TypePtr::BOTTOM, src_start,
1320 src_count, tgt_start, tgt_count);
1321 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1322 } else {
1323 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1324 region, phi, ae);
1325 }
1326 if (result != nullptr) {
1327 // The result is index relative to from_index if substring was found, -1 otherwise.
1328 // Generate code which will fold into cmove.
1329 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1330 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1331
1332 Node* if_lt = generate_slow_guard(bol, nullptr);
1333 if (if_lt != nullptr) {
1334 // result == -1
1335 phi->init_req(3, result);
1336 region->init_req(3, if_lt);
1337 }
1338 if (!stopped()) {
1339 result = _gvn.transform(new AddINode(result, from_index));
1340 phi->init_req(4, result);
1341 region->init_req(4, control());
1342 }
1343 }
1344
1345 set_control(_gvn.transform(region));
1346 record_for_igvn(region);
1347 set_result(_gvn.transform(phi));
1348 clear_upper_avx();
1349
1350 return true;
1351 }
1352
1353 // Create StrIndexOfNode with fast path checks
1354 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1355 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1356 // Check for substr count > string count
1357 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1358 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1359 Node* if_gt = generate_slow_guard(bol, nullptr);
1360 if (if_gt != nullptr) {
1361 phi->init_req(1, intcon(-1));
1362 region->init_req(1, if_gt);
1363 }
1364 if (!stopped()) {
1365 // Check for substr count == 0
1366 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1367 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1368 Node* if_zero = generate_slow_guard(bol, nullptr);
1369 if (if_zero != nullptr) {
1370 phi->init_req(2, intcon(0));
1371 region->init_req(2, if_zero);
1372 }
1373 }
1374 if (!stopped()) {
1375 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1376 }
1377 return nullptr;
1378 }
1379
1380 //-----------------------------inline_string_indexOfChar-----------------------
1381 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1382 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1383 return false;
1384 }
1385 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1386 return false;
1387 }
1388 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1389 Node* src = argument(0); // byte[]
1390 Node* int_ch = argument(1);
1391 Node* from_index = argument(2);
1392 Node* max = argument(3);
1393
1394 src = must_be_not_null(src, true);
1395
1396 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1397 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1398 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1399
1400 // Range checks
1401 RegionNode* bailout = create_bailout();
1402 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U, bailout);
1403 if (check_bailout(bailout)) {
1404 return true;
1405 }
1406
1407 // Check for int_ch >= 0
1408 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1409 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1410 {
1411 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1412 uncommon_trap(Deoptimization::Reason_intrinsic,
1413 Deoptimization::Action_maybe_recompile);
1414 }
1415 if (stopped()) {
1416 return true;
1417 }
1418
1419 RegionNode* region = new RegionNode(3);
1420 Node* phi = new PhiNode(region, TypeInt::INT);
1421
1422 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1423 C->set_has_split_ifs(true); // Has chance for split-if optimization
1424 _gvn.transform(result);
1425
1426 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1427 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1428
1429 Node* if_lt = generate_slow_guard(bol, nullptr);
1430 if (if_lt != nullptr) {
1431 // result == -1
1432 phi->init_req(2, result);
1433 region->init_req(2, if_lt);
1434 }
1435 if (!stopped()) {
1436 result = _gvn.transform(new AddINode(result, from_index));
1437 phi->init_req(1, result);
1438 region->init_req(1, control());
1439 }
1440 set_control(_gvn.transform(region));
1441 record_for_igvn(region);
1442 set_result(_gvn.transform(phi));
1443 clear_upper_avx();
1444
1445 return true;
1446 }
1447 //---------------------------inline_string_copy---------------------
1448 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1449 // int StringUTF16.compress0(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1450 // int StringUTF16.compress0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1451 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1452 // void StringLatin1.inflate0(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1453 // void StringLatin1.inflate0(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1454 bool LibraryCallKit::inline_string_copy(bool compress) {
1455 int nargs = 5; // 2 oops, 3 ints
1456 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1457
1458 Node* src = argument(0);
1459 Node* src_offset = argument(1);
1460 Node* dst = argument(2);
1461 Node* dst_offset = argument(3);
1462 Node* length = argument(4);
1463
1464 // Check for allocation before we add nodes that would confuse
1465 // tightly_coupled_allocation()
1466 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1467
1468 // Figure out the size and type of the elements we will be copying.
1469 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1470 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1471 if (src_type == nullptr || dst_type == nullptr) {
1472 return false;
1473 }
1474 BasicType src_elem = src_type->elem()->array_element_basic_type();
1475 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1476 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1477 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1478 "Unsupported array types for inline_string_copy");
1479
1480 src = must_be_not_null(src, true);
1481 dst = must_be_not_null(dst, true);
1482
1483 // Convert char[] offsets to byte[] offsets
1484 bool convert_src = (compress && src_elem == T_BYTE);
1485 bool convert_dst = (!compress && dst_elem == T_BYTE);
1486 if (convert_src) {
1487 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1488 } else if (convert_dst) {
1489 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1490 }
1491
1492 // Range checks
1493 RegionNode* bailout = create_bailout();
1494 generate_string_range_check(src, src_offset, length, convert_src, bailout);
1495 generate_string_range_check(dst, dst_offset, length, convert_dst, bailout);
1496 if (check_bailout(bailout)) {
1497 return true;
1498 }
1499
1500 Node* src_start = array_element_address(src, src_offset, src_elem);
1501 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1502 // 'src_start' points to src array + scaled offset
1503 // 'dst_start' points to dst array + scaled offset
1504 Node* count = nullptr;
1505 if (compress) {
1506 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1507 } else {
1508 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1509 }
1510
1511 if (alloc != nullptr) {
1512 if (alloc->maybe_set_complete(&_gvn)) {
1513 // "You break it, you buy it."
1514 InitializeNode* init = alloc->initialization();
1515 assert(init->is_complete(), "we just did this");
1516 init->set_complete_with_arraycopy();
1517 assert(dst->is_CheckCastPP(), "sanity");
1518 assert(dst->in(0)->in(0) == init, "dest pinned");
1519 }
1520 // Do not let stores that initialize this object be reordered with
1521 // a subsequent store that would make this object accessible by
1522 // other threads.
1523 // Record what AllocateNode this StoreStore protects so that
1524 // escape analysis can go from the MemBarStoreStoreNode to the
1525 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1526 // based on the escape status of the AllocateNode.
1527 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1528 }
1529 if (compress) {
1530 set_result(_gvn.transform(count));
1531 }
1532 clear_upper_avx();
1533
1534 return true;
1535 }
1536
1537 #ifdef _LP64
1538 #define XTOP ,top() /*additional argument*/
1539 #else //_LP64
1540 #define XTOP /*no additional argument*/
1541 #endif //_LP64
1542
1543 //------------------------inline_string_toBytesU--------------------------
1544 // public static byte[] StringUTF16.toBytes0(char[] value, int off, int len)
1545 bool LibraryCallKit::inline_string_toBytesU() {
1546 // Get the arguments.
1547 assert(callee()->signature()->size() == 3, "character array encoder requires 3 arguments");
1548 Node* value = argument(0);
1549 Node* offset = argument(1);
1550 Node* length = argument(2);
1551
1552 Node* newcopy = nullptr;
1553
1554 // Set the original stack and the reexecute bit for the interpreter to reexecute
1555 // the bytecode that invokes StringUTF16.toBytes0() if deoptimization happens.
1556 { PreserveReexecuteState preexecs(this);
1557 jvms()->set_should_reexecute(true);
1558
1559 value = must_be_not_null(value, true);
1560 RegionNode* bailout = create_bailout();
1561 generate_negative_guard(offset, bailout, nullptr, true);
1562 generate_negative_guard(length, bailout, nullptr, true);
1563 generate_limit_guard(offset, length, load_array_length(value), bailout, true);
1564 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1565 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout, true);
1566 if (check_bailout(bailout)) {
1567 return true;
1568 }
1569
1570 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1571 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1572 newcopy = new_array(klass_node, size, 0); // no arguments to push
1573 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1574 guarantee(alloc != nullptr, "created above");
1575
1576 // Calculate starting addresses.
1577 Node* src_start = array_element_address(value, offset, T_CHAR);
1578 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1579
1580 // Check if dst array address is aligned to HeapWordSize
1581 bool aligned = (arrayOopDesc::base_offset_in_bytes(T_BYTE) % HeapWordSize == 0);
1582 // If true, then check if src array address is aligned to HeapWordSize
1583 if (aligned) {
1584 const TypeInt* toffset = gvn().type(offset)->is_int();
1585 aligned = toffset->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) +
1586 toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1587 }
1588
1589 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1590 const char* copyfunc_name = "arraycopy";
1591 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1592 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1593 OptoRuntime::fast_arraycopy_Type(),
1594 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1595 src_start, dst_start, ConvI2X(length) XTOP);
1596 // Do not let reads from the cloned object float above the arraycopy.
1597 if (alloc->maybe_set_complete(&_gvn)) {
1598 // "You break it, you buy it."
1599 InitializeNode* init = alloc->initialization();
1600 assert(init->is_complete(), "we just did this");
1601 init->set_complete_with_arraycopy();
1602 assert(newcopy->is_CheckCastPP(), "sanity");
1603 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1604 }
1605 // Do not let stores that initialize this object be reordered with
1606 // a subsequent store that would make this object accessible by
1607 // other threads.
1608 // Record what AllocateNode this StoreStore protects so that
1609 // escape analysis can go from the MemBarStoreStoreNode to the
1610 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1611 // based on the escape status of the AllocateNode.
1612 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1613 } // original reexecute is set back here
1614
1615 C->set_has_split_ifs(true); // Has chance for split-if optimization
1616 if (!stopped()) {
1617 set_result(newcopy);
1618 }
1619 clear_upper_avx();
1620
1621 return true;
1622 }
1623
1624 //------------------------inline_string_getCharsU--------------------------
1625 // public void StringUTF16.getChars0(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1626 bool LibraryCallKit::inline_string_getCharsU() {
1627 assert(callee()->signature()->size() == 5, "StringUTF16.getChars0() has 5 arguments");
1628 // Get the arguments.
1629 Node* src = argument(0);
1630 Node* src_begin = argument(1);
1631 Node* src_end = argument(2); // exclusive offset (i < src_end)
1632 Node* dst = argument(3);
1633 Node* dst_begin = argument(4);
1634
1635 // Check for allocation before we add nodes that would confuse
1636 // tightly_coupled_allocation()
1637 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1638
1639 // Check if a null path was taken unconditionally.
1640 src = must_be_not_null(src, true);
1641 dst = must_be_not_null(dst, true);
1642 if (stopped()) {
1643 return true;
1644 }
1645
1646 // Get length and convert char[] offset to byte[] offset
1647 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1648 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1649
1650 // Range checks
1651 RegionNode* bailout = create_bailout();
1652 generate_string_range_check(src, src_begin, length, true, bailout);
1653 generate_string_range_check(dst, dst_begin, length, false, bailout);
1654 if (check_bailout(bailout)) {
1655 return true;
1656 }
1657
1658 // Calculate starting addresses.
1659 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1660 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1661
1662 // Check if array addresses are aligned to HeapWordSize
1663 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1664 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1665 bool aligned = tsrc->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_BYTE) + tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1666 tdst->is_con() && ((arrayOopDesc::base_offset_in_bytes(T_CHAR) + tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1667
1668 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1669 const char* copyfunc_name = "arraycopy";
1670 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1671 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1672 OptoRuntime::fast_arraycopy_Type(),
1673 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1674 src_start, dst_start, ConvI2X(length) XTOP);
1675 // Do not let reads from the cloned object float above the arraycopy.
1676 if (alloc != nullptr) {
1677 if (alloc->maybe_set_complete(&_gvn)) {
1678 // "You break it, you buy it."
1679 InitializeNode* init = alloc->initialization();
1680 assert(init->is_complete(), "we just did this");
1681 init->set_complete_with_arraycopy();
1682 assert(dst->is_CheckCastPP(), "sanity");
1683 assert(dst->in(0)->in(0) == init, "dest pinned");
1684 }
1685 // Do not let stores that initialize this object be reordered with
1686 // a subsequent store that would make this object accessible by
1687 // other threads.
1688 // Record what AllocateNode this StoreStore protects so that
1689 // escape analysis can go from the MemBarStoreStoreNode to the
1690 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1691 // based on the escape status of the AllocateNode.
1692 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1693 } else {
1694 insert_mem_bar(Op_MemBarCPUOrder);
1695 }
1696
1697 C->set_has_split_ifs(true); // Has chance for split-if optimization
1698 return true;
1699 }
1700
1701 //----------------------inline_string_char_access----------------------------
1702 // Store/Load char to/from byte[] array.
1703 // static void StringUTF16.putChar(byte[] val, int index, int c)
1704 // static char StringUTF16.getChar(byte[] val, int index)
1705 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1706 Node* ch;
1707 if (is_store) {
1708 assert(callee()->signature()->size() == 3, "StringUTF16.putChar() has 3 arguments");
1709 ch = argument(2);
1710 } else {
1711 assert(callee()->signature()->size() == 2, "StringUTF16.getChar() has 2 arguments");
1712 ch = nullptr;
1713 }
1714 Node* value = argument(0);
1715 Node* index = argument(1);
1716
1717 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1718 // correctly requires matched array shapes.
1719 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1720 "sanity: byte[] and char[] bases agree");
1721 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1722 "sanity: byte[] and char[] scales agree");
1723
1724 // Bail when getChar over constants is requested: constant folding would
1725 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1726 // Java method would constant fold nicely instead.
1727 if (!is_store && value->is_Con() && index->is_Con()) {
1728 return false;
1729 }
1730
1731 // Save state and restore on bailout
1732 SavedState old_state(this);
1733
1734 value = must_be_not_null(value, true);
1735
1736 Node* adr = array_element_address(value, index, T_CHAR);
1737 if (adr->is_top()) {
1738 return false;
1739 }
1740 old_state.discard();
1741 if (is_store) {
1742 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1743 } else {
1744 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);
1745 set_result(ch);
1746 }
1747 return true;
1748 }
1749
1750
1751 //------------------------------inline_math-----------------------------------
1752 // public static double Math.abs(double)
1753 // public static double Math.sqrt(double)
1754 // public static double Math.log(double)
1755 // public static double Math.log10(double)
1756 // public static double Math.round(double)
1757 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1758 Node* arg = argument(0);
1759 Node* n = nullptr;
1760 switch (id) {
1761 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1762 case vmIntrinsics::_dsqrt:
1763 case vmIntrinsics::_dsqrt_strict:
1764 n = new SqrtDNode(C, control(), arg); break;
1765 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1766 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1767 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1768 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1769 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1770 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1771 default: fatal_unexpected_iid(id); break;
1772 }
1773 set_result(_gvn.transform(n));
1774 return true;
1775 }
1776
1777 //------------------------------inline_math-----------------------------------
1778 // public static float Math.abs(float)
1779 // public static int Math.abs(int)
1780 // public static long Math.abs(long)
1781 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1782 Node* arg = argument(0);
1783 Node* n = nullptr;
1784 switch (id) {
1785 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1786 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1787 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1788 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1789 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1790 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1791 default: fatal_unexpected_iid(id); break;
1792 }
1793 set_result(_gvn.transform(n));
1794 return true;
1795 }
1796
1797 //------------------------------runtime_math-----------------------------
1798 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1799 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1800 "must be (DD)D or (D)D type");
1801
1802 // Inputs
1803 Node* a = argument(0);
1804 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1805
1806 const TypePtr* no_memory_effects = nullptr;
1807 Node* trig = make_runtime_call(RC_LEAF | RC_PURE, call_type, funcAddr, funcName,
1808 no_memory_effects,
1809 a, top(), b, b ? top() : nullptr);
1810 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1811 #ifdef ASSERT
1812 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1813 assert(value_top == top(), "second value must be top");
1814 #endif
1815
1816 set_result(value);
1817 return true;
1818 }
1819
1820 //------------------------------inline_math_pow-----------------------------
1821 bool LibraryCallKit::inline_math_pow() {
1822 Node* exp = argument(2);
1823 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1824 if (d != nullptr) {
1825 if (d->getd() == 2.0) {
1826 // Special case: pow(x, 2.0) => x * x
1827 Node* base = argument(0);
1828 set_result(_gvn.transform(new MulDNode(base, base)));
1829 return true;
1830 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1831 // Special case: pow(x, 0.5) => sqrt(x)
1832 Node* base = argument(0);
1833 Node* zero = _gvn.zerocon(T_DOUBLE);
1834
1835 RegionNode* region = new RegionNode(3);
1836 Node* phi = new PhiNode(region, Type::DOUBLE);
1837
1838 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1839 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1840 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1841 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1842 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1843
1844 Node* if_pow = generate_slow_guard(test, nullptr);
1845 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1846 phi->init_req(1, value_sqrt);
1847 region->init_req(1, control());
1848
1849 if (if_pow != nullptr) {
1850 set_control(if_pow);
1851 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1852 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1853 const TypePtr* no_memory_effects = nullptr;
1854 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1855 no_memory_effects, base, top(), exp, top());
1856 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1857 #ifdef ASSERT
1858 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1859 assert(value_top == top(), "second value must be top");
1860 #endif
1861 phi->init_req(2, value_pow);
1862 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1863 }
1864
1865 C->set_has_split_ifs(true); // Has chance for split-if optimization
1866 set_control(_gvn.transform(region));
1867 record_for_igvn(region);
1868 set_result(_gvn.transform(phi));
1869
1870 return true;
1871 }
1872 }
1873
1874 return StubRoutines::dpow() != nullptr ?
1875 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1876 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1877 }
1878
1879 //------------------------------inline_math_native-----------------------------
1880 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1881 switch (id) {
1882 case vmIntrinsics::_dsin:
1883 return StubRoutines::dsin() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1886 case vmIntrinsics::_dcos:
1887 return StubRoutines::dcos() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1889 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1890 case vmIntrinsics::_dtan:
1891 return StubRoutines::dtan() != nullptr ?
1892 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1893 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1894 case vmIntrinsics::_dsinh:
1895 return StubRoutines::dsinh() != nullptr ?
1896 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsinh(), "dsinh") : false;
1897 case vmIntrinsics::_dtanh:
1898 return StubRoutines::dtanh() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1900 case vmIntrinsics::_dcbrt:
1901 return StubRoutines::dcbrt() != nullptr ?
1902 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcbrt(), "dcbrt") : false;
1903 case vmIntrinsics::_dexp:
1904 return StubRoutines::dexp() != nullptr ?
1905 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1906 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1907 case vmIntrinsics::_dlog:
1908 return StubRoutines::dlog() != nullptr ?
1909 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1910 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1911 case vmIntrinsics::_dlog10:
1912 return StubRoutines::dlog10() != nullptr ?
1913 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1914 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1915
1916 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1917 case vmIntrinsics::_ceil:
1918 case vmIntrinsics::_floor:
1919 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1920
1921 case vmIntrinsics::_dsqrt:
1922 case vmIntrinsics::_dsqrt_strict:
1923 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1924 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1925 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1926 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1927 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1928
1929 case vmIntrinsics::_dpow: return inline_math_pow();
1930 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1931 case vmIntrinsics::_fcopySign: return inline_math(id);
1932 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1933 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1934 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1935
1936 // These intrinsics are not yet correctly implemented
1937 case vmIntrinsics::_datan2:
1938 return false;
1939
1940 default:
1941 fatal_unexpected_iid(id);
1942 return false;
1943 }
1944 }
1945
1946 //----------------------------inline_notify-----------------------------------*
1947 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1948 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1949 address func;
1950 if (id == vmIntrinsics::_notify) {
1951 func = OptoRuntime::monitor_notify_Java();
1952 } else {
1953 func = OptoRuntime::monitor_notifyAll_Java();
1954 }
1955 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1956 make_slow_call_ex(call, env()->Throwable_klass(), false);
1957 return true;
1958 }
1959
1960
1961 //----------------------------inline_min_max-----------------------------------
1962 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1963 Node* a = nullptr;
1964 Node* b = nullptr;
1965 Node* n = nullptr;
1966 switch (id) {
1967 case vmIntrinsics::_min:
1968 case vmIntrinsics::_max:
1969 case vmIntrinsics::_minF:
1970 case vmIntrinsics::_maxF:
1971 case vmIntrinsics::_minF_strict:
1972 case vmIntrinsics::_maxF_strict:
1973 case vmIntrinsics::_min_strict:
1974 case vmIntrinsics::_max_strict:
1975 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1976 a = argument(0);
1977 b = argument(1);
1978 break;
1979 case vmIntrinsics::_minD:
1980 case vmIntrinsics::_maxD:
1981 case vmIntrinsics::_minD_strict:
1982 case vmIntrinsics::_maxD_strict:
1983 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1984 a = argument(0);
1985 b = argument(2);
1986 break;
1987 case vmIntrinsics::_minL:
1988 case vmIntrinsics::_maxL:
1989 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1990 a = argument(0);
1991 b = argument(2);
1992 break;
1993 default:
1994 fatal_unexpected_iid(id);
1995 break;
1996 }
1997
1998 switch (id) {
1999 case vmIntrinsics::_min:
2000 case vmIntrinsics::_min_strict:
2001 n = new MinINode(a, b);
2002 break;
2003 case vmIntrinsics::_max:
2004 case vmIntrinsics::_max_strict:
2005 n = new MaxINode(a, b);
2006 break;
2007 case vmIntrinsics::_minF:
2008 case vmIntrinsics::_minF_strict:
2009 n = new MinFNode(a, b);
2010 break;
2011 case vmIntrinsics::_maxF:
2012 case vmIntrinsics::_maxF_strict:
2013 n = new MaxFNode(a, b);
2014 break;
2015 case vmIntrinsics::_minD:
2016 case vmIntrinsics::_minD_strict:
2017 n = new MinDNode(a, b);
2018 break;
2019 case vmIntrinsics::_maxD:
2020 case vmIntrinsics::_maxD_strict:
2021 n = new MaxDNode(a, b);
2022 break;
2023 case vmIntrinsics::_minL:
2024 n = new MinLNode(_gvn.C, a, b);
2025 break;
2026 case vmIntrinsics::_maxL:
2027 n = new MaxLNode(_gvn.C, a, b);
2028 break;
2029 default:
2030 fatal_unexpected_iid(id);
2031 break;
2032 }
2033
2034 set_result(_gvn.transform(n));
2035 return true;
2036 }
2037
2038 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2039 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2040 env()->ArithmeticException_instance())) {
2041 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2042 // so let's bail out intrinsic rather than risking deopting again.
2043 return false;
2044 }
2045
2046 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2047 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2048 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2049 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2050
2051 {
2052 PreserveJVMState pjvms(this);
2053 PreserveReexecuteState preexecs(this);
2054 jvms()->set_should_reexecute(true);
2055
2056 set_control(slow_path);
2057 set_i_o(i_o());
2058
2059 builtin_throw(Deoptimization::Reason_intrinsic,
2060 env()->ArithmeticException_instance(),
2061 /*allow_too_many_traps*/ false);
2062 }
2063
2064 set_control(fast_path);
2065 set_result(math);
2066 return true;
2067 }
2068
2069 template <typename OverflowOp>
2070 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2071 typedef typename OverflowOp::MathOp MathOp;
2072
2073 MathOp* mathOp = new MathOp(arg1, arg2);
2074 Node* operation = _gvn.transform( mathOp );
2075 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2076 return inline_math_mathExact(operation, ofcheck);
2077 }
2078
2079 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2080 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2081 }
2082
2083 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2084 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2085 }
2086
2087 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2088 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2089 }
2090
2091 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2092 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2093 }
2094
2095 bool LibraryCallKit::inline_math_negateExactI() {
2096 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2097 }
2098
2099 bool LibraryCallKit::inline_math_negateExactL() {
2100 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2101 }
2102
2103 bool LibraryCallKit::inline_math_multiplyExactI() {
2104 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2105 }
2106
2107 bool LibraryCallKit::inline_math_multiplyExactL() {
2108 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2109 }
2110
2111 bool LibraryCallKit::inline_math_multiplyHigh() {
2112 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2113 return true;
2114 }
2115
2116 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2117 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2118 return true;
2119 }
2120
2121 inline int
2122 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2123 const TypePtr* base_type = TypePtr::NULL_PTR;
2124 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2125 if (base_type == nullptr) {
2126 // Unknown type.
2127 return Type::AnyPtr;
2128 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2129 // Since this is a null+long form, we have to switch to a rawptr.
2130 base = _gvn.transform(new CastX2PNode(offset));
2131 offset = MakeConX(0);
2132 return Type::RawPtr;
2133 } else if (base_type->base() == Type::RawPtr) {
2134 return Type::RawPtr;
2135 } else if (base_type->isa_oopptr()) {
2136 // Base is never null => always a heap address.
2137 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2138 return Type::OopPtr;
2139 }
2140 // Offset is small => always a heap address.
2141 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2142 if (offset_type != nullptr &&
2143 base_type->offset() == 0 && // (should always be?)
2144 offset_type->_lo >= 0 &&
2145 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2146 return Type::OopPtr;
2147 } else if (type == T_OBJECT) {
2148 // off heap access to an oop doesn't make any sense. Has to be on
2149 // heap.
2150 return Type::OopPtr;
2151 }
2152 // Otherwise, it might either be oop+off or null+addr.
2153 return Type::AnyPtr;
2154 } else {
2155 // No information:
2156 return Type::AnyPtr;
2157 }
2158 }
2159
2160 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2161 Node* uncasted_base = base;
2162 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2163 if (kind == Type::RawPtr) {
2164 return basic_plus_adr(top(), uncasted_base, offset);
2165 } else if (kind == Type::AnyPtr) {
2166 assert(base == uncasted_base, "unexpected base change");
2167 if (can_cast) {
2168 if (!_gvn.type(base)->speculative_maybe_null() &&
2169 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2170 // According to profiling, this access is always on
2171 // heap. Casting the base to not null and thus avoiding membars
2172 // around the access should allow better optimizations
2173 Node* null_ctl = top();
2174 base = null_check_oop(base, &null_ctl, true, true, true);
2175 assert(null_ctl->is_top(), "no null control here");
2176 return basic_plus_adr(base, offset);
2177 } else if (_gvn.type(base)->speculative_always_null() &&
2178 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2179 // According to profiling, this access is always off
2180 // heap.
2181 base = null_assert(base);
2182 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2183 offset = MakeConX(0);
2184 return basic_plus_adr(top(), raw_base, offset);
2185 }
2186 }
2187 // We don't know if it's an on heap or off heap access. Fall back
2188 // to raw memory access.
2189 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2190 return basic_plus_adr(top(), raw, offset);
2191 } else {
2192 assert(base == uncasted_base, "unexpected base change");
2193 // We know it's an on heap access so base can't be null
2194 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2195 base = must_be_not_null(base, true);
2196 }
2197 return basic_plus_adr(base, offset);
2198 }
2199 }
2200
2201 //--------------------------inline_number_methods-----------------------------
2202 // inline int Integer.numberOfLeadingZeros(int)
2203 // inline int Long.numberOfLeadingZeros(long)
2204 //
2205 // inline int Integer.numberOfTrailingZeros(int)
2206 // inline int Long.numberOfTrailingZeros(long)
2207 //
2208 // inline int Integer.bitCount(int)
2209 // inline int Long.bitCount(long)
2210 //
2211 // inline char Character.reverseBytes(char)
2212 // inline short Short.reverseBytes(short)
2213 // inline int Integer.reverseBytes(int)
2214 // inline long Long.reverseBytes(long)
2215 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2216 Node* arg = argument(0);
2217 Node* n = nullptr;
2218 switch (id) {
2219 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2220 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2221 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2222 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2223 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2224 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2225 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2226 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2227 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2228 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2229 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2230 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2231 default: fatal_unexpected_iid(id); break;
2232 }
2233 set_result(_gvn.transform(n));
2234 return true;
2235 }
2236
2237 //--------------------------inline_bitshuffle_methods-----------------------------
2238 // inline int Integer.compress(int, int)
2239 // inline int Integer.expand(int, int)
2240 // inline long Long.compress(long, long)
2241 // inline long Long.expand(long, long)
2242 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2243 Node* n = nullptr;
2244 switch (id) {
2245 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2246 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2247 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2248 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2249 default: fatal_unexpected_iid(id); break;
2250 }
2251 set_result(_gvn.transform(n));
2252 return true;
2253 }
2254
2255 //--------------------------inline_number_methods-----------------------------
2256 // inline int Integer.compareUnsigned(int, int)
2257 // inline int Long.compareUnsigned(long, long)
2258 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2259 Node* arg1 = argument(0);
2260 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2261 Node* n = nullptr;
2262 switch (id) {
2263 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2264 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2265 default: fatal_unexpected_iid(id); break;
2266 }
2267 set_result(_gvn.transform(n));
2268 return true;
2269 }
2270
2271 //--------------------------inline_unsigned_divmod_methods-----------------------------
2272 // inline int Integer.divideUnsigned(int, int)
2273 // inline int Integer.remainderUnsigned(int, int)
2274 // inline long Long.divideUnsigned(long, long)
2275 // inline long Long.remainderUnsigned(long, long)
2276 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2277 Node* n = nullptr;
2278 switch (id) {
2279 case vmIntrinsics::_divideUnsigned_i: {
2280 zero_check_int(argument(1));
2281 // Compile-time detect of null-exception
2282 if (stopped()) {
2283 return true; // keep the graph constructed so far
2284 }
2285 n = new UDivINode(control(), argument(0), argument(1));
2286 break;
2287 }
2288 case vmIntrinsics::_divideUnsigned_l: {
2289 zero_check_long(argument(2));
2290 // Compile-time detect of null-exception
2291 if (stopped()) {
2292 return true; // keep the graph constructed so far
2293 }
2294 n = new UDivLNode(control(), argument(0), argument(2));
2295 break;
2296 }
2297 case vmIntrinsics::_remainderUnsigned_i: {
2298 zero_check_int(argument(1));
2299 // Compile-time detect of null-exception
2300 if (stopped()) {
2301 return true; // keep the graph constructed so far
2302 }
2303 n = new UModINode(control(), argument(0), argument(1));
2304 break;
2305 }
2306 case vmIntrinsics::_remainderUnsigned_l: {
2307 zero_check_long(argument(2));
2308 // Compile-time detect of null-exception
2309 if (stopped()) {
2310 return true; // keep the graph constructed so far
2311 }
2312 n = new UModLNode(control(), argument(0), argument(2));
2313 break;
2314 }
2315 default: fatal_unexpected_iid(id); break;
2316 }
2317 set_result(_gvn.transform(n));
2318 return true;
2319 }
2320
2321 //----------------------------inline_unsafe_access----------------------------
2322
2323 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2324 // Attempt to infer a sharper value type from the offset and base type.
2325 ciKlass* sharpened_klass = nullptr;
2326
2327 // See if it is an instance field, with an object type.
2328 if (alias_type->field() != nullptr) {
2329 if (alias_type->field()->type()->is_klass()) {
2330 sharpened_klass = alias_type->field()->type()->as_klass();
2331 }
2332 }
2333
2334 const TypeOopPtr* result = nullptr;
2335 // See if it is a narrow oop array.
2336 if (adr_type->isa_aryptr()) {
2337 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2338 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2339 if (elem_type != nullptr && elem_type->is_loaded()) {
2340 // Sharpen the value type.
2341 result = elem_type;
2342 }
2343 }
2344 }
2345
2346 // The sharpened class might be unloaded if there is no class loader
2347 // contraint in place.
2348 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2349 // Sharpen the value type.
2350 result = TypeOopPtr::make_from_klass(sharpened_klass);
2351 }
2352 if (result != nullptr) {
2353 #ifndef PRODUCT
2354 if (C->print_intrinsics() || C->print_inlining()) {
2355 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2356 tty->print(" sharpened value: "); result->dump(); tty->cr();
2357 }
2358 #endif
2359 }
2360 return result;
2361 }
2362
2363 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2364 switch (kind) {
2365 case Relaxed:
2366 return MO_UNORDERED;
2367 case Opaque:
2368 return MO_RELAXED;
2369 case Acquire:
2370 return MO_ACQUIRE;
2371 case Release:
2372 return MO_RELEASE;
2373 case Volatile:
2374 return MO_SEQ_CST;
2375 default:
2376 ShouldNotReachHere();
2377 return 0;
2378 }
2379 }
2380
2381 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2382 if (callee()->is_static()) return false; // caller must have the capability!
2383 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2384 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2385 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2386 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2387
2388 if (is_reference_type(type)) {
2389 decorators |= ON_UNKNOWN_OOP_REF;
2390 }
2391
2392 if (unaligned) {
2393 decorators |= C2_UNALIGNED;
2394 }
2395
2396 #ifndef PRODUCT
2397 {
2398 ResourceMark rm;
2399 // Check the signatures.
2400 ciSignature* sig = callee()->signature();
2401 #ifdef ASSERT
2402 if (!is_store) {
2403 // Object getReference(Object base, int/long offset), etc.
2404 BasicType rtype = sig->return_type()->basic_type();
2405 assert(rtype == type, "getter must return the expected value");
2406 assert(sig->count() == 2, "oop getter has 2 arguments");
2407 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2408 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2409 } else {
2410 // void putReference(Object base, int/long offset, Object x), etc.
2411 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2412 assert(sig->count() == 3, "oop putter has 3 arguments");
2413 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2414 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2415 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2416 assert(vtype == type, "putter must accept the expected value");
2417 }
2418 #endif // ASSERT
2419 }
2420 #endif //PRODUCT
2421
2422 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2423
2424 Node* receiver = argument(0); // type: oop
2425
2426 // Build address expression.
2427 Node* heap_base_oop = top();
2428
2429 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2430 Node* base = argument(1); // type: oop
2431 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2432 Node* offset = argument(2); // type: long
2433 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2434 // to be plain byte offsets, which are also the same as those accepted
2435 // by oopDesc::field_addr.
2436 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2437 "fieldOffset must be byte-scaled");
2438 // 32-bit machines ignore the high half!
2439 offset = ConvL2X(offset);
2440
2441 // Save state and restore on bailout
2442 SavedState old_state(this);
2443
2444 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2445 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2446
2447 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2448 if (type != T_OBJECT) {
2449 decorators |= IN_NATIVE; // off-heap primitive access
2450 } else {
2451 return false; // off-heap oop accesses are not supported
2452 }
2453 } else {
2454 heap_base_oop = base; // on-heap or mixed access
2455 }
2456
2457 // Can base be null? Otherwise, always on-heap access.
2458 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2459
2460 if (!can_access_non_heap) {
2461 decorators |= IN_HEAP;
2462 }
2463
2464 Node* val = is_store ? argument(4) : nullptr;
2465
2466 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2467 if (adr_type == TypePtr::NULL_PTR) {
2468 return false; // off-heap access with zero address
2469 }
2470
2471 // Try to categorize the address.
2472 Compile::AliasType* alias_type = C->alias_type(adr_type);
2473 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2474
2475 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2476 alias_type->adr_type() == TypeAryPtr::RANGE) {
2477 return false; // not supported
2478 }
2479
2480 bool mismatched = false;
2481 BasicType bt = alias_type->basic_type();
2482 if (bt != T_ILLEGAL) {
2483 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2484 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2485 // Alias type doesn't differentiate between byte[] and boolean[]).
2486 // Use address type to get the element type.
2487 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2488 }
2489 if (is_reference_type(bt, true)) {
2490 // accessing an array field with getReference is not a mismatch
2491 bt = T_OBJECT;
2492 }
2493 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2494 // Don't intrinsify mismatched object accesses
2495 return false;
2496 }
2497 mismatched = (bt != type);
2498 } else if (alias_type->adr_type()->isa_oopptr()) {
2499 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2500 }
2501
2502 old_state.discard();
2503 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2504
2505 if (mismatched) {
2506 decorators |= C2_MISMATCHED;
2507 }
2508
2509 // First guess at the value type.
2510 const Type *value_type = Type::get_const_basic_type(type);
2511
2512 // Figure out the memory ordering.
2513 decorators |= mo_decorator_for_access_kind(kind);
2514
2515 if (!is_store && type == T_OBJECT) {
2516 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2517 if (tjp != nullptr) {
2518 value_type = tjp;
2519 }
2520 }
2521
2522 receiver = null_check(receiver);
2523 if (stopped()) {
2524 return true;
2525 }
2526 // Heap pointers get a null-check from the interpreter,
2527 // as a courtesy. However, this is not guaranteed by Unsafe,
2528 // and it is not possible to fully distinguish unintended nulls
2529 // from intended ones in this API.
2530
2531 if (!is_store) {
2532 Node* p = nullptr;
2533 // Try to constant fold a load from a constant field
2534 ciField* field = alias_type->field();
2535 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2536 // final or stable field
2537 p = make_constant_from_field(field, heap_base_oop);
2538 }
2539
2540 if (p == nullptr) { // Could not constant fold the load
2541 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2542 // Normalize the value returned by getBoolean in the following cases
2543 if (type == T_BOOLEAN &&
2544 (mismatched ||
2545 heap_base_oop == top() || // - heap_base_oop is null or
2546 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2547 // and the unsafe access is made to large offset
2548 // (i.e., larger than the maximum offset necessary for any
2549 // field access)
2550 ) {
2551 IdealKit ideal = IdealKit(this);
2552 #define __ ideal.
2553 IdealVariable normalized_result(ideal);
2554 __ declarations_done();
2555 __ set(normalized_result, p);
2556 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2557 __ set(normalized_result, ideal.ConI(1));
2558 ideal.end_if();
2559 final_sync(ideal);
2560 p = __ value(normalized_result);
2561 #undef __
2562 }
2563 }
2564 if (type == T_ADDRESS) {
2565 p = gvn().transform(new CastP2XNode(nullptr, p));
2566 p = ConvX2UL(p);
2567 }
2568 // The load node has the control of the preceding MemBarCPUOrder. All
2569 // following nodes will have the control of the MemBarCPUOrder inserted at
2570 // the end of this method. So, pushing the load onto the stack at a later
2571 // point is fine.
2572 set_result(p);
2573 } else {
2574 if (bt == T_ADDRESS) {
2575 // Repackage the long as a pointer.
2576 val = ConvL2X(val);
2577 val = gvn().transform(new CastX2PNode(val));
2578 }
2579 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2580 }
2581
2582 return true;
2583 }
2584
2585 //----------------------------inline_unsafe_load_store----------------------------
2586 // This method serves a couple of different customers (depending on LoadStoreKind):
2587 //
2588 // LS_cmp_swap:
2589 //
2590 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2591 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2592 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2593 //
2594 // LS_cmp_swap_weak:
2595 //
2596 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2597 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2598 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2599 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2600 //
2601 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2602 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2603 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2604 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2605 //
2606 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2607 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2608 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2609 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2610 //
2611 // LS_cmp_exchange:
2612 //
2613 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2615 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2616 //
2617 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2619 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2620 //
2621 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2622 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2623 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2624 //
2625 // LS_get_add:
2626 //
2627 // int getAndAddInt( Object o, long offset, int delta)
2628 // long getAndAddLong(Object o, long offset, long delta)
2629 //
2630 // LS_get_set:
2631 //
2632 // int getAndSet(Object o, long offset, int newValue)
2633 // long getAndSet(Object o, long offset, long newValue)
2634 // Object getAndSet(Object o, long offset, Object newValue)
2635 //
2636 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2637 // This basic scheme here is the same as inline_unsafe_access, but
2638 // differs in enough details that combining them would make the code
2639 // overly confusing. (This is a true fact! I originally combined
2640 // them, but even I was confused by it!) As much code/comments as
2641 // possible are retained from inline_unsafe_access though to make
2642 // the correspondences clearer. - dl
2643
2644 if (callee()->is_static()) return false; // caller must have the capability!
2645
2646 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2647 decorators |= mo_decorator_for_access_kind(access_kind);
2648
2649 #ifndef PRODUCT
2650 BasicType rtype;
2651 {
2652 ResourceMark rm;
2653 // Check the signatures.
2654 ciSignature* sig = callee()->signature();
2655 rtype = sig->return_type()->basic_type();
2656 switch(kind) {
2657 case LS_get_add:
2658 case LS_get_set: {
2659 // Check the signatures.
2660 #ifdef ASSERT
2661 assert(rtype == type, "get and set must return the expected type");
2662 assert(sig->count() == 3, "get and set has 3 arguments");
2663 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2664 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2665 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2666 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2667 #endif // ASSERT
2668 break;
2669 }
2670 case LS_cmp_swap:
2671 case LS_cmp_swap_weak: {
2672 // Check the signatures.
2673 #ifdef ASSERT
2674 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2675 assert(sig->count() == 4, "CAS has 4 arguments");
2676 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2677 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2678 #endif // ASSERT
2679 break;
2680 }
2681 case LS_cmp_exchange: {
2682 // Check the signatures.
2683 #ifdef ASSERT
2684 assert(rtype == type, "CAS must return the expected type");
2685 assert(sig->count() == 4, "CAS has 4 arguments");
2686 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2687 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2688 #endif // ASSERT
2689 break;
2690 }
2691 default:
2692 ShouldNotReachHere();
2693 }
2694 }
2695 #endif //PRODUCT
2696
2697 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2698
2699 // Get arguments:
2700 Node* receiver = nullptr;
2701 Node* base = nullptr;
2702 Node* offset = nullptr;
2703 Node* oldval = nullptr;
2704 Node* newval = nullptr;
2705 switch(kind) {
2706 case LS_cmp_swap:
2707 case LS_cmp_swap_weak:
2708 case LS_cmp_exchange: {
2709 const bool two_slot_type = type2size[type] == 2;
2710 receiver = argument(0); // type: oop
2711 base = argument(1); // type: oop
2712 offset = argument(2); // type: long
2713 oldval = argument(4); // type: oop, int, or long
2714 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2715 break;
2716 }
2717 case LS_get_add:
2718 case LS_get_set: {
2719 receiver = argument(0); // type: oop
2720 base = argument(1); // type: oop
2721 offset = argument(2); // type: long
2722 oldval = nullptr;
2723 newval = argument(4); // type: oop, int, or long
2724 break;
2725 }
2726 default:
2727 ShouldNotReachHere();
2728 }
2729
2730 // Build field offset expression.
2731 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2732 // to be plain byte offsets, which are also the same as those accepted
2733 // by oopDesc::field_addr.
2734 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2735 // 32-bit machines ignore the high half of long offsets
2736 offset = ConvL2X(offset);
2737 // Save state and restore on bailout
2738 SavedState old_state(this);
2739 Node* adr = make_unsafe_address(base, offset,type, false);
2740 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2741
2742 Compile::AliasType* alias_type = C->alias_type(adr_type);
2743 BasicType bt = alias_type->basic_type();
2744 if (bt != T_ILLEGAL &&
2745 (is_reference_type(bt) != (type == T_OBJECT))) {
2746 // Don't intrinsify mismatched object accesses.
2747 return false;
2748 }
2749
2750 old_state.discard();
2751
2752 // For CAS, unlike inline_unsafe_access, there seems no point in
2753 // trying to refine types. Just use the coarse types here.
2754 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2755 const Type *value_type = Type::get_const_basic_type(type);
2756
2757 switch (kind) {
2758 case LS_get_set:
2759 case LS_cmp_exchange: {
2760 if (type == T_OBJECT) {
2761 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2762 if (tjp != nullptr) {
2763 value_type = tjp;
2764 }
2765 }
2766 break;
2767 }
2768 case LS_cmp_swap:
2769 case LS_cmp_swap_weak:
2770 case LS_get_add:
2771 break;
2772 default:
2773 ShouldNotReachHere();
2774 }
2775
2776 // Null check receiver.
2777 receiver = null_check(receiver);
2778 if (stopped()) {
2779 return true;
2780 }
2781
2782 int alias_idx = C->get_alias_index(adr_type);
2783
2784 if (is_reference_type(type)) {
2785 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2786
2787 // Transformation of a value which could be null pointer (CastPP #null)
2788 // could be delayed during Parse (for example, in adjust_map_after_if()).
2789 // Execute transformation here to avoid barrier generation in such case.
2790 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2791 newval = _gvn.makecon(TypePtr::NULL_PTR);
2792
2793 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2794 // Refine the value to a null constant, when it is known to be null
2795 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2796 }
2797 }
2798
2799 Node* result = nullptr;
2800 switch (kind) {
2801 case LS_cmp_exchange: {
2802 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2803 oldval, newval, value_type, type, decorators);
2804 break;
2805 }
2806 case LS_cmp_swap_weak:
2807 decorators |= C2_WEAK_CMPXCHG;
2808 case LS_cmp_swap: {
2809 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2810 oldval, newval, value_type, type, decorators);
2811 break;
2812 }
2813 case LS_get_set: {
2814 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2815 newval, value_type, type, decorators);
2816 break;
2817 }
2818 case LS_get_add: {
2819 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2820 newval, value_type, type, decorators);
2821 break;
2822 }
2823 default:
2824 ShouldNotReachHere();
2825 }
2826
2827 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2828 set_result(result);
2829 return true;
2830 }
2831
2832 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2833 // Regardless of form, don't allow previous ld/st to move down,
2834 // then issue acquire, release, or volatile mem_bar.
2835 insert_mem_bar(Op_MemBarCPUOrder);
2836 switch(id) {
2837 case vmIntrinsics::_loadFence:
2838 insert_mem_bar(Op_LoadFence);
2839 return true;
2840 case vmIntrinsics::_storeFence:
2841 insert_mem_bar(Op_StoreFence);
2842 return true;
2843 case vmIntrinsics::_storeStoreFence:
2844 insert_mem_bar(Op_StoreStoreFence);
2845 return true;
2846 case vmIntrinsics::_fullFence:
2847 insert_mem_bar(Op_MemBarFull);
2848 return true;
2849 default:
2850 fatal_unexpected_iid(id);
2851 return false;
2852 }
2853 }
2854
2855 bool LibraryCallKit::inline_onspinwait() {
2856 insert_mem_bar(Op_OnSpinWait);
2857 return true;
2858 }
2859
2860 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2861 if (!kls->is_Con()) {
2862 return true;
2863 }
2864 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2865 if (klsptr == nullptr) {
2866 return true;
2867 }
2868 ciInstanceKlass* ik = klsptr->instance_klass();
2869 // don't need a guard for a klass that is already initialized
2870 return !ik->is_initialized();
2871 }
2872
2873 //----------------------------inline_unsafe_writeback0-------------------------
2874 // public native void Unsafe.writeback0(long address)
2875 bool LibraryCallKit::inline_unsafe_writeback0() {
2876 if (!Matcher::has_match_rule(Op_CacheWB)) {
2877 return false;
2878 }
2879 #ifndef PRODUCT
2880 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2881 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2882 ciSignature* sig = callee()->signature();
2883 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2884 #endif
2885 null_check_receiver(); // null-check, then ignore
2886 Node *addr = argument(1);
2887 addr = new CastX2PNode(addr);
2888 addr = _gvn.transform(addr);
2889 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2890 flush = _gvn.transform(flush);
2891 set_memory(flush, TypeRawPtr::BOTTOM);
2892 return true;
2893 }
2894
2895 //----------------------------inline_unsafe_writeback0-------------------------
2896 // public native void Unsafe.writeback0(long address)
2897 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2898 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2899 return false;
2900 }
2901 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2902 return false;
2903 }
2904 #ifndef PRODUCT
2905 assert(Matcher::has_match_rule(Op_CacheWB),
2906 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2907 : "found match rule for CacheWBPostSync but not CacheWB"));
2908
2909 #endif
2910 null_check_receiver(); // null-check, then ignore
2911 Node *sync;
2912 if (is_pre) {
2913 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2914 } else {
2915 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2916 }
2917 sync = _gvn.transform(sync);
2918 set_memory(sync, TypeRawPtr::BOTTOM);
2919 return true;
2920 }
2921
2922 //----------------------------inline_unsafe_allocate---------------------------
2923 // public native Object Unsafe.allocateInstance(Class<?> cls);
2924 bool LibraryCallKit::inline_unsafe_allocate() {
2925
2926 #if INCLUDE_JVMTI
2927 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2928 return false;
2929 }
2930 #endif //INCLUDE_JVMTI
2931
2932 if (callee()->is_static()) return false; // caller must have the capability!
2933
2934 null_check_receiver(); // null-check, then ignore
2935 Node* cls = null_check(argument(1));
2936 if (stopped()) return true;
2937
2938 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2939 kls = null_check(kls);
2940 if (stopped()) return true; // argument was like int.class
2941
2942 #if INCLUDE_JVMTI
2943 // Don't try to access new allocated obj in the intrinsic.
2944 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2945 // Deoptimize and allocate in interpreter instead.
2946 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2947 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2948 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2949 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2950 {
2951 BuildCutout unless(this, tst, PROB_MAX);
2952 uncommon_trap(Deoptimization::Reason_intrinsic,
2953 Deoptimization::Action_make_not_entrant);
2954 }
2955 if (stopped()) {
2956 return true;
2957 }
2958 #endif //INCLUDE_JVMTI
2959
2960 Node* test = nullptr;
2961 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2962 // Note: The argument might still be an illegal value like
2963 // Serializable.class or Object[].class. The runtime will handle it.
2964 // But we must make an explicit check for initialization.
2965 Node* insp = basic_plus_adr(top(), kls, in_bytes(InstanceKlass::init_state_offset()));
2966 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2967 // can generate code to load it as unsigned byte.
2968 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2969 Node* bits = intcon(InstanceKlass::fully_initialized);
2970 test = _gvn.transform(new SubINode(inst, bits));
2971 // The 'test' is non-zero if we need to take a slow path.
2972 }
2973
2974 Node* obj = new_instance(kls, test);
2975 set_result(obj);
2976 return true;
2977 }
2978
2979 //------------------------inline_native_time_funcs--------------
2980 // inline code for System.currentTimeMillis() and System.nanoTime()
2981 // these have the same type and signature
2982 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2983 const TypeFunc* tf = OptoRuntime::void_long_Type();
2984 const TypePtr* no_memory_effects = nullptr;
2985 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2986 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2987 #ifdef ASSERT
2988 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2989 assert(value_top == top(), "second value must be top");
2990 #endif
2991 set_result(value);
2992 return true;
2993 }
2994
2995 //--------------------inline_native_vthread_start_transition--------------------
2996 // inline void startTransition(boolean is_mount);
2997 // inline void startFinalTransition();
2998 // Pseudocode of implementation:
2999 //
3000 // java_lang_Thread::set_is_in_vthread_transition(vthread, true);
3001 // carrier->set_is_in_vthread_transition(true);
3002 // OrderAccess::storeload();
3003 // int disable_requests = java_lang_Thread::vthread_transition_disable_count(vthread)
3004 // + global_vthread_transition_disable_count();
3005 // if (disable_requests > 0) {
3006 // slow path: runtime call
3007 // }
3008 bool LibraryCallKit::inline_native_vthread_start_transition(address funcAddr, const char* funcName, bool is_final_transition) {
3009 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3010 IdealKit ideal(this);
3011
3012 Node* thread = ideal.thread();
3013 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3014 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3015 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3016 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(1), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3017 insert_mem_bar(Op_MemBarStoreLoad);
3018 ideal.sync_kit(this);
3019
3020 Node* global_disable_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::global_vthread_transition_disable_count_address()));
3021 Node* global_disable = ideal.load(ideal.ctrl(), global_disable_addr, TypeInt::INT, T_INT, Compile::AliasIdxRaw, true /*require_atomic_access*/);
3022 Node* vt_disable_addr = basic_plus_adr(vt_oop, java_lang_Thread::vthread_transition_disable_count_offset());
3023 const TypePtr* vt_disable_addr_t = _gvn.type(vt_disable_addr)->is_ptr();
3024 Node* vt_disable = ideal.load(ideal.ctrl(), vt_disable_addr, TypeInt::INT, T_INT, C->get_alias_index(vt_disable_addr_t), true /*require_atomic_access*/);
3025 Node* disabled = _gvn.transform(new AddINode(global_disable, vt_disable));
3026
3027 ideal.if_then(disabled, BoolTest::ne, ideal.ConI(0)); {
3028 sync_kit(ideal);
3029 Node* is_mount = is_final_transition ? ideal.ConI(0) : argument(1);
3030 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3031 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3032 ideal.sync_kit(this);
3033 }
3034 ideal.end_if();
3035
3036 final_sync(ideal);
3037 return true;
3038 }
3039
3040 bool LibraryCallKit::inline_native_vthread_end_transition(address funcAddr, const char* funcName, bool is_first_transition) {
3041 Node* vt_oop = must_be_not_null(argument(0), true); // VirtualThread this argument
3042 IdealKit ideal(this);
3043
3044 Node* _notify_jvmti_addr = makecon(TypeRawPtr::make((address)MountUnmountDisabler::notify_jvmti_events_address()));
3045 Node* _notify_jvmti = ideal.load(ideal.ctrl(), _notify_jvmti_addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3046
3047 ideal.if_then(_notify_jvmti, BoolTest::eq, ideal.ConI(1)); {
3048 sync_kit(ideal);
3049 Node* is_mount = is_first_transition ? ideal.ConI(1) : argument(1);
3050 const TypeFunc* tf = OptoRuntime::vthread_transition_Type();
3051 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, is_mount);
3052 ideal.sync_kit(this);
3053 } ideal.else_(); {
3054 Node* thread = ideal.thread();
3055 Node* jt_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_in_vthread_transition_offset()));
3056 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_vthread_transition_offset());
3057
3058 sync_kit(ideal);
3059 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3060 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), ideal.ConI(0), TypeInt::BOOL, T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3061 ideal.sync_kit(this);
3062 } ideal.end_if();
3063
3064 final_sync(ideal);
3065 return true;
3066 }
3067
3068 #if INCLUDE_JVMTI
3069
3070 // Always update the is_disable_suspend bit.
3071 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3072 if (!DoJVMTIVirtualThreadTransitions) {
3073 return true;
3074 }
3075 IdealKit ideal(this);
3076
3077 {
3078 // unconditionally update the is_disable_suspend bit in current JavaThread
3079 Node* thread = ideal.thread();
3080 Node* arg = argument(0); // argument for notification
3081 Node* addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3082 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3083
3084 sync_kit(ideal);
3085 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3086 ideal.sync_kit(this);
3087 }
3088 final_sync(ideal);
3089
3090 return true;
3091 }
3092
3093 #endif // INCLUDE_JVMTI
3094
3095 #ifdef JFR_HAVE_INTRINSICS
3096
3097 /**
3098 * if oop->klass != null
3099 * // normal class
3100 * epoch = _epoch_state ? 2 : 1
3101 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3102 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3103 * }
3104 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3105 * else
3106 * // primitive class
3107 * if oop->array_klass != null
3108 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3109 * else
3110 * id = LAST_TYPE_ID + 1 // void class path
3111 * if (!signaled)
3112 * signaled = true
3113 */
3114 bool LibraryCallKit::inline_native_classID() {
3115 Node* cls = argument(0);
3116
3117 IdealKit ideal(this);
3118 #define __ ideal.
3119 IdealVariable result(ideal); __ declarations_done();
3120 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3121 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3122 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3123
3124
3125 __ if_then(kls, BoolTest::ne, null()); {
3126 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3127 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3128
3129 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3130 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3131 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3132 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3133 mask = _gvn.transform(new OrLNode(mask, epoch));
3134 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3135
3136 float unlikely = PROB_UNLIKELY(0.999);
3137 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3138 sync_kit(ideal);
3139 make_runtime_call(RC_LEAF,
3140 OptoRuntime::class_id_load_barrier_Type(),
3141 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3142 "class id load barrier",
3143 TypePtr::BOTTOM,
3144 kls);
3145 ideal.sync_kit(this);
3146 } __ end_if();
3147
3148 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3149 } __ else_(); {
3150 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3151 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3152 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3153 __ if_then(array_kls, BoolTest::ne, null()); {
3154 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3155 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3156 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3157 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3158 } __ else_(); {
3159 // void class case
3160 ideal.set(result, longcon(LAST_TYPE_ID + 1));
3161 } __ end_if();
3162
3163 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3164 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3165 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3166 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3167 } __ end_if();
3168 } __ end_if();
3169
3170 final_sync(ideal);
3171 set_result(ideal.value(result));
3172 #undef __
3173 return true;
3174 }
3175
3176 //------------------------inline_native_jvm_commit------------------
3177 bool LibraryCallKit::inline_native_jvm_commit() {
3178 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3179
3180 // Save input memory and i_o state.
3181 Node* input_memory_state = reset_memory();
3182 set_all_memory(input_memory_state);
3183 Node* input_io_state = i_o();
3184
3185 // TLS.
3186 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3187 // Jfr java buffer.
3188 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR))));
3189 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3190 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET))));
3191
3192 // Load the current value of the notified field in the JfrThreadLocal.
3193 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3194 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3195
3196 // Test for notification.
3197 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3198 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3199 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3200
3201 // True branch, is notified.
3202 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3203 set_control(is_notified);
3204
3205 // Reset notified state.
3206 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3207 Node* notified_reset_memory = reset_memory();
3208
3209 // 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.
3210 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3211 // Convert the machine-word to a long.
3212 Node* current_pos = ConvX2L(current_pos_X);
3213
3214 // False branch, not notified.
3215 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3216 set_control(not_notified);
3217 set_all_memory(input_memory_state);
3218
3219 // Arg is the next position as a long.
3220 Node* arg = argument(0);
3221 // Convert long to machine-word.
3222 Node* next_pos_X = ConvL2X(arg);
3223
3224 // Store the next_position to the underlying jfr java buffer.
3225 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3226
3227 Node* commit_memory = reset_memory();
3228 set_all_memory(commit_memory);
3229
3230 // 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.
3231 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET))));
3232 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3233 Node* lease_constant = _gvn.intcon(4);
3234
3235 // And flags with lease constant.
3236 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3237
3238 // Branch on lease to conditionalize returning the leased java buffer.
3239 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3240 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3241 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3242
3243 // False branch, not a lease.
3244 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3245
3246 // True branch, is lease.
3247 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3248 set_control(is_lease);
3249
3250 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3251 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3252 OptoRuntime::void_void_Type(),
3253 SharedRuntime::jfr_return_lease(),
3254 "return_lease", TypePtr::BOTTOM);
3255 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3256
3257 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3258 record_for_igvn(lease_compare_rgn);
3259 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3260 record_for_igvn(lease_compare_mem);
3261 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3262 record_for_igvn(lease_compare_io);
3263 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3264 record_for_igvn(lease_result_value);
3265
3266 // Update control and phi nodes.
3267 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3268 lease_compare_rgn->init_req(_false_path, not_lease);
3269
3270 lease_compare_mem->init_req(_true_path, reset_memory());
3271 lease_compare_mem->init_req(_false_path, commit_memory);
3272
3273 lease_compare_io->init_req(_true_path, i_o());
3274 lease_compare_io->init_req(_false_path, input_io_state);
3275
3276 lease_result_value->init_req(_true_path, _gvn.longcon(0)); // if the lease was returned, return 0L.
3277 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3278
3279 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3280 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3281 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3282 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3283
3284 // Update control and phi nodes.
3285 result_rgn->init_req(_true_path, is_notified);
3286 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3287
3288 result_mem->init_req(_true_path, notified_reset_memory);
3289 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3290
3291 result_io->init_req(_true_path, input_io_state);
3292 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3293
3294 result_value->init_req(_true_path, current_pos);
3295 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3296
3297 // Set output state.
3298 set_control(_gvn.transform(result_rgn));
3299 set_all_memory(_gvn.transform(result_mem));
3300 set_i_o(_gvn.transform(result_io));
3301 set_result(result_rgn, result_value);
3302 return true;
3303 }
3304
3305 /*
3306 * The intrinsic is a model of this pseudo-code:
3307 *
3308 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3309 * jobject h_event_writer = tl->java_event_writer();
3310 * if (h_event_writer == nullptr) {
3311 * return nullptr;
3312 * }
3313 * oop threadObj = Thread::threadObj();
3314 * oop vthread = java_lang_Thread::vthread(threadObj);
3315 * traceid tid;
3316 * bool pinVirtualThread;
3317 * bool excluded;
3318 * if (vthread != threadObj) { // i.e. current thread is virtual
3319 * tid = java_lang_Thread::tid(vthread);
3320 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3321 * pinVirtualThread = VMContinuations;
3322 * excluded = vthread_epoch_raw & excluded_mask;
3323 * if (!excluded) {
3324 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3325 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3326 * if (vthread_epoch != current_epoch) {
3327 * write_checkpoint();
3328 * }
3329 * }
3330 * } else {
3331 * tid = java_lang_Thread::tid(threadObj);
3332 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3333 * pinVirtualThread = false;
3334 * excluded = thread_epoch_raw & excluded_mask;
3335 * }
3336 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3337 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3338 * if (tid_in_event_writer != tid) {
3339 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3340 * setField(event_writer, "excluded", excluded);
3341 * setField(event_writer, "threadID", tid);
3342 * }
3343 * return event_writer
3344 */
3345 bool LibraryCallKit::inline_native_getEventWriter() {
3346 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3347
3348 // Save input memory and i_o state.
3349 Node* input_memory_state = reset_memory();
3350 set_all_memory(input_memory_state);
3351 Node* input_io_state = i_o();
3352
3353 // The most significant bit of the u2 is used to denote thread exclusion
3354 Node* excluded_shift = _gvn.intcon(15);
3355 Node* excluded_mask = _gvn.intcon(1 << 15);
3356 // The epoch generation is the range [1-32767]
3357 Node* epoch_mask = _gvn.intcon(32767);
3358
3359 // TLS
3360 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3361
3362 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3363 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3364
3365 // Load the eventwriter jobject handle.
3366 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3367
3368 // Null check the jobject handle.
3369 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3370 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3371 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3372
3373 // False path, jobj is null.
3374 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3375
3376 // True path, jobj is not null.
3377 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3378
3379 set_control(jobj_is_not_null);
3380
3381 // Load the threadObj for the CarrierThread.
3382 Node* threadObj = generate_current_thread(tls_ptr);
3383
3384 // Load the vthread.
3385 Node* vthread = generate_virtual_thread(tls_ptr);
3386
3387 // If vthread != threadObj, this is a virtual thread.
3388 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3389 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3390 IfNode* iff_vthread_not_equal_threadObj =
3391 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3392
3393 // False branch, fallback to threadObj.
3394 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3395 set_control(vthread_equal_threadObj);
3396
3397 // Load the tid field from the vthread object.
3398 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3399
3400 // Load the raw epoch value from the threadObj.
3401 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3402 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3403 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3404 TypeInt::CHAR, T_CHAR,
3405 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3406
3407 // Mask off the excluded information from the epoch.
3408 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3409
3410 // True branch, this is a virtual thread.
3411 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3412 set_control(vthread_not_equal_threadObj);
3413
3414 // Load the tid field from the vthread object.
3415 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3416
3417 // Continuation support determines if a virtual thread should be pinned.
3418 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3419 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3420
3421 // Load the raw epoch value from the vthread.
3422 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3423 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3424 TypeInt::CHAR, T_CHAR,
3425 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3426
3427 // Mask off the excluded information from the epoch.
3428 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, excluded_mask));
3429
3430 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3431 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, excluded_mask));
3432 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3433 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3434
3435 // False branch, vthread is excluded, no need to write epoch info.
3436 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3437
3438 // True branch, vthread is included, update epoch info.
3439 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3440 set_control(included);
3441
3442 // Get epoch value.
3443 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, epoch_mask));
3444
3445 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3446 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3447 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3448
3449 // Compare the epoch in the vthread to the current epoch generation.
3450 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3451 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3452 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3453
3454 // False path, epoch is equal, checkpoint information is valid.
3455 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3456
3457 // True path, epoch is not equal, write a checkpoint for the vthread.
3458 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3459
3460 set_control(epoch_is_not_equal);
3461
3462 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3463 // The call also updates the native thread local thread id and the vthread with the current epoch.
3464 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3465 OptoRuntime::jfr_write_checkpoint_Type(),
3466 SharedRuntime::jfr_write_checkpoint(),
3467 "write_checkpoint", TypePtr::BOTTOM);
3468 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3469
3470 // vthread epoch != current epoch
3471 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3472 record_for_igvn(epoch_compare_rgn);
3473 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3474 record_for_igvn(epoch_compare_mem);
3475 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3476 record_for_igvn(epoch_compare_io);
3477
3478 // Update control and phi nodes.
3479 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3480 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3481 epoch_compare_mem->init_req(_true_path, reset_memory());
3482 epoch_compare_mem->init_req(_false_path, input_memory_state);
3483 epoch_compare_io->init_req(_true_path, i_o());
3484 epoch_compare_io->init_req(_false_path, input_io_state);
3485
3486 // excluded != true
3487 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3488 record_for_igvn(exclude_compare_rgn);
3489 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3490 record_for_igvn(exclude_compare_mem);
3491 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3492 record_for_igvn(exclude_compare_io);
3493
3494 // Update control and phi nodes.
3495 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3496 exclude_compare_rgn->init_req(_false_path, excluded);
3497 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3498 exclude_compare_mem->init_req(_false_path, input_memory_state);
3499 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3500 exclude_compare_io->init_req(_false_path, input_io_state);
3501
3502 // vthread != threadObj
3503 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3504 record_for_igvn(vthread_compare_rgn);
3505 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3506 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3507 record_for_igvn(vthread_compare_io);
3508 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3509 record_for_igvn(tid);
3510 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3511 record_for_igvn(exclusion);
3512 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3513 record_for_igvn(pinVirtualThread);
3514
3515 // Update control and phi nodes.
3516 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3517 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3518 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3519 vthread_compare_mem->init_req(_false_path, input_memory_state);
3520 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3521 vthread_compare_io->init_req(_false_path, input_io_state);
3522 tid->init_req(_true_path, vthread_tid);
3523 tid->init_req(_false_path, thread_obj_tid);
3524 exclusion->init_req(_true_path, vthread_is_excluded);
3525 exclusion->init_req(_false_path, threadObj_is_excluded);
3526 pinVirtualThread->init_req(_true_path, continuation_support);
3527 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3528
3529 // Update branch state.
3530 set_control(_gvn.transform(vthread_compare_rgn));
3531 set_all_memory(_gvn.transform(vthread_compare_mem));
3532 set_i_o(_gvn.transform(vthread_compare_io));
3533
3534 // Load the event writer oop by dereferencing the jobject handle.
3535 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3536 assert(klass_EventWriter->is_loaded(), "invariant");
3537 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3538 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3539 const TypeOopPtr* const xtype = aklass->as_instance_type();
3540 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3541 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3542
3543 // Load the current thread id from the event writer object.
3544 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3545 // Get the field offset to, conditionally, store an updated tid value later.
3546 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3547 // Get the field offset to, conditionally, store an updated exclusion value later.
3548 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3549 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3550 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3551
3552 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3553 record_for_igvn(event_writer_tid_compare_rgn);
3554 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3555 record_for_igvn(event_writer_tid_compare_mem);
3556 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3557 record_for_igvn(event_writer_tid_compare_io);
3558
3559 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3560 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3561 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3562 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3563
3564 // False path, tids are the same.
3565 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3566
3567 // True path, tid is not equal, need to update the tid in the event writer.
3568 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3569 record_for_igvn(tid_is_not_equal);
3570
3571 // Store the pin state to the event writer.
3572 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3573
3574 // Store the exclusion state to the event writer.
3575 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3576 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3577
3578 // Store the tid to the event writer.
3579 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3580
3581 // Update control and phi nodes.
3582 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3583 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3584 event_writer_tid_compare_mem->init_req(_true_path, reset_memory());
3585 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3586 event_writer_tid_compare_io->init_req(_true_path, i_o());
3587 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3588
3589 // Result of top level CFG, Memory, IO and Value.
3590 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3591 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3592 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3593 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3594
3595 // Result control.
3596 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3597 result_rgn->init_req(_false_path, jobj_is_null);
3598
3599 // Result memory.
3600 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3601 result_mem->init_req(_false_path, input_memory_state);
3602
3603 // Result IO.
3604 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3605 result_io->init_req(_false_path, input_io_state);
3606
3607 // Result value.
3608 result_value->init_req(_true_path, event_writer); // return event writer oop
3609 result_value->init_req(_false_path, null()); // return null
3610
3611 // Set output state.
3612 set_control(_gvn.transform(result_rgn));
3613 set_all_memory(_gvn.transform(result_mem));
3614 set_i_o(_gvn.transform(result_io));
3615 set_result(result_rgn, result_value);
3616 return true;
3617 }
3618
3619 /*
3620 * The intrinsic is a model of this pseudo-code:
3621 *
3622 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3623 * if (carrierThread != thread) { // is virtual thread
3624 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3625 * bool excluded = vthread_epoch_raw & excluded_mask;
3626 * AtomicAccess::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3627 * AtomicAccess::store(&tl->_contextual_thread_excluded, is_excluded);
3628 * if (!excluded) {
3629 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3630 * AtomicAccess::store(&tl->_vthread_epoch, vthread_epoch);
3631 * }
3632 * AtomicAccess::release_store(&tl->_vthread, true);
3633 * return;
3634 * }
3635 * AtomicAccess::release_store(&tl->_vthread, false);
3636 */
3637 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3638 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3639
3640 Node* input_memory_state = reset_memory();
3641 set_all_memory(input_memory_state);
3642
3643 // The most significant bit of the u2 is used to denote thread exclusion
3644 Node* excluded_mask = _gvn.intcon(1 << 15);
3645 // The epoch generation is the range [1-32767]
3646 Node* epoch_mask = _gvn.intcon(32767);
3647
3648 Node* const carrierThread = generate_current_thread(jt);
3649 // If thread != carrierThread, this is a virtual thread.
3650 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3651 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3652 IfNode* iff_thread_not_equal_carrierThread =
3653 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3654
3655 Node* vthread_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3656
3657 // False branch, is carrierThread.
3658 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3659 // Store release
3660 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3661
3662 set_all_memory(input_memory_state);
3663
3664 // True branch, is virtual thread.
3665 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3666 set_control(thread_not_equal_carrierThread);
3667
3668 // Load the raw epoch value from the vthread.
3669 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3670 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3671 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3672
3673 // Mask off the excluded information from the epoch.
3674 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, excluded_mask));
3675
3676 // Load the tid field from the thread.
3677 Node* tid = load_field_from_object(thread, "tid", "J");
3678
3679 // Store the vthread tid to the jfr thread local.
3680 Node* thread_id_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3681 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3682
3683 // Branch is_excluded to conditionalize updating the epoch .
3684 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, excluded_mask));
3685 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3686 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3687
3688 // True branch, vthread is excluded, no need to write epoch info.
3689 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3690 set_control(excluded);
3691 Node* vthread_is_excluded = _gvn.intcon(1);
3692
3693 // False branch, vthread is included, update epoch info.
3694 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3695 set_control(included);
3696 Node* vthread_is_included = _gvn.intcon(0);
3697
3698 // Get epoch value.
3699 Node* epoch = _gvn.transform(new AndINode(epoch_raw, epoch_mask));
3700
3701 // Store the vthread epoch to the jfr thread local.
3702 Node* vthread_epoch_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3703 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3704
3705 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3706 record_for_igvn(excluded_rgn);
3707 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3708 record_for_igvn(excluded_mem);
3709 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3710 record_for_igvn(exclusion);
3711
3712 // Merge the excluded control and memory.
3713 excluded_rgn->init_req(_true_path, excluded);
3714 excluded_rgn->init_req(_false_path, included);
3715 excluded_mem->init_req(_true_path, tid_memory);
3716 excluded_mem->init_req(_false_path, included_memory);
3717 exclusion->init_req(_true_path, vthread_is_excluded);
3718 exclusion->init_req(_false_path, vthread_is_included);
3719
3720 // Set intermediate state.
3721 set_control(_gvn.transform(excluded_rgn));
3722 set_all_memory(excluded_mem);
3723
3724 // Store the vthread exclusion state to the jfr thread local.
3725 Node* thread_local_excluded_offset = basic_plus_adr(top(), jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3726 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3727
3728 // Store release
3729 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3730
3731 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3732 record_for_igvn(thread_compare_rgn);
3733 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3734 record_for_igvn(thread_compare_mem);
3735 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3736 record_for_igvn(vthread);
3737
3738 // Merge the thread_compare control and memory.
3739 thread_compare_rgn->init_req(_true_path, control());
3740 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3741 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3742 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3743
3744 // Set output state.
3745 set_control(_gvn.transform(thread_compare_rgn));
3746 set_all_memory(_gvn.transform(thread_compare_mem));
3747 }
3748
3749 #endif // JFR_HAVE_INTRINSICS
3750
3751 //------------------------inline_native_currentCarrierThread------------------
3752 bool LibraryCallKit::inline_native_currentCarrierThread() {
3753 Node* junk = nullptr;
3754 set_result(generate_current_thread(junk));
3755 return true;
3756 }
3757
3758 //------------------------inline_native_currentThread------------------
3759 bool LibraryCallKit::inline_native_currentThread() {
3760 Node* junk = nullptr;
3761 set_result(generate_virtual_thread(junk));
3762 return true;
3763 }
3764
3765 //------------------------inline_native_setVthread------------------
3766 bool LibraryCallKit::inline_native_setCurrentThread() {
3767 assert(C->method()->changes_current_thread(),
3768 "method changes current Thread but is not annotated ChangesCurrentThread");
3769 Node* arr = argument(1);
3770 Node* thread = _gvn.transform(new ThreadLocalNode());
3771 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3772 Node* thread_obj_handle
3773 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3774 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3775 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3776
3777 // Change the _monitor_owner_id of the JavaThread
3778 Node* tid = load_field_from_object(arr, "tid", "J");
3779 Node* monitor_owner_id_offset = basic_plus_adr(top(), thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3780 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3781
3782 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3783 return true;
3784 }
3785
3786 const Type* LibraryCallKit::scopedValueCache_type() {
3787 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3788 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3789 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3790
3791 // Because we create the scopedValue cache lazily we have to make the
3792 // type of the result BotPTR.
3793 bool xk = etype->klass_is_exact();
3794 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3795 return objects_type;
3796 }
3797
3798 Node* LibraryCallKit::scopedValueCache_helper() {
3799 Node* thread = _gvn.transform(new ThreadLocalNode());
3800 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3801 // We cannot use immutable_memory() because we might flip onto a
3802 // different carrier thread, at which point we'll need to use that
3803 // carrier thread's cache.
3804 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3805 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3806 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3807 }
3808
3809 //------------------------inline_native_scopedValueCache------------------
3810 bool LibraryCallKit::inline_native_scopedValueCache() {
3811 Node* cache_obj_handle = scopedValueCache_helper();
3812 const Type* objects_type = scopedValueCache_type();
3813 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3814
3815 return true;
3816 }
3817
3818 //------------------------inline_native_setScopedValueCache------------------
3819 bool LibraryCallKit::inline_native_setScopedValueCache() {
3820 Node* arr = argument(0);
3821 Node* cache_obj_handle = scopedValueCache_helper();
3822 const Type* objects_type = scopedValueCache_type();
3823
3824 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3825 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3826
3827 return true;
3828 }
3829
3830 //------------------------inline_native_Continuation_pin and unpin-----------
3831
3832 // Shared implementation routine for both pin and unpin.
3833 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3834 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3835
3836 // Save input memory.
3837 Node* input_memory_state = reset_memory();
3838 set_all_memory(input_memory_state);
3839
3840 // TLS
3841 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3842 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3843 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3844
3845 // Null check the last continuation object.
3846 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3847 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3848 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3849
3850 // False path, last continuation is null.
3851 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3852
3853 // True path, last continuation is not null.
3854 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3855
3856 set_control(continuation_is_not_null);
3857
3858 // Load the pin count from the last continuation.
3859 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3860 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3861
3862 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3863 Node* pin_count_rhs;
3864 if (unpin) {
3865 pin_count_rhs = _gvn.intcon(0);
3866 } else {
3867 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3868 }
3869 Node* pin_count_cmp = _gvn.transform(new CmpUNode(pin_count, pin_count_rhs));
3870 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3871 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3872
3873 // True branch, pin count over/underflow.
3874 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3875 {
3876 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3877 // which will throw IllegalStateException for pin count over/underflow.
3878 // No memory changed so far - we can use memory create by reset_memory()
3879 // at the beginning of this intrinsic. No need to call reset_memory() again.
3880 PreserveJVMState pjvms(this);
3881 set_control(pin_count_over_underflow);
3882 uncommon_trap(Deoptimization::Reason_intrinsic,
3883 Deoptimization::Action_none);
3884 assert(stopped(), "invariant");
3885 }
3886
3887 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3888 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3889 set_control(valid_pin_count);
3890
3891 Node* next_pin_count;
3892 if (unpin) {
3893 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3894 } else {
3895 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3896 }
3897
3898 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3899
3900 // Result of top level CFG and Memory.
3901 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3902 record_for_igvn(result_rgn);
3903 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3904 record_for_igvn(result_mem);
3905
3906 result_rgn->init_req(_true_path, valid_pin_count);
3907 result_rgn->init_req(_false_path, continuation_is_null);
3908 result_mem->init_req(_true_path, reset_memory());
3909 result_mem->init_req(_false_path, input_memory_state);
3910
3911 // Set output state.
3912 set_control(_gvn.transform(result_rgn));
3913 set_all_memory(_gvn.transform(result_mem));
3914
3915 return true;
3916 }
3917
3918 //---------------------------load_mirror_from_klass----------------------------
3919 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3920 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3921 Node* p = basic_plus_adr(top(), klass, in_bytes(Klass::java_mirror_offset()));
3922 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3923 // mirror = ((OopHandle)mirror)->resolve();
3924 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3925 }
3926
3927 //-----------------------load_klass_from_mirror_common-------------------------
3928 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3929 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3930 // and branch to the given path on the region.
3931 // If never_see_null, take an uncommon trap on null, so we can optimistically
3932 // compile for the non-null case.
3933 // If the region is null, force never_see_null = true.
3934 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3935 bool never_see_null,
3936 RegionNode* region,
3937 int null_path,
3938 int offset) {
3939 if (region == nullptr) never_see_null = true;
3940 Node* p = basic_plus_adr(mirror, offset);
3941 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3942 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3943 Node* null_ctl = top();
3944 kls = null_check_oop(kls, &null_ctl, never_see_null);
3945 if (region != nullptr) {
3946 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3947 region->init_req(null_path, null_ctl);
3948 } else {
3949 assert(null_ctl == top(), "no loose ends");
3950 }
3951 return kls;
3952 }
3953
3954 //--------------------(inline_native_Class_query helpers)---------------------
3955 // Use this for JVM_ACC_INTERFACE.
3956 // Fall through if (mods & mask) == bits, take the guard otherwise.
3957 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3958 ByteSize offset, const Type* type, BasicType bt) {
3959 // Branch around if the given klass has the given modifier bit set.
3960 // Like generate_guard, adds a new path onto the region.
3961 Node* modp = basic_plus_adr(top(), kls, in_bytes(offset));
3962 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3963 Node* mask = intcon(modifier_mask);
3964 Node* bits = intcon(modifier_bits);
3965 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3966 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3967 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3968 return generate_fair_guard(bol, region);
3969 }
3970 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3971 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3972 InstanceKlass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3973 }
3974
3975 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3976 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3977 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3978 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3979 }
3980
3981 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3982 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3983 }
3984
3985 //-------------------------inline_native_Class_query-------------------
3986 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3987 const Type* return_type = TypeInt::BOOL;
3988 Node* prim_return_value = top(); // what happens if it's a primitive class?
3989 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3990 bool expect_prim = false; // most of these guys expect to work on refs
3991
3992 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3993
3994 Node* mirror = argument(0);
3995 Node* obj = top();
3996
3997 switch (id) {
3998 case vmIntrinsics::_isInstance:
3999 // nothing is an instance of a primitive type
4000 prim_return_value = intcon(0);
4001 obj = argument(1);
4002 break;
4003 case vmIntrinsics::_isHidden:
4004 prim_return_value = intcon(0);
4005 break;
4006 case vmIntrinsics::_getSuperclass:
4007 prim_return_value = null();
4008 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
4009 break;
4010 default:
4011 fatal_unexpected_iid(id);
4012 break;
4013 }
4014
4015 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4016 if (mirror_con == nullptr) return false; // cannot happen?
4017
4018 #ifndef PRODUCT
4019 if (C->print_intrinsics() || C->print_inlining()) {
4020 ciType* k = mirror_con->java_mirror_type();
4021 if (k) {
4022 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
4023 k->print_name();
4024 tty->cr();
4025 }
4026 }
4027 #endif
4028
4029 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
4030 RegionNode* region = new RegionNode(PATH_LIMIT);
4031 record_for_igvn(region);
4032 PhiNode* phi = new PhiNode(region, return_type);
4033
4034 // The mirror will never be null of Reflection.getClassAccessFlags, however
4035 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4036 // if it is. See bug 4774291.
4037
4038 // For Reflection.getClassAccessFlags(), the null check occurs in
4039 // the wrong place; see inline_unsafe_access(), above, for a similar
4040 // situation.
4041 mirror = null_check(mirror);
4042 // If mirror or obj is dead, only null-path is taken.
4043 if (stopped()) return true;
4044
4045 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4046
4047 // Now load the mirror's klass metaobject, and null-check it.
4048 // Side-effects region with the control path if the klass is null.
4049 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4050 // If kls is null, we have a primitive mirror.
4051 phi->init_req(_prim_path, prim_return_value);
4052 if (stopped()) { set_result(region, phi); return true; }
4053 bool safe_for_replace = (region->in(_prim_path) == top());
4054
4055 Node* p; // handy temp
4056 Node* null_ctl;
4057
4058 // Now that we have the non-null klass, we can perform the real query.
4059 // For constant classes, the query will constant-fold in LoadNode::Value.
4060 Node* query_value = top();
4061 switch (id) {
4062 case vmIntrinsics::_isInstance:
4063 // nothing is an instance of a primitive type
4064 query_value = gen_instanceof(obj, kls, safe_for_replace);
4065 break;
4066
4067 case vmIntrinsics::_isHidden:
4068 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4069 if (generate_hidden_class_guard(kls, region) != nullptr)
4070 // A guard was added. If the guard is taken, it was an hidden class.
4071 phi->add_req(intcon(1));
4072 // If we fall through, it's a plain class.
4073 query_value = intcon(0);
4074 break;
4075
4076
4077 case vmIntrinsics::_getSuperclass:
4078 // The rules here are somewhat unfortunate, but we can still do better
4079 // with random logic than with a JNI call.
4080 // Interfaces store null or Object as _super, but must report null.
4081 // Arrays store an intermediate super as _super, but must report Object.
4082 // Other types can report the actual _super.
4083 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4084 if (generate_array_guard(kls, region) != nullptr) {
4085 // A guard was added. If the guard is taken, it was an array.
4086 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4087 }
4088 // Check for interface after array since this checks AccessFlags offset into InstanceKlass.
4089 // In other words, we are accessing subtype-specific information, so we need to determine the subtype first.
4090 if (generate_interface_guard(kls, region) != nullptr) {
4091 // A guard was added. If the guard is taken, it was an interface.
4092 phi->add_req(null());
4093 }
4094 // If we fall through, it's a plain class. Get its _super.
4095 p = basic_plus_adr(top(), kls, in_bytes(Klass::super_offset()));
4096 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4097 null_ctl = top();
4098 kls = null_check_oop(kls, &null_ctl);
4099 if (null_ctl != top()) {
4100 // If the guard is taken, Object.superClass is null (both klass and mirror).
4101 region->add_req(null_ctl);
4102 phi ->add_req(null());
4103 }
4104 if (!stopped()) {
4105 query_value = load_mirror_from_klass(kls);
4106 }
4107 break;
4108
4109 default:
4110 fatal_unexpected_iid(id);
4111 break;
4112 }
4113
4114 // Fall-through is the normal case of a query to a real class.
4115 phi->init_req(1, query_value);
4116 region->init_req(1, control());
4117
4118 C->set_has_split_ifs(true); // Has chance for split-if optimization
4119 set_result(region, phi);
4120 return true;
4121 }
4122
4123 //-------------------------inline_Class_cast-------------------
4124 bool LibraryCallKit::inline_Class_cast() {
4125 Node* mirror = argument(0); // Class
4126 Node* obj = argument(1);
4127 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4128 if (mirror_con == nullptr) {
4129 return false; // dead path (mirror->is_top()).
4130 }
4131 if (obj == nullptr || obj->is_top()) {
4132 return false; // dead path
4133 }
4134 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4135
4136 // First, see if Class.cast() can be folded statically.
4137 // java_mirror_type() returns non-null for compile-time Class constants.
4138 ciType* tm = mirror_con->java_mirror_type();
4139 if (tm != nullptr && tm->is_klass() &&
4140 tp != nullptr) {
4141 if (!tp->is_loaded()) {
4142 // Don't use intrinsic when class is not loaded.
4143 return false;
4144 } else {
4145 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4146 if (static_res == Compile::SSC_always_true) {
4147 // isInstance() is true - fold the code.
4148 set_result(obj);
4149 return true;
4150 } else if (static_res == Compile::SSC_always_false) {
4151 // Don't use intrinsic, have to throw ClassCastException.
4152 // If the reference is null, the non-intrinsic bytecode will
4153 // be optimized appropriately.
4154 return false;
4155 }
4156 }
4157 }
4158
4159 // Bailout intrinsic and do normal inlining if exception path is frequent.
4160 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4161 return false;
4162 }
4163
4164 // Generate dynamic checks.
4165 // Class.cast() is java implementation of _checkcast bytecode.
4166 // Do checkcast (Parse::do_checkcast()) optimizations here.
4167
4168 mirror = null_check(mirror);
4169 // If mirror is dead, only null-path is taken.
4170 if (stopped()) {
4171 return true;
4172 }
4173
4174 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4175 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4176 RegionNode* region = new RegionNode(PATH_LIMIT);
4177 record_for_igvn(region);
4178
4179 // Now load the mirror's klass metaobject, and null-check it.
4180 // If kls is null, we have a primitive mirror and
4181 // nothing is an instance of a primitive type.
4182 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4183
4184 Node* res = top();
4185 if (!stopped()) {
4186 Node* bad_type_ctrl = top();
4187 // Do checkcast optimizations.
4188 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4189 region->init_req(_bad_type_path, bad_type_ctrl);
4190 }
4191 if (region->in(_prim_path) != top() ||
4192 region->in(_bad_type_path) != top()) {
4193 // Let Interpreter throw ClassCastException.
4194 PreserveJVMState pjvms(this);
4195 set_control(_gvn.transform(region));
4196 uncommon_trap(Deoptimization::Reason_intrinsic,
4197 Deoptimization::Action_maybe_recompile);
4198 }
4199 if (!stopped()) {
4200 set_result(res);
4201 }
4202 return true;
4203 }
4204
4205
4206 //--------------------------inline_native_subtype_check------------------------
4207 // This intrinsic takes the JNI calls out of the heart of
4208 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4209 bool LibraryCallKit::inline_native_subtype_check() {
4210 // Pull both arguments off the stack.
4211 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4212 args[0] = argument(0);
4213 args[1] = argument(1);
4214 Node* klasses[2]; // corresponding Klasses: superk, subk
4215 klasses[0] = klasses[1] = top();
4216
4217 enum {
4218 // A full decision tree on {superc is prim, subc is prim}:
4219 _prim_0_path = 1, // {P,N} => false
4220 // {P,P} & superc!=subc => false
4221 _prim_same_path, // {P,P} & superc==subc => true
4222 _prim_1_path, // {N,P} => false
4223 _ref_subtype_path, // {N,N} & subtype check wins => true
4224 _both_ref_path, // {N,N} & subtype check loses => false
4225 PATH_LIMIT
4226 };
4227
4228 RegionNode* region = new RegionNode(PATH_LIMIT);
4229 Node* phi = new PhiNode(region, TypeInt::BOOL);
4230 record_for_igvn(region);
4231
4232 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4233 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4234 int class_klass_offset = java_lang_Class::klass_offset();
4235
4236 // First null-check both mirrors and load each mirror's klass metaobject.
4237 int which_arg;
4238 for (which_arg = 0; which_arg <= 1; which_arg++) {
4239 Node* arg = args[which_arg];
4240 arg = null_check(arg);
4241 if (stopped()) break;
4242 args[which_arg] = arg;
4243
4244 Node* p = basic_plus_adr(arg, class_klass_offset);
4245 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4246 klasses[which_arg] = _gvn.transform(kls);
4247 }
4248
4249 // Having loaded both klasses, test each for null.
4250 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4251 for (which_arg = 0; which_arg <= 1; which_arg++) {
4252 Node* kls = klasses[which_arg];
4253 Node* null_ctl = top();
4254 kls = null_check_oop(kls, &null_ctl, never_see_null);
4255 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4256 region->init_req(prim_path, null_ctl);
4257 if (stopped()) break;
4258 klasses[which_arg] = kls;
4259 }
4260
4261 if (!stopped()) {
4262 // now we have two reference types, in klasses[0..1]
4263 Node* subk = klasses[1]; // the argument to isAssignableFrom
4264 Node* superk = klasses[0]; // the receiver
4265 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4266 // now we have a successful reference subtype check
4267 region->set_req(_ref_subtype_path, control());
4268 }
4269
4270 // If both operands are primitive (both klasses null), then
4271 // we must return true when they are identical primitives.
4272 // It is convenient to test this after the first null klass check.
4273 set_control(region->in(_prim_0_path)); // go back to first null check
4274 if (!stopped()) {
4275 // Since superc is primitive, make a guard for the superc==subc case.
4276 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4277 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4278 generate_guard(bol_eq, region, PROB_FAIR);
4279 if (region->req() == PATH_LIMIT+1) {
4280 // A guard was added. If the added guard is taken, superc==subc.
4281 region->swap_edges(PATH_LIMIT, _prim_same_path);
4282 region->del_req(PATH_LIMIT);
4283 }
4284 region->set_req(_prim_0_path, control()); // Not equal after all.
4285 }
4286
4287 // these are the only paths that produce 'true':
4288 phi->set_req(_prim_same_path, intcon(1));
4289 phi->set_req(_ref_subtype_path, intcon(1));
4290
4291 // pull together the cases:
4292 assert(region->req() == PATH_LIMIT, "sane region");
4293 for (uint i = 1; i < region->req(); i++) {
4294 Node* ctl = region->in(i);
4295 if (ctl == nullptr || ctl == top()) {
4296 region->set_req(i, top());
4297 phi ->set_req(i, top());
4298 } else if (phi->in(i) == nullptr) {
4299 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4300 }
4301 }
4302
4303 set_control(_gvn.transform(region));
4304 set_result(_gvn.transform(phi));
4305 return true;
4306 }
4307
4308 //---------------------generate_array_guard_common------------------------
4309 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4310 bool obj_array, bool not_array, Node** obj) {
4311
4312 if (stopped()) {
4313 return nullptr;
4314 }
4315
4316 // If obj_array/non_array==false/false:
4317 // Branch around if the given klass is in fact an array (either obj or prim).
4318 // If obj_array/non_array==false/true:
4319 // Branch around if the given klass is not an array klass of any kind.
4320 // If obj_array/non_array==true/true:
4321 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4322 // If obj_array/non_array==true/false:
4323 // Branch around if the kls is an oop array (Object[] or subtype)
4324 //
4325 // Like generate_guard, adds a new path onto the region.
4326 jint layout_con = 0;
4327 Node* layout_val = get_layout_helper(kls, layout_con);
4328 if (layout_val == nullptr) {
4329 bool query = (obj_array
4330 ? Klass::layout_helper_is_objArray(layout_con)
4331 : Klass::layout_helper_is_array(layout_con));
4332 if (query == not_array) {
4333 return nullptr; // never a branch
4334 } else { // always a branch
4335 Node* always_branch = control();
4336 if (region != nullptr)
4337 region->add_req(always_branch);
4338 set_control(top());
4339 return always_branch;
4340 }
4341 }
4342 // Now test the correct condition.
4343 jint nval = (obj_array
4344 ? (jint)(Klass::_lh_array_tag_type_value
4345 << Klass::_lh_array_tag_shift)
4346 : Klass::_lh_neutral_value);
4347 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4348 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4349 // invert the test if we are looking for a non-array
4350 if (not_array) btest = BoolTest(btest).negate();
4351 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4352 Node* ctrl = generate_fair_guard(bol, region);
4353 Node* is_array_ctrl = not_array ? control() : ctrl;
4354 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4355 // Keep track of the fact that 'obj' is an array to prevent
4356 // array specific accesses from floating above the guard.
4357 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4358 }
4359 return ctrl;
4360 }
4361
4362
4363 //-----------------------inline_native_newArray--------------------------
4364 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4365 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4366 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4367 Node* mirror;
4368 Node* count_val;
4369 if (uninitialized) {
4370 null_check_receiver();
4371 mirror = argument(1);
4372 count_val = argument(2);
4373 } else {
4374 mirror = argument(0);
4375 count_val = argument(1);
4376 }
4377
4378 mirror = null_check(mirror);
4379 // If mirror or obj is dead, only null-path is taken.
4380 if (stopped()) return true;
4381
4382 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4383 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4384 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4385 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4386 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4387
4388 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4389 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4390 result_reg, _slow_path);
4391 Node* normal_ctl = control();
4392 Node* no_array_ctl = result_reg->in(_slow_path);
4393
4394 // Generate code for the slow case. We make a call to newArray().
4395 set_control(no_array_ctl);
4396 if (!stopped()) {
4397 // Either the input type is void.class, or else the
4398 // array klass has not yet been cached. Either the
4399 // ensuing call will throw an exception, or else it
4400 // will cache the array klass for next time.
4401 PreserveJVMState pjvms(this);
4402 CallJavaNode* slow_call = nullptr;
4403 if (uninitialized) {
4404 // Generate optimized virtual call (holder class 'Unsafe' is final)
4405 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4406 } else {
4407 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4408 }
4409 Node* slow_result = set_results_for_java_call(slow_call);
4410 // this->control() comes from set_results_for_java_call
4411 result_reg->set_req(_slow_path, control());
4412 result_val->set_req(_slow_path, slow_result);
4413 result_io ->set_req(_slow_path, i_o());
4414 result_mem->set_req(_slow_path, reset_memory());
4415 }
4416
4417 set_control(normal_ctl);
4418 if (!stopped()) {
4419 // Normal case: The array type has been cached in the java.lang.Class.
4420 // The following call works fine even if the array type is polymorphic.
4421 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4422 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4423 result_reg->init_req(_normal_path, control());
4424 result_val->init_req(_normal_path, obj);
4425 result_io ->init_req(_normal_path, i_o());
4426 result_mem->init_req(_normal_path, reset_memory());
4427
4428 if (uninitialized) {
4429 // Mark the allocation so that zeroing is skipped
4430 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4431 alloc->maybe_set_complete(&_gvn);
4432 }
4433 }
4434
4435 // Return the combined state.
4436 set_i_o( _gvn.transform(result_io) );
4437 set_all_memory( _gvn.transform(result_mem));
4438
4439 C->set_has_split_ifs(true); // Has chance for split-if optimization
4440 set_result(result_reg, result_val);
4441 return true;
4442 }
4443
4444 //----------------------inline_native_getLength--------------------------
4445 // public static native int java.lang.reflect.Array.getLength(Object array);
4446 bool LibraryCallKit::inline_native_getLength() {
4447 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4448
4449 Node* array = null_check(argument(0));
4450 // If array is dead, only null-path is taken.
4451 if (stopped()) return true;
4452
4453 // Deoptimize if it is a non-array.
4454 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4455
4456 if (non_array != nullptr) {
4457 PreserveJVMState pjvms(this);
4458 set_control(non_array);
4459 uncommon_trap(Deoptimization::Reason_intrinsic,
4460 Deoptimization::Action_maybe_recompile);
4461 }
4462
4463 // If control is dead, only non-array-path is taken.
4464 if (stopped()) return true;
4465
4466 // The works fine even if the array type is polymorphic.
4467 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4468 Node* result = load_array_length(array);
4469
4470 C->set_has_split_ifs(true); // Has chance for split-if optimization
4471 set_result(result);
4472 return true;
4473 }
4474
4475 //------------------------inline_array_copyOf----------------------------
4476 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4477 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4478 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4479 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4480
4481 // Get the arguments.
4482 Node* original = argument(0);
4483 Node* start = is_copyOfRange? argument(1): intcon(0);
4484 Node* end = is_copyOfRange? argument(2): argument(1);
4485 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4486
4487 Node* newcopy = nullptr;
4488
4489 // Set the original stack and the reexecute bit for the interpreter to reexecute
4490 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4491 { PreserveReexecuteState preexecs(this);
4492 jvms()->set_should_reexecute(true);
4493
4494 array_type_mirror = null_check(array_type_mirror);
4495 original = null_check(original);
4496
4497 // Check if a null path was taken unconditionally.
4498 if (stopped()) return true;
4499
4500 Node* orig_length = load_array_length(original);
4501
4502 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4503 klass_node = null_check(klass_node);
4504
4505 RegionNode* bailout = new RegionNode(1);
4506 record_for_igvn(bailout);
4507
4508 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4509 // Bail out if that is so.
4510 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4511 if (not_objArray != nullptr) {
4512 // Improve the klass node's type from the new optimistic assumption:
4513 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4514 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4515 Node* cast = new CastPPNode(control(), klass_node, akls);
4516 klass_node = _gvn.transform(cast);
4517 }
4518
4519 // Bail out if either start or end is negative.
4520 generate_negative_guard(start, bailout, &start);
4521 generate_negative_guard(end, bailout, &end);
4522
4523 Node* length = end;
4524 if (_gvn.type(start) != TypeInt::ZERO) {
4525 length = _gvn.transform(new SubINode(end, start));
4526 }
4527
4528 // Bail out if length is negative (i.e., if start > end).
4529 // Without this the new_array would throw
4530 // NegativeArraySizeException but IllegalArgumentException is what
4531 // should be thrown
4532 generate_negative_guard(length, bailout, &length);
4533
4534 // Bail out if start is larger than the original length
4535 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4536 generate_negative_guard(orig_tail, bailout, &orig_tail);
4537
4538 if (bailout->req() > 1) {
4539 PreserveJVMState pjvms(this);
4540 set_control(_gvn.transform(bailout));
4541 uncommon_trap(Deoptimization::Reason_intrinsic,
4542 Deoptimization::Action_maybe_recompile);
4543 }
4544
4545 if (!stopped()) {
4546 // How many elements will we copy from the original?
4547 // The answer is MinI(orig_tail, length).
4548 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4549
4550 // Generate a direct call to the right arraycopy function(s).
4551 // We know the copy is disjoint but we might not know if the
4552 // oop stores need checking.
4553 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4554 // This will fail a store-check if x contains any non-nulls.
4555
4556 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4557 // loads/stores but it is legal only if we're sure the
4558 // Arrays.copyOf would succeed. So we need all input arguments
4559 // to the copyOf to be validated, including that the copy to the
4560 // new array won't trigger an ArrayStoreException. That subtype
4561 // check can be optimized if we know something on the type of
4562 // the input array from type speculation.
4563 if (_gvn.type(klass_node)->singleton()) {
4564 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4565 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4566
4567 int test = C->static_subtype_check(superk, subk);
4568 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4569 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4570 if (t_original->speculative_type() != nullptr) {
4571 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4572 }
4573 }
4574 }
4575
4576 bool validated = false;
4577 // Reason_class_check rather than Reason_intrinsic because we
4578 // want to intrinsify even if this traps.
4579 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4580 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4581
4582 if (not_subtype_ctrl != top()) {
4583 PreserveJVMState pjvms(this);
4584 set_control(not_subtype_ctrl);
4585 uncommon_trap(Deoptimization::Reason_class_check,
4586 Deoptimization::Action_make_not_entrant);
4587 assert(stopped(), "Should be stopped");
4588 }
4589 validated = true;
4590 }
4591
4592 if (!stopped()) {
4593 newcopy = new_array(klass_node, length, 0); // no arguments to push
4594
4595 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4596 load_object_klass(original), klass_node);
4597 if (!is_copyOfRange) {
4598 ac->set_copyof(validated);
4599 } else {
4600 ac->set_copyofrange(validated);
4601 }
4602 Node* n = _gvn.transform(ac);
4603 if (n == ac) {
4604 ac->connect_outputs(this);
4605 } else {
4606 assert(validated, "shouldn't transform if all arguments not validated");
4607 set_all_memory(n);
4608 }
4609 }
4610 }
4611 } // original reexecute is set back here
4612
4613 C->set_has_split_ifs(true); // Has chance for split-if optimization
4614 if (!stopped()) {
4615 set_result(newcopy);
4616 }
4617 return true;
4618 }
4619
4620
4621 //----------------------generate_virtual_guard---------------------------
4622 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4623 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4624 RegionNode* slow_region) {
4625 ciMethod* method = callee();
4626 int vtable_index = method->vtable_index();
4627 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4628 "bad index %d", vtable_index);
4629 // Get the Method* out of the appropriate vtable entry.
4630 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4631 vtable_index*vtableEntry::size_in_bytes() +
4632 in_bytes(vtableEntry::method_offset());
4633 Node* entry_addr = basic_plus_adr(top(), obj_klass, entry_offset);
4634 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4635
4636 // Compare the target method with the expected method (e.g., Object.hashCode).
4637 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4638
4639 Node* native_call = makecon(native_call_addr);
4640 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4641 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4642
4643 return generate_slow_guard(test_native, slow_region);
4644 }
4645
4646 //-----------------------generate_method_call----------------------------
4647 // Use generate_method_call to make a slow-call to the real
4648 // method if the fast path fails. An alternative would be to
4649 // use a stub like OptoRuntime::slow_arraycopy_Java.
4650 // This only works for expanding the current library call,
4651 // not another intrinsic. (E.g., don't use this for making an
4652 // arraycopy call inside of the copyOf intrinsic.)
4653 CallJavaNode*
4654 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4655 // When compiling the intrinsic method itself, do not use this technique.
4656 guarantee(callee() != C->method(), "cannot make slow-call to self");
4657
4658 ciMethod* method = callee();
4659 // ensure the JVMS we have will be correct for this call
4660 guarantee(method_id == method->intrinsic_id(), "must match");
4661
4662 const TypeFunc* tf = TypeFunc::make(method);
4663 if (res_not_null) {
4664 assert(tf->return_type() == T_OBJECT, "");
4665 const TypeTuple* range = tf->range();
4666 const Type** fields = TypeTuple::fields(range->cnt());
4667 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4668 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4669 tf = TypeFunc::make(tf->domain(), new_range);
4670 }
4671 CallJavaNode* slow_call;
4672 if (is_static) {
4673 assert(!is_virtual, "");
4674 slow_call = new CallStaticJavaNode(C, tf,
4675 SharedRuntime::get_resolve_static_call_stub(), method);
4676 } else if (is_virtual) {
4677 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4678 int vtable_index = Method::invalid_vtable_index;
4679 if (UseInlineCaches) {
4680 // Suppress the vtable call
4681 } else {
4682 // hashCode and clone are not a miranda methods,
4683 // so the vtable index is fixed.
4684 // No need to use the linkResolver to get it.
4685 vtable_index = method->vtable_index();
4686 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4687 "bad index %d", vtable_index);
4688 }
4689 slow_call = new CallDynamicJavaNode(tf,
4690 SharedRuntime::get_resolve_virtual_call_stub(),
4691 method, vtable_index);
4692 } else { // neither virtual nor static: opt_virtual
4693 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4694 slow_call = new CallStaticJavaNode(C, tf,
4695 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4696 slow_call->set_optimized_virtual(true);
4697 }
4698 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4699 // To be able to issue a direct call (optimized virtual or virtual)
4700 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4701 // about the method being invoked should be attached to the call site to
4702 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4703 slow_call->set_override_symbolic_info(true);
4704 }
4705 set_arguments_for_java_call(slow_call);
4706 set_edges_for_java_call(slow_call);
4707 return slow_call;
4708 }
4709
4710
4711 /**
4712 * Build special case code for calls to hashCode on an object. This call may
4713 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4714 * slightly different code.
4715 */
4716 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4717 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4718 assert(!(is_virtual && is_static), "either virtual, special, or static");
4719
4720 enum { _slow_path = 1, _fast_path, _null_path, PATH_LIMIT };
4721
4722 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4723 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4724 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4725 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4726 Node* obj = nullptr;
4727 if (!is_static) {
4728 // Check for hashing null object
4729 obj = null_check_receiver();
4730 if (stopped()) return true; // unconditionally null
4731 result_reg->init_req(_null_path, top());
4732 result_val->init_req(_null_path, top());
4733 } else {
4734 // Do a null check, and return zero if null.
4735 // System.identityHashCode(null) == 0
4736 obj = argument(0);
4737 Node* null_ctl = top();
4738 obj = null_check_oop(obj, &null_ctl);
4739 result_reg->init_req(_null_path, null_ctl);
4740 result_val->init_req(_null_path, _gvn.intcon(0));
4741 }
4742
4743 // Unconditionally null? Then return right away.
4744 if (stopped()) {
4745 set_control( result_reg->in(_null_path));
4746 if (!stopped())
4747 set_result(result_val->in(_null_path));
4748 return true;
4749 }
4750
4751 // We only go to the fast case code if we pass a number of guards. The
4752 // paths which do not pass are accumulated in the slow_region.
4753 RegionNode* slow_region = new RegionNode(1);
4754 record_for_igvn(slow_region);
4755
4756 // If this is a virtual call, we generate a funny guard. We pull out
4757 // the vtable entry corresponding to hashCode() from the target object.
4758 // If the target method which we are calling happens to be the native
4759 // Object hashCode() method, we pass the guard. We do not need this
4760 // guard for non-virtual calls -- the caller is known to be the native
4761 // Object hashCode().
4762 if (is_virtual) {
4763 // After null check, get the object's klass.
4764 Node* obj_klass = load_object_klass(obj);
4765 generate_virtual_guard(obj_klass, slow_region);
4766 }
4767
4768 // Get the header out of the object, use LoadMarkNode when available
4769 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4770 // The control of the load must be null. Otherwise, the load can move before
4771 // the null check after castPP removal.
4772 Node* no_ctrl = nullptr;
4773 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4774
4775 if (!UseObjectMonitorTable) {
4776 // Test the header to see if it is safe to read w.r.t. locking.
4777 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4778 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4779 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4780 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4781 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4782
4783 generate_slow_guard(test_monitor, slow_region);
4784 }
4785
4786 // Get the hash value and check to see that it has been properly assigned.
4787 // We depend on hash_mask being at most 32 bits and avoid the use of
4788 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4789 // vm: see markWord.hpp.
4790 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4791 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4792 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4793 // This hack lets the hash bits live anywhere in the mark object now, as long
4794 // as the shift drops the relevant bits into the low 32 bits. Note that
4795 // Java spec says that HashCode is an int so there's no point in capturing
4796 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4797 hshifted_header = ConvX2I(hshifted_header);
4798 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4799
4800 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4801 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4802 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4803
4804 generate_slow_guard(test_assigned, slow_region);
4805
4806 Node* init_mem = reset_memory();
4807 // fill in the rest of the null path:
4808 result_io ->init_req(_null_path, i_o());
4809 result_mem->init_req(_null_path, init_mem);
4810
4811 result_val->init_req(_fast_path, hash_val);
4812 result_reg->init_req(_fast_path, control());
4813 result_io ->init_req(_fast_path, i_o());
4814 result_mem->init_req(_fast_path, init_mem);
4815
4816 // Generate code for the slow case. We make a call to hashCode().
4817 set_control(_gvn.transform(slow_region));
4818 if (!stopped()) {
4819 // No need for PreserveJVMState, because we're using up the present state.
4820 set_all_memory(init_mem);
4821 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4822 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4823 Node* slow_result = set_results_for_java_call(slow_call);
4824 // this->control() comes from set_results_for_java_call
4825 result_reg->init_req(_slow_path, control());
4826 result_val->init_req(_slow_path, slow_result);
4827 result_io ->set_req(_slow_path, i_o());
4828 result_mem ->set_req(_slow_path, reset_memory());
4829 }
4830
4831 // Return the combined state.
4832 set_i_o( _gvn.transform(result_io) );
4833 set_all_memory( _gvn.transform(result_mem));
4834
4835 set_result(result_reg, result_val);
4836 return true;
4837 }
4838
4839 //---------------------------inline_native_getClass----------------------------
4840 // public final native Class<?> java.lang.Object.getClass();
4841 //
4842 // Build special case code for calls to getClass on an object.
4843 bool LibraryCallKit::inline_native_getClass() {
4844 Node* obj = null_check_receiver();
4845 if (stopped()) return true;
4846 set_result(load_mirror_from_klass(load_object_klass(obj)));
4847 return true;
4848 }
4849
4850 //-----------------inline_native_Reflection_getCallerClass---------------------
4851 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4852 //
4853 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4854 //
4855 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4856 // in that it must skip particular security frames and checks for
4857 // caller sensitive methods.
4858 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4859 #ifndef PRODUCT
4860 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4861 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4862 }
4863 #endif
4864
4865 if (!jvms()->has_method()) {
4866 #ifndef PRODUCT
4867 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4868 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4869 }
4870 #endif
4871 return false;
4872 }
4873
4874 // Walk back up the JVM state to find the caller at the required
4875 // depth.
4876 JVMState* caller_jvms = jvms();
4877
4878 // Cf. JVM_GetCallerClass
4879 // NOTE: Start the loop at depth 1 because the current JVM state does
4880 // not include the Reflection.getCallerClass() frame.
4881 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
4882 ciMethod* m = caller_jvms->method();
4883 switch (n) {
4884 case 0:
4885 fatal("current JVM state does not include the Reflection.getCallerClass frame");
4886 break;
4887 case 1:
4888 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
4889 if (!m->caller_sensitive()) {
4890 #ifndef PRODUCT
4891 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4892 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
4893 }
4894 #endif
4895 return false; // bail-out; let JVM_GetCallerClass do the work
4896 }
4897 break;
4898 default:
4899 if (!m->is_ignored_by_security_stack_walk()) {
4900 // We have reached the desired frame; return the holder class.
4901 // Acquire method holder as java.lang.Class and push as constant.
4902 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
4903 ciInstance* caller_mirror = caller_klass->java_mirror();
4904 set_result(makecon(TypeInstPtr::make(caller_mirror)));
4905
4906 #ifndef PRODUCT
4907 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4908 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());
4909 tty->print_cr(" JVM state at this point:");
4910 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4911 ciMethod* m = jvms()->of_depth(i)->method();
4912 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4913 }
4914 }
4915 #endif
4916 return true;
4917 }
4918 break;
4919 }
4920 }
4921
4922 #ifndef PRODUCT
4923 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4924 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
4925 tty->print_cr(" JVM state at this point:");
4926 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
4927 ciMethod* m = jvms()->of_depth(i)->method();
4928 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
4929 }
4930 }
4931 #endif
4932
4933 return false; // bail-out; let JVM_GetCallerClass do the work
4934 }
4935
4936 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
4937 Node* arg = argument(0);
4938 Node* result = nullptr;
4939
4940 switch (id) {
4941 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
4942 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
4943 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
4944 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
4945 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
4946 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
4947
4948 case vmIntrinsics::_doubleToLongBits: {
4949 // two paths (plus control) merge in a wood
4950 RegionNode *r = new RegionNode(3);
4951 Node *phi = new PhiNode(r, TypeLong::LONG);
4952
4953 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
4954 // Build the boolean node
4955 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4956
4957 // Branch either way.
4958 // NaN case is less traveled, which makes all the difference.
4959 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
4960 Node *opt_isnan = _gvn.transform(ifisnan);
4961 assert( opt_isnan->is_If(), "Expect an IfNode");
4962 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
4963 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
4964
4965 set_control(iftrue);
4966
4967 static const jlong nan_bits = CONST64(0x7ff8000000000000);
4968 Node *slow_result = longcon(nan_bits); // return NaN
4969 phi->init_req(1, _gvn.transform( slow_result ));
4970 r->init_req(1, iftrue);
4971
4972 // Else fall through
4973 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
4974 set_control(iffalse);
4975
4976 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
4977 r->init_req(2, iffalse);
4978
4979 // Post merge
4980 set_control(_gvn.transform(r));
4981 record_for_igvn(r);
4982
4983 C->set_has_split_ifs(true); // Has chance for split-if optimization
4984 result = phi;
4985 assert(result->bottom_type()->isa_long(), "must be");
4986 break;
4987 }
4988
4989 case vmIntrinsics::_floatToIntBits: {
4990 // two paths (plus control) merge in a wood
4991 RegionNode *r = new RegionNode(3);
4992 Node *phi = new PhiNode(r, TypeInt::INT);
4993
4994 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
4995 // Build the boolean node
4996 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
4997
4998 // Branch either way.
4999 // NaN case is less traveled, which makes all the difference.
5000 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5001 Node *opt_isnan = _gvn.transform(ifisnan);
5002 assert( opt_isnan->is_If(), "Expect an IfNode");
5003 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5004 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5005
5006 set_control(iftrue);
5007
5008 static const jint nan_bits = 0x7fc00000;
5009 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5010 phi->init_req(1, _gvn.transform( slow_result ));
5011 r->init_req(1, iftrue);
5012
5013 // Else fall through
5014 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5015 set_control(iffalse);
5016
5017 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5018 r->init_req(2, iffalse);
5019
5020 // Post merge
5021 set_control(_gvn.transform(r));
5022 record_for_igvn(r);
5023
5024 C->set_has_split_ifs(true); // Has chance for split-if optimization
5025 result = phi;
5026 assert(result->bottom_type()->isa_int(), "must be");
5027 break;
5028 }
5029
5030 default:
5031 fatal_unexpected_iid(id);
5032 break;
5033 }
5034 set_result(_gvn.transform(result));
5035 return true;
5036 }
5037
5038 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5039 Node* arg = argument(0);
5040 Node* result = nullptr;
5041
5042 switch (id) {
5043 case vmIntrinsics::_floatIsInfinite:
5044 result = new IsInfiniteFNode(arg);
5045 break;
5046 case vmIntrinsics::_floatIsFinite:
5047 result = new IsFiniteFNode(arg);
5048 break;
5049 case vmIntrinsics::_doubleIsInfinite:
5050 result = new IsInfiniteDNode(arg);
5051 break;
5052 case vmIntrinsics::_doubleIsFinite:
5053 result = new IsFiniteDNode(arg);
5054 break;
5055 default:
5056 fatal_unexpected_iid(id);
5057 break;
5058 }
5059 set_result(_gvn.transform(result));
5060 return true;
5061 }
5062
5063 //----------------------inline_unsafe_copyMemory-------------------------
5064 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5065
5066 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5067 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5068 const Type* base_t = gvn.type(base);
5069
5070 bool in_native = (base_t == TypePtr::NULL_PTR);
5071 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5072 bool is_mixed = !in_heap && !in_native;
5073
5074 if (is_mixed) {
5075 return true; // mixed accesses can touch both on-heap and off-heap memory
5076 }
5077 if (in_heap) {
5078 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5079 if (!is_prim_array) {
5080 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5081 // there's not enough type information available to determine proper memory slice for it.
5082 return true;
5083 }
5084 }
5085 return false;
5086 }
5087
5088 bool LibraryCallKit::inline_unsafe_copyMemory() {
5089 if (callee()->is_static()) return false; // caller must have the capability!
5090 null_check_receiver(); // null-check receiver
5091 if (stopped()) return true;
5092
5093 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5094
5095 Node* src_base = argument(1); // type: oop
5096 Node* src_off = ConvL2X(argument(2)); // type: long
5097 Node* dst_base = argument(4); // type: oop
5098 Node* dst_off = ConvL2X(argument(5)); // type: long
5099 Node* size = ConvL2X(argument(7)); // type: long
5100
5101 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5102 "fieldOffset must be byte-scaled");
5103
5104 Node* src_addr = make_unsafe_address(src_base, src_off);
5105 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5106
5107 Node* thread = _gvn.transform(new ThreadLocalNode());
5108 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5109 BasicType doing_unsafe_access_bt = T_BYTE;
5110 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5111
5112 // update volatile field
5113 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5114
5115 int flags = RC_LEAF | RC_NO_FP;
5116
5117 const TypePtr* dst_type = TypePtr::BOTTOM;
5118
5119 // Adjust memory effects of the runtime call based on input values.
5120 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5121 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5122 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5123
5124 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5125 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5126 flags |= RC_NARROW_MEM; // narrow in memory
5127 }
5128 }
5129
5130 // Call it. Note that the length argument is not scaled.
5131 make_runtime_call(flags,
5132 OptoRuntime::fast_arraycopy_Type(),
5133 StubRoutines::unsafe_arraycopy(),
5134 "unsafe_arraycopy",
5135 dst_type,
5136 src_addr, dst_addr, size XTOP);
5137
5138 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5139
5140 return true;
5141 }
5142
5143 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5144 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5145 bool LibraryCallKit::inline_unsafe_setMemory() {
5146 if (callee()->is_static()) return false; // caller must have the capability!
5147 null_check_receiver(); // null-check receiver
5148 if (stopped()) return true;
5149
5150 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5151
5152 Node* dst_base = argument(1); // type: oop
5153 Node* dst_off = ConvL2X(argument(2)); // type: long
5154 Node* size = ConvL2X(argument(4)); // type: long
5155 Node* byte = argument(6); // type: byte
5156
5157 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5158 "fieldOffset must be byte-scaled");
5159
5160 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5161
5162 Node* thread = _gvn.transform(new ThreadLocalNode());
5163 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5164 BasicType doing_unsafe_access_bt = T_BYTE;
5165 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5166
5167 // update volatile field
5168 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5169
5170 int flags = RC_LEAF | RC_NO_FP;
5171
5172 const TypePtr* dst_type = TypePtr::BOTTOM;
5173
5174 // Adjust memory effects of the runtime call based on input values.
5175 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5176 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5177
5178 flags |= RC_NARROW_MEM; // narrow in memory
5179 }
5180
5181 // Call it. Note that the length argument is not scaled.
5182 make_runtime_call(flags,
5183 OptoRuntime::unsafe_setmemory_Type(),
5184 StubRoutines::unsafe_setmemory(),
5185 "unsafe_setmemory",
5186 dst_type,
5187 dst_addr, size XTOP, byte);
5188
5189 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5190
5191 return true;
5192 }
5193
5194 #undef XTOP
5195
5196 //------------------------clone_coping-----------------------------------
5197 // Helper function for inline_native_clone.
5198 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5199 assert(obj_size != nullptr, "");
5200 Node* raw_obj = alloc_obj->in(1);
5201 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5202
5203 AllocateNode* alloc = nullptr;
5204 if (ReduceBulkZeroing &&
5205 // If we are implementing an array clone without knowing its source type
5206 // (can happen when compiling the array-guarded branch of a reflective
5207 // Object.clone() invocation), initialize the array within the allocation.
5208 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5209 // to a runtime clone call that assumes fully initialized source arrays.
5210 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5211 // We will be completely responsible for initializing this object -
5212 // mark Initialize node as complete.
5213 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5214 // The object was just allocated - there should be no any stores!
5215 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5216 // Mark as complete_with_arraycopy so that on AllocateNode
5217 // expansion, we know this AllocateNode is initialized by an array
5218 // copy and a StoreStore barrier exists after the array copy.
5219 alloc->initialization()->set_complete_with_arraycopy();
5220 }
5221
5222 Node* size = _gvn.transform(obj_size);
5223 access_clone(obj, alloc_obj, size, is_array);
5224
5225 // Do not let reads from the cloned object float above the arraycopy.
5226 if (alloc != nullptr) {
5227 // Do not let stores that initialize this object be reordered with
5228 // a subsequent store that would make this object accessible by
5229 // other threads.
5230 // Record what AllocateNode this StoreStore protects so that
5231 // escape analysis can go from the MemBarStoreStoreNode to the
5232 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5233 // based on the escape status of the AllocateNode.
5234 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5235 } else {
5236 insert_mem_bar(Op_MemBarCPUOrder);
5237 }
5238 }
5239
5240 //------------------------inline_native_clone----------------------------
5241 // protected native Object java.lang.Object.clone();
5242 //
5243 // Here are the simple edge cases:
5244 // null receiver => normal trap
5245 // virtual and clone was overridden => slow path to out-of-line clone
5246 // not cloneable or finalizer => slow path to out-of-line Object.clone
5247 //
5248 // The general case has two steps, allocation and copying.
5249 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5250 //
5251 // Copying also has two cases, oop arrays and everything else.
5252 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5253 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5254 //
5255 // These steps fold up nicely if and when the cloned object's klass
5256 // can be sharply typed as an object array, a type array, or an instance.
5257 //
5258 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5259 PhiNode* result_val;
5260
5261 // Set the reexecute bit for the interpreter to reexecute
5262 // the bytecode that invokes Object.clone if deoptimization happens.
5263 { PreserveReexecuteState preexecs(this);
5264 jvms()->set_should_reexecute(true);
5265
5266 Node* obj = null_check_receiver();
5267 if (stopped()) return true;
5268
5269 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5270
5271 // If we are going to clone an instance, we need its exact type to
5272 // know the number and types of fields to convert the clone to
5273 // loads/stores. Maybe a speculative type can help us.
5274 if (!obj_type->klass_is_exact() &&
5275 obj_type->speculative_type() != nullptr &&
5276 obj_type->speculative_type()->is_instance_klass()) {
5277 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5278 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5279 !spec_ik->has_injected_fields()) {
5280 if (!obj_type->isa_instptr() ||
5281 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5282 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5283 }
5284 }
5285 }
5286
5287 // Conservatively insert a memory barrier on all memory slices.
5288 // Do not let writes into the original float below the clone.
5289 insert_mem_bar(Op_MemBarCPUOrder);
5290
5291 // paths into result_reg:
5292 enum {
5293 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5294 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5295 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5296 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5297 PATH_LIMIT
5298 };
5299 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5300 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5301 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5302 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5303 record_for_igvn(result_reg);
5304
5305 Node* obj_klass = load_object_klass(obj);
5306 Node* array_obj = obj;
5307 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5308 if (array_ctl != nullptr) {
5309 // It's an array.
5310 PreserveJVMState pjvms(this);
5311 set_control(array_ctl);
5312 Node* obj_length = load_array_length(array_obj);
5313 Node* array_size = nullptr; // Size of the array without object alignment padding.
5314 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5315
5316 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5317 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5318 // If it is an oop array, it requires very special treatment,
5319 // because gc barriers are required when accessing the array.
5320 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5321 if (is_obja != nullptr) {
5322 PreserveJVMState pjvms2(this);
5323 set_control(is_obja);
5324 // Generate a direct call to the right arraycopy function(s).
5325 // Clones are always tightly coupled.
5326 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5327 ac->set_clone_oop_array();
5328 Node* n = _gvn.transform(ac);
5329 assert(n == ac, "cannot disappear");
5330 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5331
5332 result_reg->init_req(_objArray_path, control());
5333 result_val->init_req(_objArray_path, alloc_obj);
5334 result_i_o ->set_req(_objArray_path, i_o());
5335 result_mem ->set_req(_objArray_path, reset_memory());
5336 }
5337 }
5338 // Otherwise, there are no barriers to worry about.
5339 // (We can dispense with card marks if we know the allocation
5340 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5341 // causes the non-eden paths to take compensating steps to
5342 // simulate a fresh allocation, so that no further
5343 // card marks are required in compiled code to initialize
5344 // the object.)
5345
5346 if (!stopped()) {
5347 copy_to_clone(array_obj, alloc_obj, array_size, true);
5348
5349 // Present the results of the copy.
5350 result_reg->init_req(_array_path, control());
5351 result_val->init_req(_array_path, alloc_obj);
5352 result_i_o ->set_req(_array_path, i_o());
5353 result_mem ->set_req(_array_path, reset_memory());
5354 }
5355 }
5356
5357 // We only go to the instance fast case code if we pass a number of guards.
5358 // The paths which do not pass are accumulated in the slow_region.
5359 RegionNode* slow_region = new RegionNode(1);
5360 record_for_igvn(slow_region);
5361 if (!stopped()) {
5362 // It's an instance (we did array above). Make the slow-path tests.
5363 // If this is a virtual call, we generate a funny guard. We grab
5364 // the vtable entry corresponding to clone() from the target object.
5365 // If the target method which we are calling happens to be the
5366 // Object clone() method, we pass the guard. We do not need this
5367 // guard for non-virtual calls; the caller is known to be the native
5368 // Object clone().
5369 if (is_virtual) {
5370 generate_virtual_guard(obj_klass, slow_region);
5371 }
5372
5373 // The object must be easily cloneable and must not have a finalizer.
5374 // Both of these conditions may be checked in a single test.
5375 // We could optimize the test further, but we don't care.
5376 generate_misc_flags_guard(obj_klass,
5377 // Test both conditions:
5378 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5379 // Must be cloneable but not finalizer:
5380 KlassFlags::_misc_is_cloneable_fast,
5381 slow_region);
5382 }
5383
5384 if (!stopped()) {
5385 // It's an instance, and it passed the slow-path tests.
5386 PreserveJVMState pjvms(this);
5387 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5388 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5389 // is reexecuted if deoptimization occurs and there could be problems when merging
5390 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5391 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5392
5393 copy_to_clone(obj, alloc_obj, obj_size, false);
5394
5395 // Present the results of the slow call.
5396 result_reg->init_req(_instance_path, control());
5397 result_val->init_req(_instance_path, alloc_obj);
5398 result_i_o ->set_req(_instance_path, i_o());
5399 result_mem ->set_req(_instance_path, reset_memory());
5400 }
5401
5402 // Generate code for the slow case. We make a call to clone().
5403 set_control(_gvn.transform(slow_region));
5404 if (!stopped()) {
5405 PreserveJVMState pjvms(this);
5406 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5407 // We need to deoptimize on exception (see comment above)
5408 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5409 // this->control() comes from set_results_for_java_call
5410 result_reg->init_req(_slow_path, control());
5411 result_val->init_req(_slow_path, slow_result);
5412 result_i_o ->set_req(_slow_path, i_o());
5413 result_mem ->set_req(_slow_path, reset_memory());
5414 }
5415
5416 // Return the combined state.
5417 set_control( _gvn.transform(result_reg));
5418 set_i_o( _gvn.transform(result_i_o));
5419 set_all_memory( _gvn.transform(result_mem));
5420 } // original reexecute is set back here
5421
5422 set_result(_gvn.transform(result_val));
5423 return true;
5424 }
5425
5426 // If we have a tightly coupled allocation, the arraycopy may take care
5427 // of the array initialization. If one of the guards we insert between
5428 // the allocation and the arraycopy causes a deoptimization, an
5429 // uninitialized array will escape the compiled method. To prevent that
5430 // we set the JVM state for uncommon traps between the allocation and
5431 // the arraycopy to the state before the allocation so, in case of
5432 // deoptimization, we'll reexecute the allocation and the
5433 // initialization.
5434 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5435 if (alloc != nullptr) {
5436 ciMethod* trap_method = alloc->jvms()->method();
5437 int trap_bci = alloc->jvms()->bci();
5438
5439 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5440 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5441 // Make sure there's no store between the allocation and the
5442 // arraycopy otherwise visible side effects could be rexecuted
5443 // in case of deoptimization and cause incorrect execution.
5444 bool no_interfering_store = true;
5445 Node* mem = alloc->in(TypeFunc::Memory);
5446 if (mem->is_MergeMem()) {
5447 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5448 Node* n = mms.memory();
5449 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5450 assert(n->is_Store(), "what else?");
5451 no_interfering_store = false;
5452 break;
5453 }
5454 }
5455 } else {
5456 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5457 Node* n = mms.memory();
5458 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5459 assert(n->is_Store(), "what else?");
5460 no_interfering_store = false;
5461 break;
5462 }
5463 }
5464 }
5465
5466 if (no_interfering_store) {
5467 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5468
5469 JVMState* saved_jvms = jvms();
5470 saved_reexecute_sp = _reexecute_sp;
5471
5472 set_jvms(sfpt->jvms());
5473 _reexecute_sp = jvms()->sp();
5474
5475 return saved_jvms;
5476 }
5477 }
5478 }
5479 return nullptr;
5480 }
5481
5482 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5483 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5484 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5485 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5486 uint size = alloc->req();
5487 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5488 old_jvms->set_map(sfpt);
5489 for (uint i = 0; i < size; i++) {
5490 sfpt->init_req(i, alloc->in(i));
5491 }
5492 // re-push array length for deoptimization
5493 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5494 old_jvms->set_sp(old_jvms->sp()+1);
5495 old_jvms->set_monoff(old_jvms->monoff()+1);
5496 old_jvms->set_scloff(old_jvms->scloff()+1);
5497 old_jvms->set_endoff(old_jvms->endoff()+1);
5498 old_jvms->set_should_reexecute(true);
5499
5500 sfpt->set_i_o(map()->i_o());
5501 sfpt->set_memory(map()->memory());
5502 sfpt->set_control(map()->control());
5503 return sfpt;
5504 }
5505
5506 // In case of a deoptimization, we restart execution at the
5507 // allocation, allocating a new array. We would leave an uninitialized
5508 // array in the heap that GCs wouldn't expect. Move the allocation
5509 // after the traps so we don't allocate the array if we
5510 // deoptimize. This is possible because tightly_coupled_allocation()
5511 // guarantees there's no observer of the allocated array at this point
5512 // and the control flow is simple enough.
5513 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5514 int saved_reexecute_sp, uint new_idx) {
5515 if (saved_jvms_before_guards != nullptr && !stopped()) {
5516 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5517
5518 assert(alloc != nullptr, "only with a tightly coupled allocation");
5519 // restore JVM state to the state at the arraycopy
5520 saved_jvms_before_guards->map()->set_control(map()->control());
5521 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5522 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5523 // If we've improved the types of some nodes (null check) while
5524 // emitting the guards, propagate them to the current state
5525 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5526 set_jvms(saved_jvms_before_guards);
5527 _reexecute_sp = saved_reexecute_sp;
5528
5529 // Remove the allocation from above the guards
5530 CallProjections callprojs;
5531 alloc->extract_projections(&callprojs, true);
5532 InitializeNode* init = alloc->initialization();
5533 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5534 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5535 init->replace_mem_projs_by(alloc_mem, C);
5536
5537 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5538 // the allocation (i.e. is only valid if the allocation succeeds):
5539 // 1) replace CastIINode with AllocateArrayNode's length here
5540 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5541 //
5542 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5543 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5544 Node* init_control = init->proj_out(TypeFunc::Control);
5545 Node* alloc_length = alloc->Ideal_length();
5546 #ifdef ASSERT
5547 Node* prev_cast = nullptr;
5548 #endif
5549 for (uint i = 0; i < init_control->outcnt(); i++) {
5550 Node* init_out = init_control->raw_out(i);
5551 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5552 #ifdef ASSERT
5553 if (prev_cast == nullptr) {
5554 prev_cast = init_out;
5555 } else {
5556 if (prev_cast->cmp(*init_out) == false) {
5557 prev_cast->dump();
5558 init_out->dump();
5559 assert(false, "not equal CastIINode");
5560 }
5561 }
5562 #endif
5563 C->gvn_replace_by(init_out, alloc_length);
5564 }
5565 }
5566 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5567
5568 // move the allocation here (after the guards)
5569 _gvn.hash_delete(alloc);
5570 alloc->set_req(TypeFunc::Control, control());
5571 alloc->set_req(TypeFunc::I_O, i_o());
5572 Node *mem = reset_memory();
5573 set_all_memory(mem);
5574 alloc->set_req(TypeFunc::Memory, mem);
5575 set_control(init->proj_out_or_null(TypeFunc::Control));
5576 set_i_o(callprojs.fallthrough_ioproj);
5577
5578 // Update memory as done in GraphKit::set_output_for_allocation()
5579 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5580 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5581 if (ary_type->isa_aryptr() && length_type != nullptr) {
5582 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5583 }
5584 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5585 int elemidx = C->get_alias_index(telemref);
5586 // Need to properly move every memory projection for the Initialize
5587 #ifdef ASSERT
5588 int mark_idx = C->get_alias_index(ary_type->add_offset(oopDesc::mark_offset_in_bytes()));
5589 int klass_idx = C->get_alias_index(ary_type->add_offset(oopDesc::klass_offset_in_bytes()));
5590 #endif
5591 auto move_proj = [&](ProjNode* proj) {
5592 int alias_idx = C->get_alias_index(proj->adr_type());
5593 assert(alias_idx == Compile::AliasIdxRaw ||
5594 alias_idx == elemidx ||
5595 alias_idx == mark_idx ||
5596 alias_idx == klass_idx, "should be raw memory or array element type");
5597 set_memory(proj, alias_idx);
5598 };
5599 init->for_each_proj(move_proj, TypeFunc::Memory);
5600
5601 Node* allocx = _gvn.transform(alloc);
5602 assert(allocx == alloc, "where has the allocation gone?");
5603 assert(dest->is_CheckCastPP(), "not an allocation result?");
5604
5605 _gvn.hash_delete(dest);
5606 dest->set_req(0, control());
5607 Node* destx = _gvn.transform(dest);
5608 assert(destx == dest, "where has the allocation result gone?");
5609
5610 array_ideal_length(alloc, ary_type, true);
5611 }
5612 }
5613
5614 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5615 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5616 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5617 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5618 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5619 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5620 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5621 JVMState* saved_jvms_before_guards) {
5622 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5623 // There is at least one unrelated uncommon trap which needs to be replaced.
5624 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5625
5626 JVMState* saved_jvms = jvms();
5627 const int saved_reexecute_sp = _reexecute_sp;
5628 set_jvms(sfpt->jvms());
5629 _reexecute_sp = jvms()->sp();
5630
5631 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5632
5633 // Restore state
5634 set_jvms(saved_jvms);
5635 _reexecute_sp = saved_reexecute_sp;
5636 }
5637 }
5638
5639 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5640 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5641 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5642 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5643 while (if_proj->is_IfProj()) {
5644 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5645 if (uncommon_trap != nullptr) {
5646 create_new_uncommon_trap(uncommon_trap);
5647 }
5648 assert(if_proj->in(0)->is_If(), "must be If");
5649 if_proj = if_proj->in(0)->in(0);
5650 }
5651 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5652 "must have reached control projection of init node");
5653 }
5654
5655 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5656 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5657 assert(trap_request != 0, "no valid UCT trap request");
5658 PreserveJVMState pjvms(this);
5659 set_control(uncommon_trap_call->in(0));
5660 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5661 Deoptimization::trap_request_action(trap_request));
5662 assert(stopped(), "Should be stopped");
5663 _gvn.hash_delete(uncommon_trap_call);
5664 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5665 }
5666
5667 // Common checks for array sorting intrinsics arguments.
5668 // Returns `true` if checks passed.
5669 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5670 // check address of the class
5671 if (elementType == nullptr || elementType->is_top()) {
5672 return false; // dead path
5673 }
5674 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5675 if (elem_klass == nullptr) {
5676 return false; // dead path
5677 }
5678 // java_mirror_type() returns non-null for compile-time Class constants only
5679 ciType* elem_type = elem_klass->java_mirror_type();
5680 if (elem_type == nullptr) {
5681 return false;
5682 }
5683 bt = elem_type->basic_type();
5684 // Disable the intrinsic if the CPU does not support SIMD sort
5685 if (!Matcher::supports_simd_sort(bt)) {
5686 return false;
5687 }
5688 // check address of the array
5689 if (obj == nullptr || obj->is_top()) {
5690 return false; // dead path
5691 }
5692 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5693 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5694 return false; // failed input validation
5695 }
5696 return true;
5697 }
5698
5699 //------------------------------inline_array_partition-----------------------
5700 bool LibraryCallKit::inline_array_partition() {
5701 address stubAddr = StubRoutines::select_array_partition_function();
5702 if (stubAddr == nullptr) {
5703 return false; // Intrinsic's stub is not implemented on this platform
5704 }
5705 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5706
5707 // no receiver because it is a static method
5708 Node* elementType = argument(0);
5709 Node* obj = argument(1);
5710 Node* offset = argument(2); // long
5711 Node* fromIndex = argument(4);
5712 Node* toIndex = argument(5);
5713 Node* indexPivot1 = argument(6);
5714 Node* indexPivot2 = argument(7);
5715 // PartitionOperation: argument(8) is ignored
5716
5717 Node* pivotIndices = nullptr;
5718 BasicType bt = T_ILLEGAL;
5719
5720 if (!check_array_sort_arguments(elementType, obj, bt)) {
5721 return false;
5722 }
5723 null_check(obj);
5724 // If obj is dead, only null-path is taken.
5725 if (stopped()) {
5726 return true;
5727 }
5728 // Set the original stack and the reexecute bit for the interpreter to reexecute
5729 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5730 { PreserveReexecuteState preexecs(this);
5731 jvms()->set_should_reexecute(true);
5732
5733 Node* obj_adr = make_unsafe_address(obj, offset);
5734
5735 // create the pivotIndices array of type int and size = 2
5736 Node* size = intcon(2);
5737 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5738 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5739 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5740 guarantee(alloc != nullptr, "created above");
5741 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5742
5743 // pass the basic type enum to the stub
5744 Node* elemType = intcon(bt);
5745
5746 // Call the stub
5747 const char *stubName = "array_partition_stub";
5748 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5749 stubAddr, stubName, TypePtr::BOTTOM,
5750 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5751 indexPivot1, indexPivot2);
5752
5753 } // original reexecute is set back here
5754
5755 if (!stopped()) {
5756 set_result(pivotIndices);
5757 }
5758
5759 return true;
5760 }
5761
5762
5763 //------------------------------inline_array_sort-----------------------
5764 bool LibraryCallKit::inline_array_sort() {
5765 address stubAddr = StubRoutines::select_arraysort_function();
5766 if (stubAddr == nullptr) {
5767 return false; // Intrinsic's stub is not implemented on this platform
5768 }
5769 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5770
5771 // no receiver because it is a static method
5772 Node* elementType = argument(0);
5773 Node* obj = argument(1);
5774 Node* offset = argument(2); // long
5775 Node* fromIndex = argument(4);
5776 Node* toIndex = argument(5);
5777 // SortOperation: argument(6) is ignored
5778
5779 BasicType bt = T_ILLEGAL;
5780
5781 if (!check_array_sort_arguments(elementType, obj, bt)) {
5782 return false;
5783 }
5784 null_check(obj);
5785 // If obj is dead, only null-path is taken.
5786 if (stopped()) {
5787 return true;
5788 }
5789 Node* obj_adr = make_unsafe_address(obj, offset);
5790
5791 // pass the basic type enum to the stub
5792 Node* elemType = intcon(bt);
5793
5794 // Call the stub.
5795 const char *stubName = "arraysort_stub";
5796 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5797 stubAddr, stubName, TypePtr::BOTTOM,
5798 obj_adr, elemType, fromIndex, toIndex);
5799
5800 return true;
5801 }
5802
5803
5804 //------------------------------inline_arraycopy-----------------------
5805 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5806 // Object dest, int destPos,
5807 // int length);
5808 bool LibraryCallKit::inline_arraycopy() {
5809 // Get the arguments.
5810 Node* src = argument(0); // type: oop
5811 Node* src_offset = argument(1); // type: int
5812 Node* dest = argument(2); // type: oop
5813 Node* dest_offset = argument(3); // type: int
5814 Node* length = argument(4); // type: int
5815
5816 uint new_idx = C->unique();
5817
5818 // Check for allocation before we add nodes that would confuse
5819 // tightly_coupled_allocation()
5820 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5821
5822 int saved_reexecute_sp = -1;
5823 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5824 // See arraycopy_restore_alloc_state() comment
5825 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5826 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5827 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5828 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5829
5830 // The following tests must be performed
5831 // (1) src and dest are arrays.
5832 // (2) src and dest arrays must have elements of the same BasicType
5833 // (3) src and dest must not be null.
5834 // (4) src_offset must not be negative.
5835 // (5) dest_offset must not be negative.
5836 // (6) length must not be negative.
5837 // (7) src_offset + length must not exceed length of src.
5838 // (8) dest_offset + length must not exceed length of dest.
5839 // (9) each element of an oop array must be assignable
5840
5841 // (3) src and dest must not be null.
5842 // always do this here because we need the JVM state for uncommon traps
5843 Node* null_ctl = top();
5844 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5845 assert(null_ctl->is_top(), "no null control here");
5846 dest = null_check(dest, T_ARRAY);
5847
5848 if (!can_emit_guards) {
5849 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5850 // guards but the arraycopy node could still take advantage of a
5851 // tightly allocated allocation. tightly_coupled_allocation() is
5852 // called again to make sure it takes the null check above into
5853 // account: the null check is mandatory and if it caused an
5854 // uncommon trap to be emitted then the allocation can't be
5855 // considered tightly coupled in this context.
5856 alloc = tightly_coupled_allocation(dest);
5857 }
5858
5859 bool validated = false;
5860
5861 const Type* src_type = _gvn.type(src);
5862 const Type* dest_type = _gvn.type(dest);
5863 const TypeAryPtr* top_src = src_type->isa_aryptr();
5864 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5865
5866 // Do we have the type of src?
5867 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5868 // Do we have the type of dest?
5869 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5870 // Is the type for src from speculation?
5871 bool src_spec = false;
5872 // Is the type for dest from speculation?
5873 bool dest_spec = false;
5874
5875 if ((!has_src || !has_dest) && can_emit_guards) {
5876 // We don't have sufficient type information, let's see if
5877 // speculative types can help. We need to have types for both src
5878 // and dest so that it pays off.
5879
5880 // Do we already have or could we have type information for src
5881 bool could_have_src = has_src;
5882 // Do we already have or could we have type information for dest
5883 bool could_have_dest = has_dest;
5884
5885 ciKlass* src_k = nullptr;
5886 if (!has_src) {
5887 src_k = src_type->speculative_type_not_null();
5888 if (src_k != nullptr && src_k->is_array_klass()) {
5889 could_have_src = true;
5890 }
5891 }
5892
5893 ciKlass* dest_k = nullptr;
5894 if (!has_dest) {
5895 dest_k = dest_type->speculative_type_not_null();
5896 if (dest_k != nullptr && dest_k->is_array_klass()) {
5897 could_have_dest = true;
5898 }
5899 }
5900
5901 if (could_have_src && could_have_dest) {
5902 // This is going to pay off so emit the required guards
5903 if (!has_src) {
5904 src = maybe_cast_profiled_obj(src, src_k, true);
5905 src_type = _gvn.type(src);
5906 top_src = src_type->isa_aryptr();
5907 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5908 src_spec = true;
5909 }
5910 if (!has_dest) {
5911 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5912 dest_type = _gvn.type(dest);
5913 top_dest = dest_type->isa_aryptr();
5914 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5915 dest_spec = true;
5916 }
5917 }
5918 }
5919
5920 if (has_src && has_dest && can_emit_guards) {
5921 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
5922 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
5923 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
5924 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
5925
5926 if (src_elem == dest_elem && src_elem == T_OBJECT) {
5927 // If both arrays are object arrays then having the exact types
5928 // for both will remove the need for a subtype check at runtime
5929 // before the call and may make it possible to pick a faster copy
5930 // routine (without a subtype check on every element)
5931 // Do we have the exact type of src?
5932 bool could_have_src = src_spec;
5933 // Do we have the exact type of dest?
5934 bool could_have_dest = dest_spec;
5935 ciKlass* src_k = nullptr;
5936 ciKlass* dest_k = nullptr;
5937 if (!src_spec) {
5938 src_k = src_type->speculative_type_not_null();
5939 if (src_k != nullptr && src_k->is_array_klass()) {
5940 could_have_src = true;
5941 }
5942 }
5943 if (!dest_spec) {
5944 dest_k = dest_type->speculative_type_not_null();
5945 if (dest_k != nullptr && dest_k->is_array_klass()) {
5946 could_have_dest = true;
5947 }
5948 }
5949 if (could_have_src && could_have_dest) {
5950 // If we can have both exact types, emit the missing guards
5951 if (could_have_src && !src_spec) {
5952 src = maybe_cast_profiled_obj(src, src_k, true);
5953 }
5954 if (could_have_dest && !dest_spec) {
5955 dest = maybe_cast_profiled_obj(dest, dest_k, true);
5956 }
5957 }
5958 }
5959 }
5960
5961 ciMethod* trap_method = method();
5962 int trap_bci = bci();
5963 if (saved_jvms_before_guards != nullptr) {
5964 trap_method = alloc->jvms()->method();
5965 trap_bci = alloc->jvms()->bci();
5966 }
5967
5968 bool negative_length_guard_generated = false;
5969
5970 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5971 can_emit_guards &&
5972 !src->is_top() && !dest->is_top()) {
5973 // validate arguments: enables transformation the ArrayCopyNode
5974 validated = true;
5975
5976 RegionNode* slow_region = new RegionNode(1);
5977 record_for_igvn(slow_region);
5978
5979 // (1) src and dest are arrays.
5980 generate_non_array_guard(load_object_klass(src), slow_region, &src);
5981 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
5982
5983 // (2) src and dest arrays must have elements of the same BasicType
5984 // done at macro expansion or at Ideal transformation time
5985
5986 // (4) src_offset must not be negative.
5987 generate_negative_guard(src_offset, slow_region);
5988
5989 // (5) dest_offset must not be negative.
5990 generate_negative_guard(dest_offset, slow_region);
5991
5992 // (7) src_offset + length must not exceed length of src.
5993 generate_limit_guard(src_offset, length,
5994 load_array_length(src),
5995 slow_region);
5996
5997 // (8) dest_offset + length must not exceed length of dest.
5998 generate_limit_guard(dest_offset, length,
5999 load_array_length(dest),
6000 slow_region);
6001
6002 // (6) length must not be negative.
6003 // This is also checked in generate_arraycopy() during macro expansion, but
6004 // we also have to check it here for the case where the ArrayCopyNode will
6005 // be eliminated by Escape Analysis.
6006 if (EliminateAllocations) {
6007 generate_negative_guard(length, slow_region);
6008 negative_length_guard_generated = true;
6009 }
6010
6011 // (9) each element of an oop array must be assignable
6012 Node* dest_klass = load_object_klass(dest);
6013 if (src != dest) {
6014 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6015
6016 if (not_subtype_ctrl != top()) {
6017 PreserveJVMState pjvms(this);
6018 set_control(not_subtype_ctrl);
6019 uncommon_trap(Deoptimization::Reason_intrinsic,
6020 Deoptimization::Action_make_not_entrant);
6021 assert(stopped(), "Should be stopped");
6022 }
6023 }
6024 {
6025 PreserveJVMState pjvms(this);
6026 set_control(_gvn.transform(slow_region));
6027 uncommon_trap(Deoptimization::Reason_intrinsic,
6028 Deoptimization::Action_make_not_entrant);
6029 assert(stopped(), "Should be stopped");
6030 }
6031
6032 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6033 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6034 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6035 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6036 }
6037
6038 if (stopped()) {
6039 return true;
6040 }
6041
6042 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6043 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6044 // so the compiler has a chance to eliminate them: during macro expansion,
6045 // we have to set their control (CastPP nodes are eliminated).
6046 load_object_klass(src), load_object_klass(dest),
6047 load_array_length(src), load_array_length(dest));
6048
6049 ac->set_arraycopy(validated);
6050
6051 Node* n = _gvn.transform(ac);
6052 if (n == ac) {
6053 ac->connect_outputs(this);
6054 } else {
6055 assert(validated, "shouldn't transform if all arguments not validated");
6056 set_all_memory(n);
6057 }
6058 clear_upper_avx();
6059
6060
6061 return true;
6062 }
6063
6064
6065 // Helper function which determines if an arraycopy immediately follows
6066 // an allocation, with no intervening tests or other escapes for the object.
6067 AllocateArrayNode*
6068 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6069 if (stopped()) return nullptr; // no fast path
6070 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6071
6072 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6073 if (alloc == nullptr) return nullptr;
6074
6075 Node* rawmem = memory(Compile::AliasIdxRaw);
6076 // Is the allocation's memory state untouched?
6077 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6078 // Bail out if there have been raw-memory effects since the allocation.
6079 // (Example: There might have been a call or safepoint.)
6080 return nullptr;
6081 }
6082 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6083 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6084 return nullptr;
6085 }
6086
6087 // There must be no unexpected observers of this allocation.
6088 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6089 Node* obs = ptr->fast_out(i);
6090 if (obs != this->map()) {
6091 return nullptr;
6092 }
6093 }
6094
6095 // This arraycopy must unconditionally follow the allocation of the ptr.
6096 Node* alloc_ctl = ptr->in(0);
6097 Node* ctl = control();
6098 while (ctl != alloc_ctl) {
6099 // There may be guards which feed into the slow_region.
6100 // Any other control flow means that we might not get a chance
6101 // to finish initializing the allocated object.
6102 // Various low-level checks bottom out in uncommon traps. These
6103 // are considered safe since we've already checked above that
6104 // there is no unexpected observer of this allocation.
6105 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6106 assert(ctl->in(0)->is_If(), "must be If");
6107 ctl = ctl->in(0)->in(0);
6108 } else {
6109 return nullptr;
6110 }
6111 }
6112
6113 // If we get this far, we have an allocation which immediately
6114 // precedes the arraycopy, and we can take over zeroing the new object.
6115 // The arraycopy will finish the initialization, and provide
6116 // a new control state to which we will anchor the destination pointer.
6117
6118 return alloc;
6119 }
6120
6121 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6122 if (node->is_IfProj()) {
6123 IfProjNode* other_proj = node->as_IfProj()->other_if_proj();
6124 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6125 Node* obs = other_proj->fast_out(j);
6126 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6127 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6128 return obs->as_CallStaticJava();
6129 }
6130 }
6131 }
6132 return nullptr;
6133 }
6134
6135 //-------------inline_encodeISOArray-----------------------------------
6136 // int sun.nio.cs.ISO_8859_1.Encoder#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6137 // int java.lang.StringCoding#encodeISOArray0(byte[] sa, int sp, byte[] da, int dp, int len)
6138 // int java.lang.StringCoding#encodeAsciiArray0(char[] sa, int sp, byte[] da, int dp, int len)
6139 // encode char[] to byte[] in ISO_8859_1 or ASCII
6140 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6141 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6142 // no receiver since it is static method
6143 Node *src = argument(0);
6144 Node *src_offset = argument(1);
6145 Node *dst = argument(2);
6146 Node *dst_offset = argument(3);
6147 Node *length = argument(4);
6148
6149 // Cast source & target arrays to not-null
6150 src = must_be_not_null(src, true);
6151 dst = must_be_not_null(dst, true);
6152 if (stopped()) {
6153 return true;
6154 }
6155
6156 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6157 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6158 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6159 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6160 // failed array check
6161 return false;
6162 }
6163
6164 // Figure out the size and type of the elements we will be copying.
6165 BasicType src_elem = src_type->elem()->array_element_basic_type();
6166 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6167 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6168 return false;
6169 }
6170
6171 // Check source & target bounds
6172 RegionNode* bailout = create_bailout();
6173 generate_string_range_check(src, src_offset, length, src_elem == T_BYTE, bailout);
6174 generate_string_range_check(dst, dst_offset, length, false, bailout);
6175 if (check_bailout(bailout)) {
6176 return true;
6177 }
6178
6179 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6180 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6181 // 'src_start' points to src array + scaled offset
6182 // 'dst_start' points to dst array + scaled offset
6183
6184 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6185 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6186 enc = _gvn.transform(enc);
6187 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6188 set_memory(res_mem, mtype);
6189 set_result(enc);
6190 clear_upper_avx();
6191
6192 return true;
6193 }
6194
6195 //-------------inline_multiplyToLen-----------------------------------
6196 bool LibraryCallKit::inline_multiplyToLen() {
6197 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6198
6199 address stubAddr = StubRoutines::multiplyToLen();
6200 if (stubAddr == nullptr) {
6201 return false; // Intrinsic's stub is not implemented on this platform
6202 }
6203 const char* stubName = "multiplyToLen";
6204
6205 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6206
6207 // no receiver because it is a static method
6208 Node* x = argument(0);
6209 Node* xlen = argument(1);
6210 Node* y = argument(2);
6211 Node* ylen = argument(3);
6212 Node* z = argument(4);
6213
6214 x = must_be_not_null(x, true);
6215 y = must_be_not_null(y, true);
6216
6217 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6218 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6219 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6220 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6221 // failed array check
6222 return false;
6223 }
6224
6225 BasicType x_elem = x_type->elem()->array_element_basic_type();
6226 BasicType y_elem = y_type->elem()->array_element_basic_type();
6227 if (x_elem != T_INT || y_elem != T_INT) {
6228 return false;
6229 }
6230
6231 Node* x_start = array_element_address(x, intcon(0), x_elem);
6232 Node* y_start = array_element_address(y, intcon(0), y_elem);
6233 // 'x_start' points to x array + scaled xlen
6234 // 'y_start' points to y array + scaled ylen
6235
6236 Node* z_start = array_element_address(z, intcon(0), T_INT);
6237
6238 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6239 OptoRuntime::multiplyToLen_Type(),
6240 stubAddr, stubName, TypePtr::BOTTOM,
6241 x_start, xlen, y_start, ylen, z_start);
6242
6243 C->set_has_split_ifs(true); // Has chance for split-if optimization
6244 set_result(z);
6245 return true;
6246 }
6247
6248 //-------------inline_squareToLen------------------------------------
6249 bool LibraryCallKit::inline_squareToLen() {
6250 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6251
6252 address stubAddr = StubRoutines::squareToLen();
6253 if (stubAddr == nullptr) {
6254 return false; // Intrinsic's stub is not implemented on this platform
6255 }
6256 const char* stubName = "squareToLen";
6257
6258 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6259
6260 Node* x = argument(0);
6261 Node* len = argument(1);
6262 Node* z = argument(2);
6263 Node* zlen = argument(3);
6264
6265 x = must_be_not_null(x, true);
6266 z = must_be_not_null(z, true);
6267
6268 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6269 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6270 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6271 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6272 // failed array check
6273 return false;
6274 }
6275
6276 BasicType x_elem = x_type->elem()->array_element_basic_type();
6277 BasicType z_elem = z_type->elem()->array_element_basic_type();
6278 if (x_elem != T_INT || z_elem != T_INT) {
6279 return false;
6280 }
6281
6282
6283 Node* x_start = array_element_address(x, intcon(0), x_elem);
6284 Node* z_start = array_element_address(z, intcon(0), z_elem);
6285
6286 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6287 OptoRuntime::squareToLen_Type(),
6288 stubAddr, stubName, TypePtr::BOTTOM,
6289 x_start, len, z_start, zlen);
6290
6291 set_result(z);
6292 return true;
6293 }
6294
6295 //-------------inline_mulAdd------------------------------------------
6296 bool LibraryCallKit::inline_mulAdd() {
6297 assert(UseMulAddIntrinsic, "not implemented on this platform");
6298
6299 address stubAddr = StubRoutines::mulAdd();
6300 if (stubAddr == nullptr) {
6301 return false; // Intrinsic's stub is not implemented on this platform
6302 }
6303 const char* stubName = "mulAdd";
6304
6305 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6306
6307 Node* out = argument(0);
6308 Node* in = argument(1);
6309 Node* offset = argument(2);
6310 Node* len = argument(3);
6311 Node* k = argument(4);
6312
6313 in = must_be_not_null(in, true);
6314 out = must_be_not_null(out, true);
6315
6316 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6317 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6318 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6319 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6320 // failed array check
6321 return false;
6322 }
6323
6324 BasicType out_elem = out_type->elem()->array_element_basic_type();
6325 BasicType in_elem = in_type->elem()->array_element_basic_type();
6326 if (out_elem != T_INT || in_elem != T_INT) {
6327 return false;
6328 }
6329
6330 Node* outlen = load_array_length(out);
6331 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6332 Node* out_start = array_element_address(out, intcon(0), out_elem);
6333 Node* in_start = array_element_address(in, intcon(0), in_elem);
6334
6335 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6336 OptoRuntime::mulAdd_Type(),
6337 stubAddr, stubName, TypePtr::BOTTOM,
6338 out_start,in_start, new_offset, len, k);
6339 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6340 set_result(result);
6341 return true;
6342 }
6343
6344 //-------------inline_montgomeryMultiply-----------------------------------
6345 bool LibraryCallKit::inline_montgomeryMultiply() {
6346 address stubAddr = StubRoutines::montgomeryMultiply();
6347 if (stubAddr == nullptr) {
6348 return false; // Intrinsic's stub is not implemented on this platform
6349 }
6350
6351 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6352 const char* stubName = "montgomery_multiply";
6353
6354 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6355
6356 Node* a = argument(0);
6357 Node* b = argument(1);
6358 Node* n = argument(2);
6359 Node* len = argument(3);
6360 Node* inv = argument(4);
6361 Node* m = argument(6);
6362
6363 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6364 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6365 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6366 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6367 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6368 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6369 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6370 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6371 // failed array check
6372 return false;
6373 }
6374
6375 BasicType a_elem = a_type->elem()->array_element_basic_type();
6376 BasicType b_elem = b_type->elem()->array_element_basic_type();
6377 BasicType n_elem = n_type->elem()->array_element_basic_type();
6378 BasicType m_elem = m_type->elem()->array_element_basic_type();
6379 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6380 return false;
6381 }
6382
6383 // Make the call
6384 {
6385 Node* a_start = array_element_address(a, intcon(0), a_elem);
6386 Node* b_start = array_element_address(b, intcon(0), b_elem);
6387 Node* n_start = array_element_address(n, intcon(0), n_elem);
6388 Node* m_start = array_element_address(m, intcon(0), m_elem);
6389
6390 Node* call = make_runtime_call(RC_LEAF,
6391 OptoRuntime::montgomeryMultiply_Type(),
6392 stubAddr, stubName, TypePtr::BOTTOM,
6393 a_start, b_start, n_start, len, inv, top(),
6394 m_start);
6395 set_result(m);
6396 }
6397
6398 return true;
6399 }
6400
6401 bool LibraryCallKit::inline_montgomerySquare() {
6402 address stubAddr = StubRoutines::montgomerySquare();
6403 if (stubAddr == nullptr) {
6404 return false; // Intrinsic's stub is not implemented on this platform
6405 }
6406
6407 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6408 const char* stubName = "montgomery_square";
6409
6410 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6411
6412 Node* a = argument(0);
6413 Node* n = argument(1);
6414 Node* len = argument(2);
6415 Node* inv = argument(3);
6416 Node* m = argument(5);
6417
6418 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6419 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6420 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6421 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6422 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6423 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6424 // failed array check
6425 return false;
6426 }
6427
6428 BasicType a_elem = a_type->elem()->array_element_basic_type();
6429 BasicType n_elem = n_type->elem()->array_element_basic_type();
6430 BasicType m_elem = m_type->elem()->array_element_basic_type();
6431 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6432 return false;
6433 }
6434
6435 // Make the call
6436 {
6437 Node* a_start = array_element_address(a, intcon(0), a_elem);
6438 Node* n_start = array_element_address(n, intcon(0), n_elem);
6439 Node* m_start = array_element_address(m, intcon(0), m_elem);
6440
6441 Node* call = make_runtime_call(RC_LEAF,
6442 OptoRuntime::montgomerySquare_Type(),
6443 stubAddr, stubName, TypePtr::BOTTOM,
6444 a_start, n_start, len, inv, top(),
6445 m_start);
6446 set_result(m);
6447 }
6448
6449 return true;
6450 }
6451
6452 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6453 address stubAddr = nullptr;
6454 const char* stubName = nullptr;
6455
6456 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6457 if (stubAddr == nullptr) {
6458 return false; // Intrinsic's stub is not implemented on this platform
6459 }
6460
6461 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6462
6463 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6464
6465 Node* newArr = argument(0);
6466 Node* oldArr = argument(1);
6467 Node* newIdx = argument(2);
6468 Node* shiftCount = argument(3);
6469 Node* numIter = argument(4);
6470
6471 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6472 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6473 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6474 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6475 return false;
6476 }
6477
6478 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6479 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6480 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6481 return false;
6482 }
6483
6484 // Make the call
6485 {
6486 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6487 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6488
6489 Node* call = make_runtime_call(RC_LEAF,
6490 OptoRuntime::bigIntegerShift_Type(),
6491 stubAddr,
6492 stubName,
6493 TypePtr::BOTTOM,
6494 newArr_start,
6495 oldArr_start,
6496 newIdx,
6497 shiftCount,
6498 numIter);
6499 }
6500
6501 return true;
6502 }
6503
6504 //-------------inline_vectorizedMismatch------------------------------
6505 bool LibraryCallKit::inline_vectorizedMismatch() {
6506 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6507
6508 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6509 Node* obja = argument(0); // Object
6510 Node* aoffset = argument(1); // long
6511 Node* objb = argument(3); // Object
6512 Node* boffset = argument(4); // long
6513 Node* length = argument(6); // int
6514 Node* scale = argument(7); // int
6515
6516 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6517 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6518 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6519 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6520 scale == top()) {
6521 return false; // failed input validation
6522 }
6523
6524 Node* obja_adr = make_unsafe_address(obja, aoffset);
6525 Node* objb_adr = make_unsafe_address(objb, boffset);
6526
6527 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6528 //
6529 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6530 // if (length <= inline_limit) {
6531 // inline_path:
6532 // vmask = VectorMaskGen length
6533 // vload1 = LoadVectorMasked obja, vmask
6534 // vload2 = LoadVectorMasked objb, vmask
6535 // result1 = VectorCmpMasked vload1, vload2, vmask
6536 // } else {
6537 // call_stub_path:
6538 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6539 // }
6540 // exit_block:
6541 // return Phi(result1, result2);
6542 //
6543 enum { inline_path = 1, // input is small enough to process it all at once
6544 stub_path = 2, // input is too large; call into the VM
6545 PATH_LIMIT = 3
6546 };
6547
6548 Node* exit_block = new RegionNode(PATH_LIMIT);
6549 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6550 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6551
6552 Node* call_stub_path = control();
6553
6554 BasicType elem_bt = T_ILLEGAL;
6555
6556 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6557 if (scale_t->is_con()) {
6558 switch (scale_t->get_con()) {
6559 case 0: elem_bt = T_BYTE; break;
6560 case 1: elem_bt = T_SHORT; break;
6561 case 2: elem_bt = T_INT; break;
6562 case 3: elem_bt = T_LONG; break;
6563
6564 default: elem_bt = T_ILLEGAL; break; // not supported
6565 }
6566 }
6567
6568 int inline_limit = 0;
6569 bool do_partial_inline = false;
6570
6571 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6572 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6573 do_partial_inline = inline_limit >= 16;
6574 }
6575
6576 if (do_partial_inline) {
6577 assert(elem_bt != T_ILLEGAL, "sanity");
6578
6579 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6580 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6581 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6582
6583 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6584 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6585 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6586
6587 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6588
6589 if (!stopped()) {
6590 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6591
6592 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6593 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6594 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6595 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6596
6597 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6598 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6599 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6600 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6601
6602 exit_block->init_req(inline_path, control());
6603 memory_phi->init_req(inline_path, map()->memory());
6604 result_phi->init_req(inline_path, result);
6605
6606 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6607 clear_upper_avx();
6608 }
6609 }
6610 }
6611
6612 if (call_stub_path != nullptr) {
6613 set_control(call_stub_path);
6614
6615 Node* call = make_runtime_call(RC_LEAF,
6616 OptoRuntime::vectorizedMismatch_Type(),
6617 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6618 obja_adr, objb_adr, length, scale);
6619
6620 exit_block->init_req(stub_path, control());
6621 memory_phi->init_req(stub_path, map()->memory());
6622 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6623 }
6624
6625 exit_block = _gvn.transform(exit_block);
6626 memory_phi = _gvn.transform(memory_phi);
6627 result_phi = _gvn.transform(result_phi);
6628
6629 record_for_igvn(exit_block);
6630 record_for_igvn(memory_phi);
6631 record_for_igvn(result_phi);
6632
6633 set_control(exit_block);
6634 set_all_memory(memory_phi);
6635 set_result(result_phi);
6636
6637 return true;
6638 }
6639
6640 //------------------------------inline_vectorizedHashcode----------------------------
6641 bool LibraryCallKit::inline_vectorizedHashCode() {
6642 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6643
6644 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6645 Node* array = argument(0);
6646 Node* offset = argument(1);
6647 Node* length = argument(2);
6648 Node* initialValue = argument(3);
6649 Node* basic_type = argument(4);
6650
6651 if (basic_type == top()) {
6652 return false; // failed input validation
6653 }
6654
6655 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6656 if (!basic_type_t->is_con()) {
6657 return false; // Only intrinsify if mode argument is constant
6658 }
6659
6660 array = must_be_not_null(array, true);
6661
6662 BasicType bt = (BasicType)basic_type_t->get_con();
6663
6664 // Resolve address of first element
6665 Node* array_start = array_element_address(array, offset, bt);
6666
6667 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6668 array_start, length, initialValue, basic_type)));
6669 clear_upper_avx();
6670
6671 return true;
6672 }
6673
6674 /**
6675 * Calculate CRC32 for byte.
6676 * int java.util.zip.CRC32.update(int crc, int b)
6677 */
6678 bool LibraryCallKit::inline_updateCRC32() {
6679 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6680 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6681 // no receiver since it is static method
6682 Node* crc = argument(0); // type: int
6683 Node* b = argument(1); // type: int
6684
6685 /*
6686 * int c = ~ crc;
6687 * b = timesXtoThe32[(b ^ c) & 0xFF];
6688 * b = b ^ (c >>> 8);
6689 * crc = ~b;
6690 */
6691
6692 Node* M1 = intcon(-1);
6693 crc = _gvn.transform(new XorINode(crc, M1));
6694 Node* result = _gvn.transform(new XorINode(crc, b));
6695 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6696
6697 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6698 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6699 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6700 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6701
6702 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6703 result = _gvn.transform(new XorINode(crc, result));
6704 result = _gvn.transform(new XorINode(result, M1));
6705 set_result(result);
6706 return true;
6707 }
6708
6709 /**
6710 * Calculate CRC32 for byte[] array.
6711 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6712 */
6713 bool LibraryCallKit::inline_updateBytesCRC32() {
6714 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6715 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6716 // no receiver since it is static method
6717 Node* crc = argument(0); // type: int
6718 Node* src = argument(1); // type: oop
6719 Node* offset = argument(2); // type: int
6720 Node* length = argument(3); // type: int
6721
6722 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6723 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6724 // failed array check
6725 return false;
6726 }
6727
6728 // Figure out the size and type of the elements we will be copying.
6729 BasicType src_elem = src_type->elem()->array_element_basic_type();
6730 if (src_elem != T_BYTE) {
6731 return false;
6732 }
6733
6734 // 'src_start' points to src array + scaled offset
6735 src = must_be_not_null(src, true);
6736 Node* src_start = array_element_address(src, offset, src_elem);
6737
6738 // We assume that range check is done by caller.
6739 // TODO: generate range check (offset+length < src.length) in debug VM.
6740
6741 // Call the stub.
6742 address stubAddr = StubRoutines::updateBytesCRC32();
6743 const char *stubName = "updateBytesCRC32";
6744
6745 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6746 stubAddr, stubName, TypePtr::BOTTOM,
6747 crc, src_start, length);
6748 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6749 set_result(result);
6750 return true;
6751 }
6752
6753 /**
6754 * Calculate CRC32 for ByteBuffer.
6755 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6756 */
6757 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6758 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6759 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6760 // no receiver since it is static method
6761 Node* crc = argument(0); // type: int
6762 Node* src = argument(1); // type: long
6763 Node* offset = argument(3); // type: int
6764 Node* length = argument(4); // type: int
6765
6766 src = ConvL2X(src); // adjust Java long to machine word
6767 Node* base = _gvn.transform(new CastX2PNode(src));
6768 offset = ConvI2X(offset);
6769
6770 // 'src_start' points to src array + scaled offset
6771 Node* src_start = basic_plus_adr(top(), base, offset);
6772
6773 // Call the stub.
6774 address stubAddr = StubRoutines::updateBytesCRC32();
6775 const char *stubName = "updateBytesCRC32";
6776
6777 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6778 stubAddr, stubName, TypePtr::BOTTOM,
6779 crc, src_start, length);
6780 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6781 set_result(result);
6782 return true;
6783 }
6784
6785 //------------------------------get_table_from_crc32c_class-----------------------
6786 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6787 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6788 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6789
6790 return table;
6791 }
6792
6793 //------------------------------inline_updateBytesCRC32C-----------------------
6794 //
6795 // Calculate CRC32C for byte[] array.
6796 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6797 //
6798 bool LibraryCallKit::inline_updateBytesCRC32C() {
6799 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6800 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6801 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6802 // no receiver since it is a static method
6803 Node* crc = argument(0); // type: int
6804 Node* src = argument(1); // type: oop
6805 Node* offset = argument(2); // type: int
6806 Node* end = argument(3); // type: int
6807
6808 Node* length = _gvn.transform(new SubINode(end, offset));
6809
6810 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6811 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6812 // failed array check
6813 return false;
6814 }
6815
6816 // Figure out the size and type of the elements we will be copying.
6817 BasicType src_elem = src_type->elem()->array_element_basic_type();
6818 if (src_elem != T_BYTE) {
6819 return false;
6820 }
6821
6822 // 'src_start' points to src array + scaled offset
6823 src = must_be_not_null(src, true);
6824 Node* src_start = array_element_address(src, offset, src_elem);
6825
6826 // static final int[] byteTable in class CRC32C
6827 Node* table = get_table_from_crc32c_class(callee()->holder());
6828 table = must_be_not_null(table, true);
6829 Node* table_start = array_element_address(table, intcon(0), T_INT);
6830
6831 // We assume that range check is done by caller.
6832 // TODO: generate range check (offset+length < src.length) in debug VM.
6833
6834 // Call the stub.
6835 address stubAddr = StubRoutines::updateBytesCRC32C();
6836 const char *stubName = "updateBytesCRC32C";
6837
6838 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6839 stubAddr, stubName, TypePtr::BOTTOM,
6840 crc, src_start, length, table_start);
6841 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6842 set_result(result);
6843 return true;
6844 }
6845
6846 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6847 //
6848 // Calculate CRC32C for DirectByteBuffer.
6849 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6850 //
6851 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6852 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6853 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6854 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6855 // no receiver since it is a static method
6856 Node* crc = argument(0); // type: int
6857 Node* src = argument(1); // type: long
6858 Node* offset = argument(3); // type: int
6859 Node* end = argument(4); // type: int
6860
6861 Node* length = _gvn.transform(new SubINode(end, offset));
6862
6863 src = ConvL2X(src); // adjust Java long to machine word
6864 Node* base = _gvn.transform(new CastX2PNode(src));
6865 offset = ConvI2X(offset);
6866
6867 // 'src_start' points to src array + scaled offset
6868 Node* src_start = basic_plus_adr(top(), base, offset);
6869
6870 // static final int[] byteTable in class CRC32C
6871 Node* table = get_table_from_crc32c_class(callee()->holder());
6872 table = must_be_not_null(table, true);
6873 Node* table_start = array_element_address(table, intcon(0), T_INT);
6874
6875 // Call the stub.
6876 address stubAddr = StubRoutines::updateBytesCRC32C();
6877 const char *stubName = "updateBytesCRC32C";
6878
6879 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6880 stubAddr, stubName, TypePtr::BOTTOM,
6881 crc, src_start, length, table_start);
6882 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6883 set_result(result);
6884 return true;
6885 }
6886
6887 //------------------------------inline_updateBytesAdler32----------------------
6888 //
6889 // Calculate Adler32 checksum for byte[] array.
6890 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6891 //
6892 bool LibraryCallKit::inline_updateBytesAdler32() {
6893 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6894 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6895 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6896 // no receiver since it is static method
6897 Node* crc = argument(0); // type: int
6898 Node* src = argument(1); // type: oop
6899 Node* offset = argument(2); // type: int
6900 Node* length = argument(3); // type: int
6901
6902 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6903 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6904 // failed array check
6905 return false;
6906 }
6907
6908 // Figure out the size and type of the elements we will be copying.
6909 BasicType src_elem = src_type->elem()->array_element_basic_type();
6910 if (src_elem != T_BYTE) {
6911 return false;
6912 }
6913
6914 // 'src_start' points to src array + scaled offset
6915 Node* src_start = array_element_address(src, offset, src_elem);
6916
6917 // We assume that range check is done by caller.
6918 // TODO: generate range check (offset+length < src.length) in debug VM.
6919
6920 // Call the stub.
6921 address stubAddr = StubRoutines::updateBytesAdler32();
6922 const char *stubName = "updateBytesAdler32";
6923
6924 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6925 stubAddr, stubName, TypePtr::BOTTOM,
6926 crc, src_start, length);
6927 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6928 set_result(result);
6929 return true;
6930 }
6931
6932 //------------------------------inline_updateByteBufferAdler32---------------
6933 //
6934 // Calculate Adler32 checksum for DirectByteBuffer.
6935 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
6936 //
6937 bool LibraryCallKit::inline_updateByteBufferAdler32() {
6938 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6939 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6940 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6941 // no receiver since it is static method
6942 Node* crc = argument(0); // type: int
6943 Node* src = argument(1); // type: long
6944 Node* offset = argument(3); // type: int
6945 Node* length = argument(4); // type: int
6946
6947 src = ConvL2X(src); // adjust Java long to machine word
6948 Node* base = _gvn.transform(new CastX2PNode(src));
6949 offset = ConvI2X(offset);
6950
6951 // 'src_start' points to src array + scaled offset
6952 Node* src_start = basic_plus_adr(top(), base, offset);
6953
6954 // Call the stub.
6955 address stubAddr = StubRoutines::updateBytesAdler32();
6956 const char *stubName = "updateBytesAdler32";
6957
6958 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
6959 stubAddr, stubName, TypePtr::BOTTOM,
6960 crc, src_start, length);
6961
6962 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6963 set_result(result);
6964 return true;
6965 }
6966
6967 //----------------------------inline_reference_get0----------------------------
6968 // public T java.lang.ref.Reference.get();
6969 bool LibraryCallKit::inline_reference_get0() {
6970 const int referent_offset = java_lang_ref_Reference::referent_offset();
6971
6972 // Get the argument:
6973 Node* reference_obj = null_check_receiver();
6974 if (stopped()) return true;
6975
6976 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
6977 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
6978 decorators, /*is_static*/ false, nullptr);
6979 if (result == nullptr) return false;
6980
6981 // Add memory barrier to prevent commoning reads from this field
6982 // across safepoint since GC can change its value.
6983 insert_mem_bar(Op_MemBarCPUOrder);
6984
6985 set_result(result);
6986 return true;
6987 }
6988
6989 //----------------------------inline_reference_refersTo0----------------------------
6990 // bool java.lang.ref.Reference.refersTo0();
6991 // bool java.lang.ref.PhantomReference.refersTo0();
6992 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
6993 // Get arguments:
6994 Node* reference_obj = null_check_receiver();
6995 Node* other_obj = argument(1);
6996 if (stopped()) return true;
6997
6998 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
6999 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7000 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7001 decorators, /*is_static*/ false, nullptr);
7002 if (referent == nullptr) return false;
7003
7004 // Add memory barrier to prevent commoning reads from this field
7005 // across safepoint since GC can change its value.
7006 insert_mem_bar(Op_MemBarCPUOrder);
7007
7008 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7009 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7010 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7011
7012 RegionNode* region = new RegionNode(3);
7013 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7014
7015 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7016 region->init_req(1, if_true);
7017 phi->init_req(1, intcon(1));
7018
7019 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7020 region->init_req(2, if_false);
7021 phi->init_req(2, intcon(0));
7022
7023 set_control(_gvn.transform(region));
7024 record_for_igvn(region);
7025 set_result(_gvn.transform(phi));
7026 return true;
7027 }
7028
7029 //----------------------------inline_reference_clear0----------------------------
7030 // void java.lang.ref.Reference.clear0();
7031 // void java.lang.ref.PhantomReference.clear0();
7032 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7033 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7034
7035 // Get arguments
7036 Node* reference_obj = null_check_receiver();
7037 if (stopped()) return true;
7038
7039 // Common access parameters
7040 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7041 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7042 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7043 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7044 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7045
7046 Node* referent = access_load_at(reference_obj,
7047 referent_field_addr,
7048 referent_field_addr_type,
7049 val_type,
7050 T_OBJECT,
7051 decorators);
7052
7053 IdealKit ideal(this);
7054 #define __ ideal.
7055 __ if_then(referent, BoolTest::ne, null());
7056 sync_kit(ideal);
7057 access_store_at(reference_obj,
7058 referent_field_addr,
7059 referent_field_addr_type,
7060 null(),
7061 val_type,
7062 T_OBJECT,
7063 decorators);
7064 __ sync_kit(this);
7065 __ end_if();
7066 final_sync(ideal);
7067 #undef __
7068
7069 return true;
7070 }
7071
7072 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7073 DecoratorSet decorators, bool is_static,
7074 ciInstanceKlass* fromKls) {
7075 if (fromKls == nullptr) {
7076 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7077 assert(tinst != nullptr, "obj is null");
7078 assert(tinst->is_loaded(), "obj is not loaded");
7079 fromKls = tinst->instance_klass();
7080 } else {
7081 assert(is_static, "only for static field access");
7082 }
7083 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7084 ciSymbol::make(fieldTypeString),
7085 is_static);
7086
7087 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7088 if (field == nullptr) return (Node *) nullptr;
7089
7090 if (is_static) {
7091 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7092 fromObj = makecon(tip);
7093 }
7094
7095 // Next code copied from Parse::do_get_xxx():
7096
7097 // Compute address and memory type.
7098 int offset = field->offset_in_bytes();
7099 bool is_vol = field->is_volatile();
7100 ciType* field_klass = field->type();
7101 assert(field_klass->is_loaded(), "should be loaded");
7102 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7103 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7104 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7105 "slice of address and input slice don't match");
7106 BasicType bt = field->layout_type();
7107
7108 // Build the resultant type of the load
7109 const Type *type;
7110 if (bt == T_OBJECT) {
7111 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7112 } else {
7113 type = Type::get_const_basic_type(bt);
7114 }
7115
7116 if (is_vol) {
7117 decorators |= MO_SEQ_CST;
7118 }
7119
7120 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7121 }
7122
7123 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7124 bool is_exact /* true */, bool is_static /* false */,
7125 ciInstanceKlass * fromKls /* nullptr */) {
7126 if (fromKls == nullptr) {
7127 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7128 assert(tinst != nullptr, "obj is null");
7129 assert(tinst->is_loaded(), "obj is not loaded");
7130 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7131 fromKls = tinst->instance_klass();
7132 }
7133 else {
7134 assert(is_static, "only for static field access");
7135 }
7136 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7137 ciSymbol::make(fieldTypeString),
7138 is_static);
7139
7140 assert(field != nullptr, "undefined field");
7141 assert(!field->is_volatile(), "not defined for volatile fields");
7142
7143 if (is_static) {
7144 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7145 fromObj = makecon(tip);
7146 }
7147
7148 // Next code copied from Parse::do_get_xxx():
7149
7150 // Compute address and memory type.
7151 int offset = field->offset_in_bytes();
7152 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7153
7154 return adr;
7155 }
7156
7157 //------------------------------inline_aescrypt_Block-----------------------
7158 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7159 address stubAddr = nullptr;
7160 const char *stubName;
7161 bool is_decrypt = false;
7162 assert(UseAES, "need AES instruction support");
7163
7164 switch(id) {
7165 case vmIntrinsics::_aescrypt_encryptBlock:
7166 stubAddr = StubRoutines::aescrypt_encryptBlock();
7167 stubName = "aescrypt_encryptBlock";
7168 break;
7169 case vmIntrinsics::_aescrypt_decryptBlock:
7170 stubAddr = StubRoutines::aescrypt_decryptBlock();
7171 stubName = "aescrypt_decryptBlock";
7172 is_decrypt = true;
7173 break;
7174 default:
7175 break;
7176 }
7177 if (stubAddr == nullptr) return false;
7178
7179 Node* aescrypt_object = argument(0);
7180 Node* src = argument(1);
7181 Node* src_offset = argument(2);
7182 Node* dest = argument(3);
7183 Node* dest_offset = argument(4);
7184
7185 src = must_be_not_null(src, true);
7186 dest = must_be_not_null(dest, true);
7187
7188 // (1) src and dest are arrays.
7189 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7190 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7191 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7192 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7193
7194 // for the quick and dirty code we will skip all the checks.
7195 // we are just trying to get the call to be generated.
7196 Node* src_start = src;
7197 Node* dest_start = dest;
7198 if (src_offset != nullptr || dest_offset != nullptr) {
7199 assert(src_offset != nullptr && dest_offset != nullptr, "");
7200 src_start = array_element_address(src, src_offset, T_BYTE);
7201 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7202 }
7203
7204 // now need to get the start of its expanded key array
7205 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7206 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7207 if (k_start == nullptr) return false;
7208
7209 // Call the stub.
7210 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7211 stubAddr, stubName, TypePtr::BOTTOM,
7212 src_start, dest_start, k_start);
7213
7214 return true;
7215 }
7216
7217 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7218 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7219 address stubAddr = nullptr;
7220 const char *stubName = nullptr;
7221 bool is_decrypt = false;
7222 assert(UseAES, "need AES instruction support");
7223
7224 switch(id) {
7225 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7226 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7227 stubName = "cipherBlockChaining_encryptAESCrypt";
7228 break;
7229 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7230 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7231 stubName = "cipherBlockChaining_decryptAESCrypt";
7232 is_decrypt = true;
7233 break;
7234 default:
7235 break;
7236 }
7237 if (stubAddr == nullptr) return false;
7238
7239 Node* cipherBlockChaining_object = argument(0);
7240 Node* src = argument(1);
7241 Node* src_offset = argument(2);
7242 Node* len = argument(3);
7243 Node* dest = argument(4);
7244 Node* dest_offset = argument(5);
7245
7246 src = must_be_not_null(src, false);
7247 dest = must_be_not_null(dest, false);
7248
7249 // (1) src and dest are arrays.
7250 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7251 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7252 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7253 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7254
7255 // checks are the responsibility of the caller
7256 Node* src_start = src;
7257 Node* dest_start = dest;
7258 if (src_offset != nullptr || dest_offset != nullptr) {
7259 assert(src_offset != nullptr && dest_offset != nullptr, "");
7260 src_start = array_element_address(src, src_offset, T_BYTE);
7261 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7262 }
7263
7264 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7265 // (because of the predicated logic executed earlier).
7266 // so we cast it here safely.
7267 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7268
7269 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7270 if (embeddedCipherObj == nullptr) return false;
7271
7272 // cast it to what we know it will be at runtime
7273 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7274 assert(tinst != nullptr, "CBC obj is null");
7275 assert(tinst->is_loaded(), "CBC obj is not loaded");
7276 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7277 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7278
7279 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7280 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7281 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7282 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7283 aescrypt_object = _gvn.transform(aescrypt_object);
7284
7285 // we need to get the start of the aescrypt_object's expanded key array
7286 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7287 if (k_start == nullptr) return false;
7288
7289 // similarly, get the start address of the r vector
7290 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7291 if (objRvec == nullptr) return false;
7292 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7293
7294 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7295 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7296 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7297 stubAddr, stubName, TypePtr::BOTTOM,
7298 src_start, dest_start, k_start, r_start, len);
7299
7300 // return cipher length (int)
7301 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7302 set_result(retvalue);
7303 return true;
7304 }
7305
7306 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7307 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7308 address stubAddr = nullptr;
7309 const char *stubName = nullptr;
7310 bool is_decrypt = false;
7311 assert(UseAES, "need AES instruction support");
7312
7313 switch (id) {
7314 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7315 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7316 stubName = "electronicCodeBook_encryptAESCrypt";
7317 break;
7318 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7319 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7320 stubName = "electronicCodeBook_decryptAESCrypt";
7321 is_decrypt = true;
7322 break;
7323 default:
7324 break;
7325 }
7326
7327 if (stubAddr == nullptr) return false;
7328
7329 Node* electronicCodeBook_object = argument(0);
7330 Node* src = argument(1);
7331 Node* src_offset = argument(2);
7332 Node* len = argument(3);
7333 Node* dest = argument(4);
7334 Node* dest_offset = argument(5);
7335
7336 // (1) src and dest are arrays.
7337 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7338 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7339 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7340 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7341
7342 // checks are the responsibility of the caller
7343 Node* src_start = src;
7344 Node* dest_start = dest;
7345 if (src_offset != nullptr || dest_offset != nullptr) {
7346 assert(src_offset != nullptr && dest_offset != nullptr, "");
7347 src_start = array_element_address(src, src_offset, T_BYTE);
7348 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7349 }
7350
7351 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7352 // (because of the predicated logic executed earlier).
7353 // so we cast it here safely.
7354 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7355
7356 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7357 if (embeddedCipherObj == nullptr) return false;
7358
7359 // cast it to what we know it will be at runtime
7360 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7361 assert(tinst != nullptr, "ECB obj is null");
7362 assert(tinst->is_loaded(), "ECB obj is not loaded");
7363 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7364 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7365
7366 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7367 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7368 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7369 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7370 aescrypt_object = _gvn.transform(aescrypt_object);
7371
7372 // we need to get the start of the aescrypt_object's expanded key array
7373 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, is_decrypt);
7374 if (k_start == nullptr) return false;
7375
7376 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7377 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7378 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7379 stubAddr, stubName, TypePtr::BOTTOM,
7380 src_start, dest_start, k_start, len);
7381
7382 // return cipher length (int)
7383 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7384 set_result(retvalue);
7385 return true;
7386 }
7387
7388 //------------------------------inline_counterMode_AESCrypt-----------------------
7389 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7390 assert(UseAES, "need AES instruction support");
7391 if (!UseAESCTRIntrinsics) return false;
7392
7393 address stubAddr = nullptr;
7394 const char *stubName = nullptr;
7395 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7396 stubAddr = StubRoutines::counterMode_AESCrypt();
7397 stubName = "counterMode_AESCrypt";
7398 }
7399 if (stubAddr == nullptr) return false;
7400
7401 Node* counterMode_object = argument(0);
7402 Node* src = argument(1);
7403 Node* src_offset = argument(2);
7404 Node* len = argument(3);
7405 Node* dest = argument(4);
7406 Node* dest_offset = argument(5);
7407
7408 // (1) src and dest are arrays.
7409 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7410 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7411 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7412 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7413
7414 // checks are the responsibility of the caller
7415 Node* src_start = src;
7416 Node* dest_start = dest;
7417 if (src_offset != nullptr || dest_offset != nullptr) {
7418 assert(src_offset != nullptr && dest_offset != nullptr, "");
7419 src_start = array_element_address(src, src_offset, T_BYTE);
7420 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7421 }
7422
7423 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7424 // (because of the predicated logic executed earlier).
7425 // so we cast it here safely.
7426 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7427 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7428 if (embeddedCipherObj == nullptr) return false;
7429 // cast it to what we know it will be at runtime
7430 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7431 assert(tinst != nullptr, "CTR obj is null");
7432 assert(tinst->is_loaded(), "CTR obj is not loaded");
7433 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7434 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7435 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7436 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7437 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7438 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7439 aescrypt_object = _gvn.transform(aescrypt_object);
7440 // we need to get the start of the aescrypt_object's expanded key array
7441 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
7442 if (k_start == nullptr) return false;
7443 // similarly, get the start address of the r vector
7444 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7445 if (obj_counter == nullptr) return false;
7446 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7447
7448 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7449 if (saved_encCounter == nullptr) return false;
7450 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7451 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7452
7453 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7454 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7455 OptoRuntime::counterMode_aescrypt_Type(),
7456 stubAddr, stubName, TypePtr::BOTTOM,
7457 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7458
7459 // return cipher length (int)
7460 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7461 set_result(retvalue);
7462 return true;
7463 }
7464
7465 //------------------------------get_key_start_from_aescrypt_object-----------------------
7466 Node* LibraryCallKit::get_key_start_from_aescrypt_object(Node* aescrypt_object, bool is_decrypt) {
7467 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7468 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7469 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7470 // The following platform specific stubs of encryption and decryption use the same round keys.
7471 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7472 bool use_decryption_key = false;
7473 #else
7474 bool use_decryption_key = is_decrypt;
7475 #endif
7476 Node* objAESCryptKey = load_field_from_object(aescrypt_object, use_decryption_key ? "sessionKd" : "sessionKe", "[I");
7477 assert(objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AES_Crypt");
7478 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7479
7480 // now have the array, need to get the start address of the selected key array
7481 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7482 return k_start;
7483 }
7484
7485 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7486 // Return node representing slow path of predicate check.
7487 // the pseudo code we want to emulate with this predicate is:
7488 // for encryption:
7489 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7490 // for decryption:
7491 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7492 // note cipher==plain is more conservative than the original java code but that's OK
7493 //
7494 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7495 // The receiver was checked for null already.
7496 Node* objCBC = argument(0);
7497
7498 Node* src = argument(1);
7499 Node* dest = argument(4);
7500
7501 // Load embeddedCipher field of CipherBlockChaining object.
7502 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7503
7504 // get AESCrypt klass for instanceOf check
7505 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7506 // will have same classloader as CipherBlockChaining object
7507 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7508 assert(tinst != nullptr, "CBCobj is null");
7509 assert(tinst->is_loaded(), "CBCobj is not loaded");
7510
7511 // we want to do an instanceof comparison against the AESCrypt class
7512 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7513 if (!klass_AESCrypt->is_loaded()) {
7514 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7515 Node* ctrl = control();
7516 set_control(top()); // no regular fast path
7517 return ctrl;
7518 }
7519
7520 src = must_be_not_null(src, true);
7521 dest = must_be_not_null(dest, true);
7522
7523 // Resolve oops to stable for CmpP below.
7524 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7525
7526 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7527 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7528 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7529
7530 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7531
7532 // for encryption, we are done
7533 if (!decrypting)
7534 return instof_false; // even if it is null
7535
7536 // for decryption, we need to add a further check to avoid
7537 // taking the intrinsic path when cipher and plain are the same
7538 // see the original java code for why.
7539 RegionNode* region = new RegionNode(3);
7540 region->init_req(1, instof_false);
7541
7542 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7543 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7544 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7545 region->init_req(2, src_dest_conjoint);
7546
7547 record_for_igvn(region);
7548 return _gvn.transform(region);
7549 }
7550
7551 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7552 // Return node representing slow path of predicate check.
7553 // the pseudo code we want to emulate with this predicate is:
7554 // for encryption:
7555 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7556 // for decryption:
7557 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7558 // note cipher==plain is more conservative than the original java code but that's OK
7559 //
7560 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7561 // The receiver was checked for null already.
7562 Node* objECB = argument(0);
7563
7564 // Load embeddedCipher field of ElectronicCodeBook object.
7565 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7566
7567 // get AESCrypt klass for instanceOf check
7568 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7569 // will have same classloader as ElectronicCodeBook object
7570 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7571 assert(tinst != nullptr, "ECBobj is null");
7572 assert(tinst->is_loaded(), "ECBobj is not loaded");
7573
7574 // we want to do an instanceof comparison against the AESCrypt class
7575 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7576 if (!klass_AESCrypt->is_loaded()) {
7577 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7578 Node* ctrl = control();
7579 set_control(top()); // no regular fast path
7580 return ctrl;
7581 }
7582 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7583
7584 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7585 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7586 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7587
7588 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7589
7590 // for encryption, we are done
7591 if (!decrypting)
7592 return instof_false; // even if it is null
7593
7594 // for decryption, we need to add a further check to avoid
7595 // taking the intrinsic path when cipher and plain are the same
7596 // see the original java code for why.
7597 RegionNode* region = new RegionNode(3);
7598 region->init_req(1, instof_false);
7599 Node* src = argument(1);
7600 Node* dest = argument(4);
7601 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7602 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7603 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7604 region->init_req(2, src_dest_conjoint);
7605
7606 record_for_igvn(region);
7607 return _gvn.transform(region);
7608 }
7609
7610 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7611 // Return node representing slow path of predicate check.
7612 // the pseudo code we want to emulate with this predicate is:
7613 // for encryption:
7614 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7615 // for decryption:
7616 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7617 // note cipher==plain is more conservative than the original java code but that's OK
7618 //
7619
7620 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7621 // The receiver was checked for null already.
7622 Node* objCTR = argument(0);
7623
7624 // Load embeddedCipher field of CipherBlockChaining object.
7625 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7626
7627 // get AESCrypt klass for instanceOf check
7628 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7629 // will have same classloader as CipherBlockChaining object
7630 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7631 assert(tinst != nullptr, "CTRobj is null");
7632 assert(tinst->is_loaded(), "CTRobj is not loaded");
7633
7634 // we want to do an instanceof comparison against the AESCrypt class
7635 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
7636 if (!klass_AESCrypt->is_loaded()) {
7637 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7638 Node* ctrl = control();
7639 set_control(top()); // no regular fast path
7640 return ctrl;
7641 }
7642
7643 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7644 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7645 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7646 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7647 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7648
7649 return instof_false; // even if it is null
7650 }
7651
7652 //------------------------------inline_ghash_processBlocks
7653 bool LibraryCallKit::inline_ghash_processBlocks() {
7654 address stubAddr;
7655 const char *stubName;
7656 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7657
7658 stubAddr = StubRoutines::ghash_processBlocks();
7659 stubName = "ghash_processBlocks";
7660
7661 Node* data = argument(0);
7662 Node* offset = argument(1);
7663 Node* len = argument(2);
7664 Node* state = argument(3);
7665 Node* subkeyH = argument(4);
7666
7667 state = must_be_not_null(state, true);
7668 subkeyH = must_be_not_null(subkeyH, true);
7669 data = must_be_not_null(data, true);
7670
7671 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7672 assert(state_start, "state is null");
7673 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7674 assert(subkeyH_start, "subkeyH is null");
7675 Node* data_start = array_element_address(data, offset, T_BYTE);
7676 assert(data_start, "data is null");
7677
7678 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7679 OptoRuntime::ghash_processBlocks_Type(),
7680 stubAddr, stubName, TypePtr::BOTTOM,
7681 state_start, subkeyH_start, data_start, len);
7682 return true;
7683 }
7684
7685 //------------------------------inline_chacha20Block
7686 bool LibraryCallKit::inline_chacha20Block() {
7687 address stubAddr;
7688 const char *stubName;
7689 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7690
7691 stubAddr = StubRoutines::chacha20Block();
7692 stubName = "chacha20Block";
7693
7694 Node* state = argument(0);
7695 Node* result = argument(1);
7696
7697 state = must_be_not_null(state, true);
7698 result = must_be_not_null(result, true);
7699
7700 Node* state_start = array_element_address(state, intcon(0), T_INT);
7701 assert(state_start, "state is null");
7702 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7703 assert(result_start, "result is null");
7704
7705 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7706 OptoRuntime::chacha20Block_Type(),
7707 stubAddr, stubName, TypePtr::BOTTOM,
7708 state_start, result_start);
7709 // return key stream length (int)
7710 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7711 set_result(retvalue);
7712 return true;
7713 }
7714
7715 //------------------------------inline_kyberNtt
7716 bool LibraryCallKit::inline_kyberNtt() {
7717 address stubAddr;
7718 const char *stubName;
7719 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7720 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7721
7722 stubAddr = StubRoutines::kyberNtt();
7723 stubName = "kyberNtt";
7724 if (!stubAddr) return false;
7725
7726 Node* coeffs = argument(0);
7727 Node* ntt_zetas = argument(1);
7728
7729 coeffs = must_be_not_null(coeffs, true);
7730 ntt_zetas = must_be_not_null(ntt_zetas, true);
7731
7732 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7733 assert(coeffs_start, "coeffs is null");
7734 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7735 assert(ntt_zetas_start, "ntt_zetas is null");
7736 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7737 OptoRuntime::kyberNtt_Type(),
7738 stubAddr, stubName, TypePtr::BOTTOM,
7739 coeffs_start, ntt_zetas_start);
7740 // return an int
7741 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7742 set_result(retvalue);
7743 return true;
7744 }
7745
7746 //------------------------------inline_kyberInverseNtt
7747 bool LibraryCallKit::inline_kyberInverseNtt() {
7748 address stubAddr;
7749 const char *stubName;
7750 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7751 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7752
7753 stubAddr = StubRoutines::kyberInverseNtt();
7754 stubName = "kyberInverseNtt";
7755 if (!stubAddr) return false;
7756
7757 Node* coeffs = argument(0);
7758 Node* zetas = argument(1);
7759
7760 coeffs = must_be_not_null(coeffs, true);
7761 zetas = must_be_not_null(zetas, true);
7762
7763 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7764 assert(coeffs_start, "coeffs is null");
7765 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7766 assert(zetas_start, "inverseNtt_zetas is null");
7767 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7768 OptoRuntime::kyberInverseNtt_Type(),
7769 stubAddr, stubName, TypePtr::BOTTOM,
7770 coeffs_start, zetas_start);
7771
7772 // return an int
7773 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7774 set_result(retvalue);
7775 return true;
7776 }
7777
7778 //------------------------------inline_kyberNttMult
7779 bool LibraryCallKit::inline_kyberNttMult() {
7780 address stubAddr;
7781 const char *stubName;
7782 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7783 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7784
7785 stubAddr = StubRoutines::kyberNttMult();
7786 stubName = "kyberNttMult";
7787 if (!stubAddr) return false;
7788
7789 Node* result = argument(0);
7790 Node* ntta = argument(1);
7791 Node* nttb = argument(2);
7792 Node* zetas = argument(3);
7793
7794 result = must_be_not_null(result, true);
7795 ntta = must_be_not_null(ntta, true);
7796 nttb = must_be_not_null(nttb, true);
7797 zetas = must_be_not_null(zetas, true);
7798
7799 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7800 assert(result_start, "result is null");
7801 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7802 assert(ntta_start, "ntta is null");
7803 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7804 assert(nttb_start, "nttb is null");
7805 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7806 assert(zetas_start, "nttMult_zetas is null");
7807 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7808 OptoRuntime::kyberNttMult_Type(),
7809 stubAddr, stubName, TypePtr::BOTTOM,
7810 result_start, ntta_start, nttb_start,
7811 zetas_start);
7812
7813 // return an int
7814 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7815 set_result(retvalue);
7816
7817 return true;
7818 }
7819
7820 //------------------------------inline_kyberAddPoly_2
7821 bool LibraryCallKit::inline_kyberAddPoly_2() {
7822 address stubAddr;
7823 const char *stubName;
7824 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7825 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7826
7827 stubAddr = StubRoutines::kyberAddPoly_2();
7828 stubName = "kyberAddPoly_2";
7829 if (!stubAddr) return false;
7830
7831 Node* result = argument(0);
7832 Node* a = argument(1);
7833 Node* b = argument(2);
7834
7835 result = must_be_not_null(result, true);
7836 a = must_be_not_null(a, true);
7837 b = must_be_not_null(b, true);
7838
7839 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7840 assert(result_start, "result is null");
7841 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7842 assert(a_start, "a is null");
7843 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7844 assert(b_start, "b is null");
7845 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7846 OptoRuntime::kyberAddPoly_2_Type(),
7847 stubAddr, stubName, TypePtr::BOTTOM,
7848 result_start, a_start, b_start);
7849 // return an int
7850 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7851 set_result(retvalue);
7852 return true;
7853 }
7854
7855 //------------------------------inline_kyberAddPoly_3
7856 bool LibraryCallKit::inline_kyberAddPoly_3() {
7857 address stubAddr;
7858 const char *stubName;
7859 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7860 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7861
7862 stubAddr = StubRoutines::kyberAddPoly_3();
7863 stubName = "kyberAddPoly_3";
7864 if (!stubAddr) return false;
7865
7866 Node* result = argument(0);
7867 Node* a = argument(1);
7868 Node* b = argument(2);
7869 Node* c = argument(3);
7870
7871 result = must_be_not_null(result, true);
7872 a = must_be_not_null(a, true);
7873 b = must_be_not_null(b, true);
7874 c = must_be_not_null(c, true);
7875
7876 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7877 assert(result_start, "result is null");
7878 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7879 assert(a_start, "a is null");
7880 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7881 assert(b_start, "b is null");
7882 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7883 assert(c_start, "c is null");
7884 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7885 OptoRuntime::kyberAddPoly_3_Type(),
7886 stubAddr, stubName, TypePtr::BOTTOM,
7887 result_start, a_start, b_start, c_start);
7888 // return an int
7889 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7890 set_result(retvalue);
7891 return true;
7892 }
7893
7894 //------------------------------inline_kyber12To16
7895 bool LibraryCallKit::inline_kyber12To16() {
7896 address stubAddr;
7897 const char *stubName;
7898 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7899 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7900
7901 stubAddr = StubRoutines::kyber12To16();
7902 stubName = "kyber12To16";
7903 if (!stubAddr) return false;
7904
7905 Node* condensed = argument(0);
7906 Node* condensedOffs = argument(1);
7907 Node* parsed = argument(2);
7908 Node* parsedLength = argument(3);
7909
7910 condensed = must_be_not_null(condensed, true);
7911 parsed = must_be_not_null(parsed, true);
7912
7913 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
7914 assert(condensed_start, "condensed is null");
7915 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
7916 assert(parsed_start, "parsed is null");
7917 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
7918 OptoRuntime::kyber12To16_Type(),
7919 stubAddr, stubName, TypePtr::BOTTOM,
7920 condensed_start, condensedOffs, parsed_start, parsedLength);
7921 // return an int
7922 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
7923 set_result(retvalue);
7924 return true;
7925
7926 }
7927
7928 //------------------------------inline_kyberBarrettReduce
7929 bool LibraryCallKit::inline_kyberBarrettReduce() {
7930 address stubAddr;
7931 const char *stubName;
7932 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7933 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
7934
7935 stubAddr = StubRoutines::kyberBarrettReduce();
7936 stubName = "kyberBarrettReduce";
7937 if (!stubAddr) return false;
7938
7939 Node* coeffs = argument(0);
7940
7941 coeffs = must_be_not_null(coeffs, true);
7942
7943 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7944 assert(coeffs_start, "coeffs is null");
7945 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
7946 OptoRuntime::kyberBarrettReduce_Type(),
7947 stubAddr, stubName, TypePtr::BOTTOM,
7948 coeffs_start);
7949 // return an int
7950 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
7951 set_result(retvalue);
7952 return true;
7953 }
7954
7955 //------------------------------inline_dilithiumAlmostNtt
7956 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
7957 address stubAddr;
7958 const char *stubName;
7959 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7960 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
7961
7962 stubAddr = StubRoutines::dilithiumAlmostNtt();
7963 stubName = "dilithiumAlmostNtt";
7964 if (!stubAddr) return false;
7965
7966 Node* coeffs = argument(0);
7967 Node* ntt_zetas = argument(1);
7968
7969 coeffs = must_be_not_null(coeffs, true);
7970 ntt_zetas = must_be_not_null(ntt_zetas, true);
7971
7972 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
7973 assert(coeffs_start, "coeffs is null");
7974 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
7975 assert(ntt_zetas_start, "ntt_zetas is null");
7976 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7977 OptoRuntime::dilithiumAlmostNtt_Type(),
7978 stubAddr, stubName, TypePtr::BOTTOM,
7979 coeffs_start, ntt_zetas_start);
7980 // return an int
7981 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
7982 set_result(retvalue);
7983 return true;
7984 }
7985
7986 //------------------------------inline_dilithiumAlmostInverseNtt
7987 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
7988 address stubAddr;
7989 const char *stubName;
7990 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
7991 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
7992
7993 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
7994 stubName = "dilithiumAlmostInverseNtt";
7995 if (!stubAddr) return false;
7996
7997 Node* coeffs = argument(0);
7998 Node* zetas = argument(1);
7999
8000 coeffs = must_be_not_null(coeffs, true);
8001 zetas = must_be_not_null(zetas, true);
8002
8003 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8004 assert(coeffs_start, "coeffs is null");
8005 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8006 assert(zetas_start, "inverseNtt_zetas is null");
8007 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8008 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8009 stubAddr, stubName, TypePtr::BOTTOM,
8010 coeffs_start, zetas_start);
8011 // return an int
8012 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8013 set_result(retvalue);
8014 return true;
8015 }
8016
8017 //------------------------------inline_dilithiumNttMult
8018 bool LibraryCallKit::inline_dilithiumNttMult() {
8019 address stubAddr;
8020 const char *stubName;
8021 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8022 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8023
8024 stubAddr = StubRoutines::dilithiumNttMult();
8025 stubName = "dilithiumNttMult";
8026 if (!stubAddr) return false;
8027
8028 Node* result = argument(0);
8029 Node* ntta = argument(1);
8030 Node* nttb = argument(2);
8031 Node* zetas = argument(3);
8032
8033 result = must_be_not_null(result, true);
8034 ntta = must_be_not_null(ntta, true);
8035 nttb = must_be_not_null(nttb, true);
8036 zetas = must_be_not_null(zetas, true);
8037
8038 Node* result_start = array_element_address(result, intcon(0), T_INT);
8039 assert(result_start, "result is null");
8040 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8041 assert(ntta_start, "ntta is null");
8042 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8043 assert(nttb_start, "nttb is null");
8044 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8045 OptoRuntime::dilithiumNttMult_Type(),
8046 stubAddr, stubName, TypePtr::BOTTOM,
8047 result_start, ntta_start, nttb_start);
8048
8049 // return an int
8050 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8051 set_result(retvalue);
8052
8053 return true;
8054 }
8055
8056 //------------------------------inline_dilithiumMontMulByConstant
8057 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8058 address stubAddr;
8059 const char *stubName;
8060 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8061 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8062
8063 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8064 stubName = "dilithiumMontMulByConstant";
8065 if (!stubAddr) return false;
8066
8067 Node* coeffs = argument(0);
8068 Node* constant = argument(1);
8069
8070 coeffs = must_be_not_null(coeffs, true);
8071
8072 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8073 assert(coeffs_start, "coeffs is null");
8074 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8075 OptoRuntime::dilithiumMontMulByConstant_Type(),
8076 stubAddr, stubName, TypePtr::BOTTOM,
8077 coeffs_start, constant);
8078
8079 // return an int
8080 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8081 set_result(retvalue);
8082 return true;
8083 }
8084
8085
8086 //------------------------------inline_dilithiumDecomposePoly
8087 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8088 address stubAddr;
8089 const char *stubName;
8090 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8091 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8092
8093 stubAddr = StubRoutines::dilithiumDecomposePoly();
8094 stubName = "dilithiumDecomposePoly";
8095 if (!stubAddr) return false;
8096
8097 Node* input = argument(0);
8098 Node* lowPart = argument(1);
8099 Node* highPart = argument(2);
8100 Node* twoGamma2 = argument(3);
8101 Node* multiplier = argument(4);
8102
8103 input = must_be_not_null(input, true);
8104 lowPart = must_be_not_null(lowPart, true);
8105 highPart = must_be_not_null(highPart, true);
8106
8107 Node* input_start = array_element_address(input, intcon(0), T_INT);
8108 assert(input_start, "input is null");
8109 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8110 assert(lowPart_start, "lowPart is null");
8111 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8112 assert(highPart_start, "highPart is null");
8113
8114 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8115 OptoRuntime::dilithiumDecomposePoly_Type(),
8116 stubAddr, stubName, TypePtr::BOTTOM,
8117 input_start, lowPart_start, highPart_start,
8118 twoGamma2, multiplier);
8119
8120 // return an int
8121 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8122 set_result(retvalue);
8123 return true;
8124 }
8125
8126 bool LibraryCallKit::inline_base64_encodeBlock() {
8127 address stubAddr;
8128 const char *stubName;
8129 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8130 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8131 stubAddr = StubRoutines::base64_encodeBlock();
8132 stubName = "encodeBlock";
8133
8134 if (!stubAddr) return false;
8135 Node* base64obj = argument(0);
8136 Node* src = argument(1);
8137 Node* offset = argument(2);
8138 Node* len = argument(3);
8139 Node* dest = argument(4);
8140 Node* dp = argument(5);
8141 Node* isURL = argument(6);
8142
8143 src = must_be_not_null(src, true);
8144 dest = must_be_not_null(dest, true);
8145
8146 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8147 assert(src_start, "source array is null");
8148 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8149 assert(dest_start, "destination array is null");
8150
8151 Node* base64 = make_runtime_call(RC_LEAF,
8152 OptoRuntime::base64_encodeBlock_Type(),
8153 stubAddr, stubName, TypePtr::BOTTOM,
8154 src_start, offset, len, dest_start, dp, isURL);
8155 return true;
8156 }
8157
8158 bool LibraryCallKit::inline_base64_decodeBlock() {
8159 address stubAddr;
8160 const char *stubName;
8161 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8162 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8163 stubAddr = StubRoutines::base64_decodeBlock();
8164 stubName = "decodeBlock";
8165
8166 if (!stubAddr) return false;
8167 Node* base64obj = argument(0);
8168 Node* src = argument(1);
8169 Node* src_offset = argument(2);
8170 Node* len = argument(3);
8171 Node* dest = argument(4);
8172 Node* dest_offset = argument(5);
8173 Node* isURL = argument(6);
8174 Node* isMIME = argument(7);
8175
8176 src = must_be_not_null(src, true);
8177 dest = must_be_not_null(dest, true);
8178
8179 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8180 assert(src_start, "source array is null");
8181 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8182 assert(dest_start, "destination array is null");
8183
8184 Node* call = make_runtime_call(RC_LEAF,
8185 OptoRuntime::base64_decodeBlock_Type(),
8186 stubAddr, stubName, TypePtr::BOTTOM,
8187 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8188 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8189 set_result(result);
8190 return true;
8191 }
8192
8193 bool LibraryCallKit::inline_poly1305_processBlocks() {
8194 address stubAddr;
8195 const char *stubName;
8196 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8197 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8198 stubAddr = StubRoutines::poly1305_processBlocks();
8199 stubName = "poly1305_processBlocks";
8200
8201 if (!stubAddr) return false;
8202 null_check_receiver(); // null-check receiver
8203 if (stopped()) return true;
8204
8205 Node* input = argument(1);
8206 Node* input_offset = argument(2);
8207 Node* len = argument(3);
8208 Node* alimbs = argument(4);
8209 Node* rlimbs = argument(5);
8210
8211 input = must_be_not_null(input, true);
8212 alimbs = must_be_not_null(alimbs, true);
8213 rlimbs = must_be_not_null(rlimbs, true);
8214
8215 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8216 assert(input_start, "input array is null");
8217 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8218 assert(acc_start, "acc array is null");
8219 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8220 assert(r_start, "r array is null");
8221
8222 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8223 OptoRuntime::poly1305_processBlocks_Type(),
8224 stubAddr, stubName, TypePtr::BOTTOM,
8225 input_start, len, acc_start, r_start);
8226 return true;
8227 }
8228
8229 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8230 address stubAddr;
8231 const char *stubName;
8232 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8233 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8234 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8235 stubName = "intpoly_montgomeryMult_P256";
8236
8237 if (!stubAddr) return false;
8238 null_check_receiver(); // null-check receiver
8239 if (stopped()) return true;
8240
8241 Node* a = argument(1);
8242 Node* b = argument(2);
8243 Node* r = argument(3);
8244
8245 a = must_be_not_null(a, true);
8246 b = must_be_not_null(b, true);
8247 r = must_be_not_null(r, true);
8248
8249 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8250 assert(a_start, "a array is null");
8251 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8252 assert(b_start, "b array is null");
8253 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8254 assert(r_start, "r array is null");
8255
8256 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8257 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8258 stubAddr, stubName, TypePtr::BOTTOM,
8259 a_start, b_start, r_start);
8260 return true;
8261 }
8262
8263 bool LibraryCallKit::inline_intpoly_assign() {
8264 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8265 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8266 const char *stubName = "intpoly_assign";
8267 address stubAddr = StubRoutines::intpoly_assign();
8268 if (!stubAddr) return false;
8269
8270 Node* set = argument(0);
8271 Node* a = argument(1);
8272 Node* b = argument(2);
8273 Node* arr_length = load_array_length(a);
8274
8275 a = must_be_not_null(a, true);
8276 b = must_be_not_null(b, true);
8277
8278 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8279 assert(a_start, "a array is null");
8280 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8281 assert(b_start, "b array is null");
8282
8283 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8284 OptoRuntime::intpoly_assign_Type(),
8285 stubAddr, stubName, TypePtr::BOTTOM,
8286 set, a_start, b_start, arr_length);
8287 return true;
8288 }
8289
8290 //------------------------------inline_digestBase_implCompress-----------------------
8291 //
8292 // Calculate MD5 for single-block byte[] array.
8293 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8294 //
8295 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8296 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8297 //
8298 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8299 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8300 //
8301 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8302 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8303 //
8304 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8305 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8306 //
8307 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8308 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8309
8310 Node* digestBase_obj = argument(0);
8311 Node* src = argument(1); // type oop
8312 Node* ofs = argument(2); // type int
8313
8314 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8315 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8316 // failed array check
8317 return false;
8318 }
8319 // Figure out the size and type of the elements we will be copying.
8320 BasicType src_elem = src_type->elem()->array_element_basic_type();
8321 if (src_elem != T_BYTE) {
8322 return false;
8323 }
8324 // 'src_start' points to src array + offset
8325 src = must_be_not_null(src, true);
8326 Node* src_start = array_element_address(src, ofs, src_elem);
8327 Node* state = nullptr;
8328 Node* block_size = nullptr;
8329 address stubAddr;
8330 const char *stubName;
8331
8332 switch(id) {
8333 case vmIntrinsics::_md5_implCompress:
8334 assert(UseMD5Intrinsics, "need MD5 instruction support");
8335 state = get_state_from_digest_object(digestBase_obj, T_INT);
8336 stubAddr = StubRoutines::md5_implCompress();
8337 stubName = "md5_implCompress";
8338 break;
8339 case vmIntrinsics::_sha_implCompress:
8340 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8341 state = get_state_from_digest_object(digestBase_obj, T_INT);
8342 stubAddr = StubRoutines::sha1_implCompress();
8343 stubName = "sha1_implCompress";
8344 break;
8345 case vmIntrinsics::_sha2_implCompress:
8346 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8347 state = get_state_from_digest_object(digestBase_obj, T_INT);
8348 stubAddr = StubRoutines::sha256_implCompress();
8349 stubName = "sha256_implCompress";
8350 break;
8351 case vmIntrinsics::_sha5_implCompress:
8352 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8353 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8354 stubAddr = StubRoutines::sha512_implCompress();
8355 stubName = "sha512_implCompress";
8356 break;
8357 case vmIntrinsics::_sha3_implCompress:
8358 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8359 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8360 stubAddr = StubRoutines::sha3_implCompress();
8361 stubName = "sha3_implCompress";
8362 block_size = get_block_size_from_digest_object(digestBase_obj);
8363 if (block_size == nullptr) return false;
8364 break;
8365 default:
8366 fatal_unexpected_iid(id);
8367 return false;
8368 }
8369 if (state == nullptr) return false;
8370
8371 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8372 if (stubAddr == nullptr) return false;
8373
8374 // Call the stub.
8375 Node* call;
8376 if (block_size == nullptr) {
8377 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8378 stubAddr, stubName, TypePtr::BOTTOM,
8379 src_start, state);
8380 } else {
8381 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8382 stubAddr, stubName, TypePtr::BOTTOM,
8383 src_start, state, block_size);
8384 }
8385
8386 return true;
8387 }
8388
8389 //------------------------------inline_double_keccak
8390 bool LibraryCallKit::inline_double_keccak() {
8391 address stubAddr;
8392 const char *stubName;
8393 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8394 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8395
8396 stubAddr = StubRoutines::double_keccak();
8397 stubName = "double_keccak";
8398 if (!stubAddr) return false;
8399
8400 Node* status0 = argument(0);
8401 Node* status1 = argument(1);
8402
8403 status0 = must_be_not_null(status0, true);
8404 status1 = must_be_not_null(status1, true);
8405
8406 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8407 assert(status0_start, "status0 is null");
8408 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8409 assert(status1_start, "status1 is null");
8410 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8411 OptoRuntime::double_keccak_Type(),
8412 stubAddr, stubName, TypePtr::BOTTOM,
8413 status0_start, status1_start);
8414 // return an int
8415 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8416 set_result(retvalue);
8417 return true;
8418 }
8419
8420
8421 //------------------------------inline_digestBase_implCompressMB-----------------------
8422 //
8423 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8424 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8425 //
8426 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8427 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8428 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8429 assert((uint)predicate < 5, "sanity");
8430 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8431
8432 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8433 Node* src = argument(1); // byte[] array
8434 Node* ofs = argument(2); // type int
8435 Node* limit = argument(3); // type int
8436
8437 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8438 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8439 // failed array check
8440 return false;
8441 }
8442 // Figure out the size and type of the elements we will be copying.
8443 BasicType src_elem = src_type->elem()->array_element_basic_type();
8444 if (src_elem != T_BYTE) {
8445 return false;
8446 }
8447 // 'src_start' points to src array + offset
8448 src = must_be_not_null(src, false);
8449 Node* src_start = array_element_address(src, ofs, src_elem);
8450
8451 const char* klass_digestBase_name = nullptr;
8452 const char* stub_name = nullptr;
8453 address stub_addr = nullptr;
8454 BasicType elem_type = T_INT;
8455
8456 switch (predicate) {
8457 case 0:
8458 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8459 klass_digestBase_name = "sun/security/provider/MD5";
8460 stub_name = "md5_implCompressMB";
8461 stub_addr = StubRoutines::md5_implCompressMB();
8462 }
8463 break;
8464 case 1:
8465 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8466 klass_digestBase_name = "sun/security/provider/SHA";
8467 stub_name = "sha1_implCompressMB";
8468 stub_addr = StubRoutines::sha1_implCompressMB();
8469 }
8470 break;
8471 case 2:
8472 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8473 klass_digestBase_name = "sun/security/provider/SHA2";
8474 stub_name = "sha256_implCompressMB";
8475 stub_addr = StubRoutines::sha256_implCompressMB();
8476 }
8477 break;
8478 case 3:
8479 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8480 klass_digestBase_name = "sun/security/provider/SHA5";
8481 stub_name = "sha512_implCompressMB";
8482 stub_addr = StubRoutines::sha512_implCompressMB();
8483 elem_type = T_LONG;
8484 }
8485 break;
8486 case 4:
8487 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8488 klass_digestBase_name = "sun/security/provider/SHA3";
8489 stub_name = "sha3_implCompressMB";
8490 stub_addr = StubRoutines::sha3_implCompressMB();
8491 elem_type = T_LONG;
8492 }
8493 break;
8494 default:
8495 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8496 }
8497 if (klass_digestBase_name != nullptr) {
8498 assert(stub_addr != nullptr, "Stub is generated");
8499 if (stub_addr == nullptr) return false;
8500
8501 // get DigestBase klass to lookup for SHA klass
8502 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8503 assert(tinst != nullptr, "digestBase_obj is not instance???");
8504 assert(tinst->is_loaded(), "DigestBase is not loaded");
8505
8506 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8507 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8508 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8509 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8510 }
8511 return false;
8512 }
8513
8514 //------------------------------inline_digestBase_implCompressMB-----------------------
8515 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8516 BasicType elem_type, address stubAddr, const char *stubName,
8517 Node* src_start, Node* ofs, Node* limit) {
8518 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8519 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8520 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8521 digest_obj = _gvn.transform(digest_obj);
8522
8523 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8524 if (state == nullptr) return false;
8525
8526 Node* block_size = nullptr;
8527 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8528 block_size = get_block_size_from_digest_object(digest_obj);
8529 if (block_size == nullptr) return false;
8530 }
8531
8532 // Call the stub.
8533 Node* call;
8534 if (block_size == nullptr) {
8535 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8536 OptoRuntime::digestBase_implCompressMB_Type(false),
8537 stubAddr, stubName, TypePtr::BOTTOM,
8538 src_start, state, ofs, limit);
8539 } else {
8540 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8541 OptoRuntime::digestBase_implCompressMB_Type(true),
8542 stubAddr, stubName, TypePtr::BOTTOM,
8543 src_start, state, block_size, ofs, limit);
8544 }
8545
8546 // return ofs (int)
8547 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8548 set_result(result);
8549
8550 return true;
8551 }
8552
8553 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8554 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8555 assert(UseAES, "need AES instruction support");
8556 address stubAddr = nullptr;
8557 const char *stubName = nullptr;
8558 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8559 stubName = "galoisCounterMode_AESCrypt";
8560
8561 if (stubAddr == nullptr) return false;
8562
8563 Node* in = argument(0);
8564 Node* inOfs = argument(1);
8565 Node* len = argument(2);
8566 Node* ct = argument(3);
8567 Node* ctOfs = argument(4);
8568 Node* out = argument(5);
8569 Node* outOfs = argument(6);
8570 Node* gctr_object = argument(7);
8571 Node* ghash_object = argument(8);
8572
8573 // (1) in, ct and out are arrays.
8574 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8575 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8576 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8577 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8578 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8579 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8580
8581 // checks are the responsibility of the caller
8582 Node* in_start = in;
8583 Node* ct_start = ct;
8584 Node* out_start = out;
8585 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8586 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8587 in_start = array_element_address(in, inOfs, T_BYTE);
8588 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8589 out_start = array_element_address(out, outOfs, T_BYTE);
8590 }
8591
8592 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8593 // (because of the predicated logic executed earlier).
8594 // so we cast it here safely.
8595 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8596 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8597 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8598 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8599 Node* state = load_field_from_object(ghash_object, "state", "[J");
8600
8601 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8602 return false;
8603 }
8604 // cast it to what we know it will be at runtime
8605 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8606 assert(tinst != nullptr, "GCTR obj is null");
8607 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8608 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8609 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8610 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8611 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8612 const TypeOopPtr* xtype = aklass->as_instance_type();
8613 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8614 aescrypt_object = _gvn.transform(aescrypt_object);
8615 // we need to get the start of the aescrypt_object's expanded key array
8616 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object, /* is_decrypt */ false);
8617 if (k_start == nullptr) return false;
8618 // similarly, get the start address of the r vector
8619 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8620 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8621 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8622
8623
8624 // Call the stub, passing params
8625 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8626 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8627 stubAddr, stubName, TypePtr::BOTTOM,
8628 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8629
8630 // return cipher length (int)
8631 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8632 set_result(retvalue);
8633
8634 return true;
8635 }
8636
8637 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8638 // Return node representing slow path of predicate check.
8639 // the pseudo code we want to emulate with this predicate is:
8640 // for encryption:
8641 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8642 // for decryption:
8643 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8644 // note cipher==plain is more conservative than the original java code but that's OK
8645 //
8646
8647 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8648 // The receiver was checked for null already.
8649 Node* objGCTR = argument(7);
8650 // Load embeddedCipher field of GCTR object.
8651 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8652 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8653
8654 // get AESCrypt klass for instanceOf check
8655 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8656 // will have same classloader as CipherBlockChaining object
8657 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8658 assert(tinst != nullptr, "GCTR obj is null");
8659 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8660
8661 // we want to do an instanceof comparison against the AESCrypt class
8662 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AES_Crypt"));
8663 if (!klass_AESCrypt->is_loaded()) {
8664 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8665 Node* ctrl = control();
8666 set_control(top()); // no regular fast path
8667 return ctrl;
8668 }
8669
8670 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8671 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8672 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8673 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8674 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8675
8676 return instof_false; // even if it is null
8677 }
8678
8679 //------------------------------get_state_from_digest_object-----------------------
8680 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8681 const char* state_type;
8682 switch (elem_type) {
8683 case T_BYTE: state_type = "[B"; break;
8684 case T_INT: state_type = "[I"; break;
8685 case T_LONG: state_type = "[J"; break;
8686 default: ShouldNotReachHere();
8687 }
8688 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8689 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8690 if (digest_state == nullptr) return (Node *) nullptr;
8691
8692 // now have the array, need to get the start address of the state array
8693 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8694 return state;
8695 }
8696
8697 //------------------------------get_block_size_from_sha3_object----------------------------------
8698 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8699 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8700 assert (block_size != nullptr, "sanity");
8701 return block_size;
8702 }
8703
8704 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8705 // Return node representing slow path of predicate check.
8706 // the pseudo code we want to emulate with this predicate is:
8707 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8708 //
8709 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8710 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8711 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8712 assert((uint)predicate < 5, "sanity");
8713
8714 // The receiver was checked for null already.
8715 Node* digestBaseObj = argument(0);
8716
8717 // get DigestBase klass for instanceOf check
8718 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8719 assert(tinst != nullptr, "digestBaseObj is null");
8720 assert(tinst->is_loaded(), "DigestBase is not loaded");
8721
8722 const char* klass_name = nullptr;
8723 switch (predicate) {
8724 case 0:
8725 if (UseMD5Intrinsics) {
8726 // we want to do an instanceof comparison against the MD5 class
8727 klass_name = "sun/security/provider/MD5";
8728 }
8729 break;
8730 case 1:
8731 if (UseSHA1Intrinsics) {
8732 // we want to do an instanceof comparison against the SHA class
8733 klass_name = "sun/security/provider/SHA";
8734 }
8735 break;
8736 case 2:
8737 if (UseSHA256Intrinsics) {
8738 // we want to do an instanceof comparison against the SHA2 class
8739 klass_name = "sun/security/provider/SHA2";
8740 }
8741 break;
8742 case 3:
8743 if (UseSHA512Intrinsics) {
8744 // we want to do an instanceof comparison against the SHA5 class
8745 klass_name = "sun/security/provider/SHA5";
8746 }
8747 break;
8748 case 4:
8749 if (UseSHA3Intrinsics) {
8750 // we want to do an instanceof comparison against the SHA3 class
8751 klass_name = "sun/security/provider/SHA3";
8752 }
8753 break;
8754 default:
8755 fatal("unknown SHA intrinsic predicate: %d", predicate);
8756 }
8757
8758 ciKlass* klass = nullptr;
8759 if (klass_name != nullptr) {
8760 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8761 }
8762 if ((klass == nullptr) || !klass->is_loaded()) {
8763 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8764 Node* ctrl = control();
8765 set_control(top()); // no intrinsic path
8766 return ctrl;
8767 }
8768 ciInstanceKlass* instklass = klass->as_instance_klass();
8769
8770 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8771 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8772 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8773 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8774
8775 return instof_false; // even if it is null
8776 }
8777
8778 //-------------inline_fma-----------------------------------
8779 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8780 Node *a = nullptr;
8781 Node *b = nullptr;
8782 Node *c = nullptr;
8783 Node* result = nullptr;
8784 switch (id) {
8785 case vmIntrinsics::_fmaD:
8786 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8787 // no receiver since it is static method
8788 a = argument(0);
8789 b = argument(2);
8790 c = argument(4);
8791 result = _gvn.transform(new FmaDNode(a, b, c));
8792 break;
8793 case vmIntrinsics::_fmaF:
8794 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8795 a = argument(0);
8796 b = argument(1);
8797 c = argument(2);
8798 result = _gvn.transform(new FmaFNode(a, b, c));
8799 break;
8800 default:
8801 fatal_unexpected_iid(id); break;
8802 }
8803 set_result(result);
8804 return true;
8805 }
8806
8807 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8808 // argument(0) is receiver
8809 Node* codePoint = argument(1);
8810 Node* n = nullptr;
8811
8812 switch (id) {
8813 case vmIntrinsics::_isDigit :
8814 n = new DigitNode(control(), codePoint);
8815 break;
8816 case vmIntrinsics::_isLowerCase :
8817 n = new LowerCaseNode(control(), codePoint);
8818 break;
8819 case vmIntrinsics::_isUpperCase :
8820 n = new UpperCaseNode(control(), codePoint);
8821 break;
8822 case vmIntrinsics::_isWhitespace :
8823 n = new WhitespaceNode(control(), codePoint);
8824 break;
8825 default:
8826 fatal_unexpected_iid(id);
8827 }
8828
8829 set_result(_gvn.transform(n));
8830 return true;
8831 }
8832
8833 bool LibraryCallKit::inline_profileBoolean() {
8834 Node* counts = argument(1);
8835 const TypeAryPtr* ary = nullptr;
8836 ciArray* aobj = nullptr;
8837 if (counts->is_Con()
8838 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8839 && (aobj = ary->const_oop()->as_array()) != nullptr
8840 && (aobj->length() == 2)) {
8841 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8842 jint false_cnt = aobj->element_value(0).as_int();
8843 jint true_cnt = aobj->element_value(1).as_int();
8844
8845 if (C->log() != nullptr) {
8846 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8847 false_cnt, true_cnt);
8848 }
8849
8850 if (false_cnt + true_cnt == 0) {
8851 // According to profile, never executed.
8852 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8853 Deoptimization::Action_reinterpret);
8854 return true;
8855 }
8856
8857 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8858 // is a number of each value occurrences.
8859 Node* result = argument(0);
8860 if (false_cnt == 0 || true_cnt == 0) {
8861 // According to profile, one value has been never seen.
8862 int expected_val = (false_cnt == 0) ? 1 : 0;
8863
8864 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8865 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8866
8867 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8868 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8869 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8870
8871 { // Slow path: uncommon trap for never seen value and then reexecute
8872 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8873 // the value has been seen at least once.
8874 PreserveJVMState pjvms(this);
8875 PreserveReexecuteState preexecs(this);
8876 jvms()->set_should_reexecute(true);
8877
8878 set_control(slow_path);
8879 set_i_o(i_o());
8880
8881 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8882 Deoptimization::Action_reinterpret);
8883 }
8884 // The guard for never seen value enables sharpening of the result and
8885 // returning a constant. It allows to eliminate branches on the same value
8886 // later on.
8887 set_control(fast_path);
8888 result = intcon(expected_val);
8889 }
8890 // Stop profiling.
8891 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8892 // By replacing method body with profile data (represented as ProfileBooleanNode
8893 // on IR level) we effectively disable profiling.
8894 // It enables full speed execution once optimized code is generated.
8895 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8896 C->record_for_igvn(profile);
8897 set_result(profile);
8898 return true;
8899 } else {
8900 // Continue profiling.
8901 // Profile data isn't available at the moment. So, execute method's bytecode version.
8902 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8903 // is compiled and counters aren't available since corresponding MethodHandle
8904 // isn't a compile-time constant.
8905 return false;
8906 }
8907 }
8908
8909 bool LibraryCallKit::inline_isCompileConstant() {
8910 Node* n = argument(0);
8911 set_result(n->is_Con() ? intcon(1) : intcon(0));
8912 return true;
8913 }
8914
8915 //------------------------------- inline_getObjectSize --------------------------------------
8916 //
8917 // Calculate the runtime size of the object/array.
8918 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
8919 //
8920 bool LibraryCallKit::inline_getObjectSize() {
8921 Node* obj = argument(3);
8922 Node* klass_node = load_object_klass(obj);
8923
8924 jint layout_con = Klass::_lh_neutral_value;
8925 Node* layout_val = get_layout_helper(klass_node, layout_con);
8926 int layout_is_con = (layout_val == nullptr);
8927
8928 if (layout_is_con) {
8929 // Layout helper is constant, can figure out things at compile time.
8930
8931 if (Klass::layout_helper_is_instance(layout_con)) {
8932 // Instance case: layout_con contains the size itself.
8933 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
8934 set_result(size);
8935 } else {
8936 // Array case: size is round(header + element_size*arraylength).
8937 // Since arraylength is different for every array instance, we have to
8938 // compute the whole thing at runtime.
8939
8940 Node* arr_length = load_array_length(obj);
8941
8942 int round_mask = MinObjAlignmentInBytes - 1;
8943 int hsize = Klass::layout_helper_header_size(layout_con);
8944 int eshift = Klass::layout_helper_log2_element_size(layout_con);
8945
8946 if ((round_mask & ~right_n_bits(eshift)) == 0) {
8947 round_mask = 0; // strength-reduce it if it goes away completely
8948 }
8949 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
8950 Node* header_size = intcon(hsize + round_mask);
8951
8952 Node* lengthx = ConvI2X(arr_length);
8953 Node* headerx = ConvI2X(header_size);
8954
8955 Node* abody = lengthx;
8956 if (eshift != 0) {
8957 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
8958 }
8959 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
8960 if (round_mask != 0) {
8961 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
8962 }
8963 size = ConvX2L(size);
8964 set_result(size);
8965 }
8966 } else {
8967 // Layout helper is not constant, need to test for array-ness at runtime.
8968
8969 enum { _instance_path = 1, _array_path, PATH_LIMIT };
8970 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
8971 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
8972 record_for_igvn(result_reg);
8973
8974 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
8975 if (array_ctl != nullptr) {
8976 // Array case: size is round(header + element_size*arraylength).
8977 // Since arraylength is different for every array instance, we have to
8978 // compute the whole thing at runtime.
8979
8980 PreserveJVMState pjvms(this);
8981 set_control(array_ctl);
8982 Node* arr_length = load_array_length(obj);
8983
8984 int round_mask = MinObjAlignmentInBytes - 1;
8985 Node* mask = intcon(round_mask);
8986
8987 Node* hss = intcon(Klass::_lh_header_size_shift);
8988 Node* hsm = intcon(Klass::_lh_header_size_mask);
8989 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
8990 header_size = _gvn.transform(new AndINode(header_size, hsm));
8991 header_size = _gvn.transform(new AddINode(header_size, mask));
8992
8993 // There is no need to mask or shift this value.
8994 // The semantics of LShiftINode include an implicit mask to 0x1F.
8995 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
8996 Node* elem_shift = layout_val;
8997
8998 Node* lengthx = ConvI2X(arr_length);
8999 Node* headerx = ConvI2X(header_size);
9000
9001 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9002 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9003 if (round_mask != 0) {
9004 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9005 }
9006 size = ConvX2L(size);
9007
9008 result_reg->init_req(_array_path, control());
9009 result_val->init_req(_array_path, size);
9010 }
9011
9012 if (!stopped()) {
9013 // Instance case: the layout helper gives us instance size almost directly,
9014 // but we need to mask out the _lh_instance_slow_path_bit.
9015 Node* size = ConvI2X(layout_val);
9016 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9017 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9018 size = _gvn.transform(new AndXNode(size, mask));
9019 size = ConvX2L(size);
9020
9021 result_reg->init_req(_instance_path, control());
9022 result_val->init_req(_instance_path, size);
9023 }
9024
9025 set_result(result_reg, result_val);
9026 }
9027
9028 return true;
9029 }
9030
9031 //------------------------------- inline_blackhole --------------------------------------
9032 //
9033 // Make sure all arguments to this node are alive.
9034 // This matches methods that were requested to be blackholed through compile commands.
9035 //
9036 bool LibraryCallKit::inline_blackhole() {
9037 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9038 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9039 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9040
9041 // Blackhole node pinches only the control, not memory. This allows
9042 // the blackhole to be pinned in the loop that computes blackholed
9043 // values, but have no other side effects, like breaking the optimizations
9044 // across the blackhole.
9045
9046 Node* bh = _gvn.transform(new BlackholeNode(control()));
9047 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9048
9049 // Bind call arguments as blackhole arguments to keep them alive
9050 uint nargs = callee()->arg_size();
9051 for (uint i = 0; i < nargs; i++) {
9052 bh->add_req(argument(i));
9053 }
9054
9055 return true;
9056 }
9057
9058 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9059 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9060 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9061 return nullptr; // box klass is not Float16
9062 }
9063
9064 // Null check; get notnull casted pointer
9065 Node* null_ctl = top();
9066 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9067 // If not_null_box is dead, only null-path is taken
9068 if (stopped()) {
9069 set_control(null_ctl);
9070 return nullptr;
9071 }
9072 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9073 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9074 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9075 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9076 }
9077
9078 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9079 PreserveReexecuteState preexecs(this);
9080 jvms()->set_should_reexecute(true);
9081
9082 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9083 Node* klass_node = makecon(klass_type);
9084 Node* box = new_instance(klass_node);
9085
9086 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9087 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9088
9089 Node* field_store = _gvn.transform(access_store_at(box,
9090 value_field,
9091 value_adr_type,
9092 value,
9093 TypeInt::SHORT,
9094 T_SHORT,
9095 IN_HEAP));
9096 set_memory(field_store, value_adr_type);
9097 return box;
9098 }
9099
9100 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9101 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9102 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9103 return false;
9104 }
9105
9106 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9107 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9108 return false;
9109 }
9110
9111 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9112 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9113 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9114 ciSymbols::short_signature(),
9115 false);
9116 assert(field != nullptr, "");
9117
9118 // Transformed nodes
9119 Node* fld1 = nullptr;
9120 Node* fld2 = nullptr;
9121 Node* fld3 = nullptr;
9122 switch(num_args) {
9123 case 3:
9124 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9125 if (fld3 == nullptr) {
9126 return false;
9127 }
9128 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9129 // fall-through
9130 case 2:
9131 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9132 if (fld2 == nullptr) {
9133 return false;
9134 }
9135 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9136 // fall-through
9137 case 1:
9138 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9139 if (fld1 == nullptr) {
9140 return false;
9141 }
9142 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9143 break;
9144 default: fatal("Unsupported number of arguments %d", num_args);
9145 }
9146
9147 Node* result = nullptr;
9148 switch (id) {
9149 // Unary operations
9150 case vmIntrinsics::_sqrt_float16:
9151 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9152 break;
9153 // Ternary operations
9154 case vmIntrinsics::_fma_float16:
9155 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9156 break;
9157 default:
9158 fatal_unexpected_iid(id);
9159 break;
9160 }
9161 result = _gvn.transform(new ReinterpretHF2SNode(result));
9162 set_result(box_fp16_value(float16_box_type, field, result));
9163 return true;
9164 }
9165