1 /*
2 * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "asm/macroAssembler.hpp"
26 #include "ci/ciUtilities.inline.hpp"
27 #include "ci/ciSymbols.hpp"
28 #include "classfile/vmIntrinsics.hpp"
29 #include "compiler/compileBroker.hpp"
30 #include "compiler/compileLog.hpp"
31 #include "gc/shared/barrierSet.hpp"
32 #include "jfr/support/jfrIntrinsics.hpp"
33 #include "memory/resourceArea.hpp"
34 #include "oops/klass.inline.hpp"
35 #include "oops/objArrayKlass.hpp"
36 #include "opto/addnode.hpp"
37 #include "opto/arraycopynode.hpp"
38 #include "opto/c2compiler.hpp"
39 #include "opto/castnode.hpp"
40 #include "opto/cfgnode.hpp"
41 #include "opto/convertnode.hpp"
42 #include "opto/countbitsnode.hpp"
43 #include "opto/idealKit.hpp"
44 #include "opto/library_call.hpp"
45 #include "opto/mathexactnode.hpp"
46 #include "opto/mulnode.hpp"
47 #include "opto/narrowptrnode.hpp"
48 #include "opto/opaquenode.hpp"
49 #include "opto/parse.hpp"
50 #include "opto/runtime.hpp"
51 #include "opto/rootnode.hpp"
52 #include "opto/subnode.hpp"
53 #include "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/objectMonitor.hpp"
59 #include "runtime/sharedRuntime.hpp"
60 #include "runtime/stubRoutines.hpp"
61 #include "utilities/macros.hpp"
62 #include "utilities/powerOfTwo.hpp"
63
64 //---------------------------make_vm_intrinsic----------------------------
65 CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
66 vmIntrinsicID id = m->intrinsic_id();
67 assert(id != vmIntrinsics::_none, "must be a VM intrinsic");
68
69 if (!m->is_loaded()) {
70 // Do not attempt to inline unloaded methods.
71 return nullptr;
72 }
73
74 C2Compiler* compiler = (C2Compiler*)CompileBroker::compiler(CompLevel_full_optimization);
75 bool is_available = false;
76
77 {
78 // For calling is_intrinsic_supported and is_intrinsic_disabled_by_flag
79 // the compiler must transition to '_thread_in_vm' state because both
80 // methods access VM-internal data.
81 VM_ENTRY_MARK;
82 methodHandle mh(THREAD, m->get_Method());
83 is_available = compiler != nullptr && compiler->is_intrinsic_available(mh, C->directive());
84 if (is_available && is_virtual) {
85 is_available = vmIntrinsics::does_virtual_dispatch(id);
86 }
87 }
88
89 if (is_available) {
90 assert(id <= vmIntrinsics::LAST_COMPILER_INLINE, "caller responsibility");
91 assert(id != vmIntrinsics::_Object_init && id != vmIntrinsics::_invoke, "enum out of order?");
92 return new LibraryIntrinsic(m, is_virtual,
93 vmIntrinsics::predicates_needed(id),
94 vmIntrinsics::does_virtual_dispatch(id),
95 id);
96 } else {
97 return nullptr;
98 }
99 }
100
101 JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
102 LibraryCallKit kit(jvms, this);
103 Compile* C = kit.C;
104 int nodes = C->unique();
105 #ifndef PRODUCT
106 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
107 char buf[1000];
108 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
109 tty->print_cr("Intrinsic %s", str);
110 }
111 #endif
112 ciMethod* callee = kit.callee();
113 const int bci = kit.bci();
114 #ifdef ASSERT
115 Node* ctrl = kit.control();
116 #endif
117 // Try to inline the intrinsic.
118 if (callee->check_intrinsic_candidate() &&
119 kit.try_to_inline(_last_predicate)) {
120 const char *inline_msg = is_virtual() ? "(intrinsic, virtual)"
121 : "(intrinsic)";
122 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
123 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
124 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
125 if (C->log()) {
126 C->log()->elem("intrinsic id='%s'%s nodes='%d'",
127 vmIntrinsics::name_at(intrinsic_id()),
128 (is_virtual() ? " virtual='1'" : ""),
129 C->unique() - nodes);
130 }
131 // Push the result from the inlined method onto the stack.
132 kit.push_result();
133 return kit.transfer_exceptions_into_jvms();
134 }
135
136 // The intrinsic bailed out
137 assert(ctrl == kit.control(), "Control flow was added although the intrinsic bailed out");
138 if (jvms->has_method()) {
139 // Not a root compile.
140 const char* msg;
141 if (callee->intrinsic_candidate()) {
142 msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
143 } else {
144 msg = is_virtual() ? "failed to inline (intrinsic, virtual), method not annotated"
145 : "failed to inline (intrinsic), method not annotated";
146 }
147 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
148 C->inline_printer()->record(callee, jvms, InliningResult::FAILURE, msg);
149 } else {
150 // Root compile
151 ResourceMark rm;
152 stringStream msg_stream;
153 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
154 vmIntrinsics::name_at(intrinsic_id()),
155 is_virtual() ? " (virtual)" : "", bci);
156 const char *msg = msg_stream.freeze();
157 log_debug(jit, inlining)("%s", msg);
158 if (C->print_intrinsics() || C->print_inlining()) {
159 tty->print("%s", msg);
160 }
161 }
162 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
163
164 return nullptr;
165 }
166
167 Node* LibraryIntrinsic::generate_predicate(JVMState* jvms, int predicate) {
168 LibraryCallKit kit(jvms, this);
169 Compile* C = kit.C;
170 int nodes = C->unique();
171 _last_predicate = predicate;
172 #ifndef PRODUCT
173 assert(is_predicated() && predicate < predicates_count(), "sanity");
174 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
175 char buf[1000];
176 const char* str = vmIntrinsics::short_name_as_C_string(intrinsic_id(), buf, sizeof(buf));
177 tty->print_cr("Predicate for intrinsic %s", str);
178 }
179 #endif
180 ciMethod* callee = kit.callee();
181 const int bci = kit.bci();
182
183 Node* slow_ctl = kit.try_to_predicate(predicate);
184 if (!kit.failing()) {
185 const char *inline_msg = is_virtual() ? "(intrinsic, virtual, predicate)"
186 : "(intrinsic, predicate)";
187 CompileTask::print_inlining_ul(callee, jvms->depth() - 1, bci, InliningResult::SUCCESS, inline_msg);
188 C->inline_printer()->record(callee, jvms, InliningResult::SUCCESS, inline_msg);
189
190 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_worked);
191 if (C->log()) {
192 C->log()->elem("predicate_intrinsic id='%s'%s nodes='%d'",
193 vmIntrinsics::name_at(intrinsic_id()),
194 (is_virtual() ? " virtual='1'" : ""),
195 C->unique() - nodes);
196 }
197 return slow_ctl; // Could be null if the check folds.
198 }
199
200 // The intrinsic bailed out
201 if (jvms->has_method()) {
202 // Not a root compile.
203 const char* msg = "failed to generate predicate for intrinsic";
204 CompileTask::print_inlining_ul(kit.callee(), jvms->depth() - 1, bci, InliningResult::FAILURE, msg);
205 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
206 } else {
207 // Root compile
208 ResourceMark rm;
209 stringStream msg_stream;
210 msg_stream.print("Did not generate intrinsic %s%s at bci:%d in",
211 vmIntrinsics::name_at(intrinsic_id()),
212 is_virtual() ? " (virtual)" : "", bci);
213 const char *msg = msg_stream.freeze();
214 log_debug(jit, inlining)("%s", msg);
215 C->inline_printer()->record(kit.callee(), jvms, InliningResult::FAILURE, msg);
216 }
217 C->gather_intrinsic_statistics(intrinsic_id(), is_virtual(), Compile::_intrinsic_failed);
218 return nullptr;
219 }
220
221 bool LibraryCallKit::try_to_inline(int predicate) {
222 // Handle symbolic names for otherwise undistinguished boolean switches:
223 const bool is_store = true;
224 const bool is_compress = true;
225 const bool is_static = true;
226 const bool is_volatile = true;
227
228 if (!jvms()->has_method()) {
229 // Root JVMState has a null method.
230 assert(map()->memory()->Opcode() == Op_Parm, "");
231 // Insert the memory aliasing node
232 set_all_memory(reset_memory());
233 }
234 assert(merged_memory(), "");
235
236 switch (intrinsic_id()) {
237 case vmIntrinsics::_hashCode: return inline_native_hashcode(intrinsic()->is_virtual(), !is_static);
238 case vmIntrinsics::_identityHashCode: return inline_native_hashcode(/*!virtual*/ false, is_static);
239 case vmIntrinsics::_getClass: return inline_native_getClass();
240
241 case vmIntrinsics::_ceil:
242 case vmIntrinsics::_floor:
243 case vmIntrinsics::_rint:
244 case vmIntrinsics::_dsin:
245 case vmIntrinsics::_dcos:
246 case vmIntrinsics::_dtan:
247 case vmIntrinsics::_dtanh:
248 case vmIntrinsics::_dabs:
249 case vmIntrinsics::_fabs:
250 case vmIntrinsics::_iabs:
251 case vmIntrinsics::_labs:
252 case vmIntrinsics::_datan2:
253 case vmIntrinsics::_dsqrt:
254 case vmIntrinsics::_dsqrt_strict:
255 case vmIntrinsics::_dexp:
256 case vmIntrinsics::_dlog:
257 case vmIntrinsics::_dlog10:
258 case vmIntrinsics::_dpow:
259 case vmIntrinsics::_dcopySign:
260 case vmIntrinsics::_fcopySign:
261 case vmIntrinsics::_dsignum:
262 case vmIntrinsics::_roundF:
263 case vmIntrinsics::_roundD:
264 case vmIntrinsics::_fsignum: return inline_math_native(intrinsic_id());
265
266 case vmIntrinsics::_notify:
267 case vmIntrinsics::_notifyAll:
268 return inline_notify(intrinsic_id());
269
270 case vmIntrinsics::_addExactI: return inline_math_addExactI(false /* add */);
271 case vmIntrinsics::_addExactL: return inline_math_addExactL(false /* add */);
272 case vmIntrinsics::_decrementExactI: return inline_math_subtractExactI(true /* decrement */);
273 case vmIntrinsics::_decrementExactL: return inline_math_subtractExactL(true /* decrement */);
274 case vmIntrinsics::_incrementExactI: return inline_math_addExactI(true /* increment */);
275 case vmIntrinsics::_incrementExactL: return inline_math_addExactL(true /* increment */);
276 case vmIntrinsics::_multiplyExactI: return inline_math_multiplyExactI();
277 case vmIntrinsics::_multiplyExactL: return inline_math_multiplyExactL();
278 case vmIntrinsics::_multiplyHigh: return inline_math_multiplyHigh();
279 case vmIntrinsics::_unsignedMultiplyHigh: return inline_math_unsignedMultiplyHigh();
280 case vmIntrinsics::_negateExactI: return inline_math_negateExactI();
281 case vmIntrinsics::_negateExactL: return inline_math_negateExactL();
282 case vmIntrinsics::_subtractExactI: return inline_math_subtractExactI(false /* subtract */);
283 case vmIntrinsics::_subtractExactL: return inline_math_subtractExactL(false /* subtract */);
284
285 case vmIntrinsics::_arraycopy: return inline_arraycopy();
286
287 case vmIntrinsics::_arraySort: return inline_array_sort();
288 case vmIntrinsics::_arrayPartition: return inline_array_partition();
289
290 case vmIntrinsics::_compareToL: return inline_string_compareTo(StrIntrinsicNode::LL);
291 case vmIntrinsics::_compareToU: return inline_string_compareTo(StrIntrinsicNode::UU);
292 case vmIntrinsics::_compareToLU: return inline_string_compareTo(StrIntrinsicNode::LU);
293 case vmIntrinsics::_compareToUL: return inline_string_compareTo(StrIntrinsicNode::UL);
294
295 case vmIntrinsics::_indexOfL: return inline_string_indexOf(StrIntrinsicNode::LL);
296 case vmIntrinsics::_indexOfU: return inline_string_indexOf(StrIntrinsicNode::UU);
297 case vmIntrinsics::_indexOfUL: return inline_string_indexOf(StrIntrinsicNode::UL);
298 case vmIntrinsics::_indexOfIL: return inline_string_indexOfI(StrIntrinsicNode::LL);
299 case vmIntrinsics::_indexOfIU: return inline_string_indexOfI(StrIntrinsicNode::UU);
300 case vmIntrinsics::_indexOfIUL: return inline_string_indexOfI(StrIntrinsicNode::UL);
301 case vmIntrinsics::_indexOfU_char: return inline_string_indexOfChar(StrIntrinsicNode::U);
302 case vmIntrinsics::_indexOfL_char: return inline_string_indexOfChar(StrIntrinsicNode::L);
303
304 case vmIntrinsics::_equalsL: return inline_string_equals(StrIntrinsicNode::LL);
305
306 case vmIntrinsics::_vectorizedHashCode: return inline_vectorizedHashCode();
307
308 case vmIntrinsics::_toBytesStringU: return inline_string_toBytesU();
309 case vmIntrinsics::_getCharsStringU: return inline_string_getCharsU();
310 case vmIntrinsics::_getCharStringU: return inline_string_char_access(!is_store);
311 case vmIntrinsics::_putCharStringU: return inline_string_char_access( is_store);
312
313 case vmIntrinsics::_compressStringC:
314 case vmIntrinsics::_compressStringB: return inline_string_copy( is_compress);
315 case vmIntrinsics::_inflateStringC:
316 case vmIntrinsics::_inflateStringB: return inline_string_copy(!is_compress);
317
318 case vmIntrinsics::_getReference: return inline_unsafe_access(!is_store, T_OBJECT, Relaxed, false);
319 case vmIntrinsics::_getBoolean: return inline_unsafe_access(!is_store, T_BOOLEAN, Relaxed, false);
320 case vmIntrinsics::_getByte: return inline_unsafe_access(!is_store, T_BYTE, Relaxed, false);
321 case vmIntrinsics::_getShort: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, false);
322 case vmIntrinsics::_getChar: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, false);
323 case vmIntrinsics::_getInt: return inline_unsafe_access(!is_store, T_INT, Relaxed, false);
324 case vmIntrinsics::_getLong: return inline_unsafe_access(!is_store, T_LONG, Relaxed, false);
325 case vmIntrinsics::_getFloat: return inline_unsafe_access(!is_store, T_FLOAT, Relaxed, false);
326 case vmIntrinsics::_getDouble: return inline_unsafe_access(!is_store, T_DOUBLE, Relaxed, false);
327
328 case vmIntrinsics::_putReference: return inline_unsafe_access( is_store, T_OBJECT, Relaxed, false);
329 case vmIntrinsics::_putBoolean: return inline_unsafe_access( is_store, T_BOOLEAN, Relaxed, false);
330 case vmIntrinsics::_putByte: return inline_unsafe_access( is_store, T_BYTE, Relaxed, false);
331 case vmIntrinsics::_putShort: return inline_unsafe_access( is_store, T_SHORT, Relaxed, false);
332 case vmIntrinsics::_putChar: return inline_unsafe_access( is_store, T_CHAR, Relaxed, false);
333 case vmIntrinsics::_putInt: return inline_unsafe_access( is_store, T_INT, Relaxed, false);
334 case vmIntrinsics::_putLong: return inline_unsafe_access( is_store, T_LONG, Relaxed, false);
335 case vmIntrinsics::_putFloat: return inline_unsafe_access( is_store, T_FLOAT, Relaxed, false);
336 case vmIntrinsics::_putDouble: return inline_unsafe_access( is_store, T_DOUBLE, Relaxed, false);
337
338 case vmIntrinsics::_getReferenceVolatile: return inline_unsafe_access(!is_store, T_OBJECT, Volatile, false);
339 case vmIntrinsics::_getBooleanVolatile: return inline_unsafe_access(!is_store, T_BOOLEAN, Volatile, false);
340 case vmIntrinsics::_getByteVolatile: return inline_unsafe_access(!is_store, T_BYTE, Volatile, false);
341 case vmIntrinsics::_getShortVolatile: return inline_unsafe_access(!is_store, T_SHORT, Volatile, false);
342 case vmIntrinsics::_getCharVolatile: return inline_unsafe_access(!is_store, T_CHAR, Volatile, false);
343 case vmIntrinsics::_getIntVolatile: return inline_unsafe_access(!is_store, T_INT, Volatile, false);
344 case vmIntrinsics::_getLongVolatile: return inline_unsafe_access(!is_store, T_LONG, Volatile, false);
345 case vmIntrinsics::_getFloatVolatile: return inline_unsafe_access(!is_store, T_FLOAT, Volatile, false);
346 case vmIntrinsics::_getDoubleVolatile: return inline_unsafe_access(!is_store, T_DOUBLE, Volatile, false);
347
348 case vmIntrinsics::_putReferenceVolatile: return inline_unsafe_access( is_store, T_OBJECT, Volatile, false);
349 case vmIntrinsics::_putBooleanVolatile: return inline_unsafe_access( is_store, T_BOOLEAN, Volatile, false);
350 case vmIntrinsics::_putByteVolatile: return inline_unsafe_access( is_store, T_BYTE, Volatile, false);
351 case vmIntrinsics::_putShortVolatile: return inline_unsafe_access( is_store, T_SHORT, Volatile, false);
352 case vmIntrinsics::_putCharVolatile: return inline_unsafe_access( is_store, T_CHAR, Volatile, false);
353 case vmIntrinsics::_putIntVolatile: return inline_unsafe_access( is_store, T_INT, Volatile, false);
354 case vmIntrinsics::_putLongVolatile: return inline_unsafe_access( is_store, T_LONG, Volatile, false);
355 case vmIntrinsics::_putFloatVolatile: return inline_unsafe_access( is_store, T_FLOAT, Volatile, false);
356 case vmIntrinsics::_putDoubleVolatile: return inline_unsafe_access( is_store, T_DOUBLE, Volatile, false);
357
358 case vmIntrinsics::_getShortUnaligned: return inline_unsafe_access(!is_store, T_SHORT, Relaxed, true);
359 case vmIntrinsics::_getCharUnaligned: return inline_unsafe_access(!is_store, T_CHAR, Relaxed, true);
360 case vmIntrinsics::_getIntUnaligned: return inline_unsafe_access(!is_store, T_INT, Relaxed, true);
361 case vmIntrinsics::_getLongUnaligned: return inline_unsafe_access(!is_store, T_LONG, Relaxed, true);
362
363 case vmIntrinsics::_putShortUnaligned: return inline_unsafe_access( is_store, T_SHORT, Relaxed, true);
364 case vmIntrinsics::_putCharUnaligned: return inline_unsafe_access( is_store, T_CHAR, Relaxed, true);
365 case vmIntrinsics::_putIntUnaligned: return inline_unsafe_access( is_store, T_INT, Relaxed, true);
366 case vmIntrinsics::_putLongUnaligned: return inline_unsafe_access( is_store, T_LONG, Relaxed, true);
367
368 case vmIntrinsics::_getReferenceAcquire: return inline_unsafe_access(!is_store, T_OBJECT, Acquire, false);
369 case vmIntrinsics::_getBooleanAcquire: return inline_unsafe_access(!is_store, T_BOOLEAN, Acquire, false);
370 case vmIntrinsics::_getByteAcquire: return inline_unsafe_access(!is_store, T_BYTE, Acquire, false);
371 case vmIntrinsics::_getShortAcquire: return inline_unsafe_access(!is_store, T_SHORT, Acquire, false);
372 case vmIntrinsics::_getCharAcquire: return inline_unsafe_access(!is_store, T_CHAR, Acquire, false);
373 case vmIntrinsics::_getIntAcquire: return inline_unsafe_access(!is_store, T_INT, Acquire, false);
374 case vmIntrinsics::_getLongAcquire: return inline_unsafe_access(!is_store, T_LONG, Acquire, false);
375 case vmIntrinsics::_getFloatAcquire: return inline_unsafe_access(!is_store, T_FLOAT, Acquire, false);
376 case vmIntrinsics::_getDoubleAcquire: return inline_unsafe_access(!is_store, T_DOUBLE, Acquire, false);
377
378 case vmIntrinsics::_putReferenceRelease: return inline_unsafe_access( is_store, T_OBJECT, Release, false);
379 case vmIntrinsics::_putBooleanRelease: return inline_unsafe_access( is_store, T_BOOLEAN, Release, false);
380 case vmIntrinsics::_putByteRelease: return inline_unsafe_access( is_store, T_BYTE, Release, false);
381 case vmIntrinsics::_putShortRelease: return inline_unsafe_access( is_store, T_SHORT, Release, false);
382 case vmIntrinsics::_putCharRelease: return inline_unsafe_access( is_store, T_CHAR, Release, false);
383 case vmIntrinsics::_putIntRelease: return inline_unsafe_access( is_store, T_INT, Release, false);
384 case vmIntrinsics::_putLongRelease: return inline_unsafe_access( is_store, T_LONG, Release, false);
385 case vmIntrinsics::_putFloatRelease: return inline_unsafe_access( is_store, T_FLOAT, Release, false);
386 case vmIntrinsics::_putDoubleRelease: return inline_unsafe_access( is_store, T_DOUBLE, Release, false);
387
388 case vmIntrinsics::_getReferenceOpaque: return inline_unsafe_access(!is_store, T_OBJECT, Opaque, false);
389 case vmIntrinsics::_getBooleanOpaque: return inline_unsafe_access(!is_store, T_BOOLEAN, Opaque, false);
390 case vmIntrinsics::_getByteOpaque: return inline_unsafe_access(!is_store, T_BYTE, Opaque, false);
391 case vmIntrinsics::_getShortOpaque: return inline_unsafe_access(!is_store, T_SHORT, Opaque, false);
392 case vmIntrinsics::_getCharOpaque: return inline_unsafe_access(!is_store, T_CHAR, Opaque, false);
393 case vmIntrinsics::_getIntOpaque: return inline_unsafe_access(!is_store, T_INT, Opaque, false);
394 case vmIntrinsics::_getLongOpaque: return inline_unsafe_access(!is_store, T_LONG, Opaque, false);
395 case vmIntrinsics::_getFloatOpaque: return inline_unsafe_access(!is_store, T_FLOAT, Opaque, false);
396 case vmIntrinsics::_getDoubleOpaque: return inline_unsafe_access(!is_store, T_DOUBLE, Opaque, false);
397
398 case vmIntrinsics::_putReferenceOpaque: return inline_unsafe_access( is_store, T_OBJECT, Opaque, false);
399 case vmIntrinsics::_putBooleanOpaque: return inline_unsafe_access( is_store, T_BOOLEAN, Opaque, false);
400 case vmIntrinsics::_putByteOpaque: return inline_unsafe_access( is_store, T_BYTE, Opaque, false);
401 case vmIntrinsics::_putShortOpaque: return inline_unsafe_access( is_store, T_SHORT, Opaque, false);
402 case vmIntrinsics::_putCharOpaque: return inline_unsafe_access( is_store, T_CHAR, Opaque, false);
403 case vmIntrinsics::_putIntOpaque: return inline_unsafe_access( is_store, T_INT, Opaque, false);
404 case vmIntrinsics::_putLongOpaque: return inline_unsafe_access( is_store, T_LONG, Opaque, false);
405 case vmIntrinsics::_putFloatOpaque: return inline_unsafe_access( is_store, T_FLOAT, Opaque, false);
406 case vmIntrinsics::_putDoubleOpaque: return inline_unsafe_access( is_store, T_DOUBLE, Opaque, false);
407
408 case vmIntrinsics::_compareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap, Volatile);
409 case vmIntrinsics::_compareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap, Volatile);
410 case vmIntrinsics::_compareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap, Volatile);
411 case vmIntrinsics::_compareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap, Volatile);
412 case vmIntrinsics::_compareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap, Volatile);
413
414 case vmIntrinsics::_weakCompareAndSetReferencePlain: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Relaxed);
415 case vmIntrinsics::_weakCompareAndSetReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Acquire);
416 case vmIntrinsics::_weakCompareAndSetReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Release);
417 case vmIntrinsics::_weakCompareAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_swap_weak, Volatile);
418 case vmIntrinsics::_weakCompareAndSetBytePlain: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Relaxed);
419 case vmIntrinsics::_weakCompareAndSetByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Acquire);
420 case vmIntrinsics::_weakCompareAndSetByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Release);
421 case vmIntrinsics::_weakCompareAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_swap_weak, Volatile);
422 case vmIntrinsics::_weakCompareAndSetShortPlain: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Relaxed);
423 case vmIntrinsics::_weakCompareAndSetShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Acquire);
424 case vmIntrinsics::_weakCompareAndSetShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Release);
425 case vmIntrinsics::_weakCompareAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_swap_weak, Volatile);
426 case vmIntrinsics::_weakCompareAndSetIntPlain: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Relaxed);
427 case vmIntrinsics::_weakCompareAndSetIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Acquire);
428 case vmIntrinsics::_weakCompareAndSetIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Release);
429 case vmIntrinsics::_weakCompareAndSetInt: return inline_unsafe_load_store(T_INT, LS_cmp_swap_weak, Volatile);
430 case vmIntrinsics::_weakCompareAndSetLongPlain: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Relaxed);
431 case vmIntrinsics::_weakCompareAndSetLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Acquire);
432 case vmIntrinsics::_weakCompareAndSetLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Release);
433 case vmIntrinsics::_weakCompareAndSetLong: return inline_unsafe_load_store(T_LONG, LS_cmp_swap_weak, Volatile);
434
435 case vmIntrinsics::_compareAndExchangeReference: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Volatile);
436 case vmIntrinsics::_compareAndExchangeReferenceAcquire: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Acquire);
437 case vmIntrinsics::_compareAndExchangeReferenceRelease: return inline_unsafe_load_store(T_OBJECT, LS_cmp_exchange, Release);
438 case vmIntrinsics::_compareAndExchangeByte: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Volatile);
439 case vmIntrinsics::_compareAndExchangeByteAcquire: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Acquire);
440 case vmIntrinsics::_compareAndExchangeByteRelease: return inline_unsafe_load_store(T_BYTE, LS_cmp_exchange, Release);
441 case vmIntrinsics::_compareAndExchangeShort: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Volatile);
442 case vmIntrinsics::_compareAndExchangeShortAcquire: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Acquire);
443 case vmIntrinsics::_compareAndExchangeShortRelease: return inline_unsafe_load_store(T_SHORT, LS_cmp_exchange, Release);
444 case vmIntrinsics::_compareAndExchangeInt: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Volatile);
445 case vmIntrinsics::_compareAndExchangeIntAcquire: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Acquire);
446 case vmIntrinsics::_compareAndExchangeIntRelease: return inline_unsafe_load_store(T_INT, LS_cmp_exchange, Release);
447 case vmIntrinsics::_compareAndExchangeLong: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Volatile);
448 case vmIntrinsics::_compareAndExchangeLongAcquire: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Acquire);
449 case vmIntrinsics::_compareAndExchangeLongRelease: return inline_unsafe_load_store(T_LONG, LS_cmp_exchange, Release);
450
451 case vmIntrinsics::_getAndAddByte: return inline_unsafe_load_store(T_BYTE, LS_get_add, Volatile);
452 case vmIntrinsics::_getAndAddShort: return inline_unsafe_load_store(T_SHORT, LS_get_add, Volatile);
453 case vmIntrinsics::_getAndAddInt: return inline_unsafe_load_store(T_INT, LS_get_add, Volatile);
454 case vmIntrinsics::_getAndAddLong: return inline_unsafe_load_store(T_LONG, LS_get_add, Volatile);
455
456 case vmIntrinsics::_getAndSetByte: return inline_unsafe_load_store(T_BYTE, LS_get_set, Volatile);
457 case vmIntrinsics::_getAndSetShort: return inline_unsafe_load_store(T_SHORT, LS_get_set, Volatile);
458 case vmIntrinsics::_getAndSetInt: return inline_unsafe_load_store(T_INT, LS_get_set, Volatile);
459 case vmIntrinsics::_getAndSetLong: return inline_unsafe_load_store(T_LONG, LS_get_set, Volatile);
460 case vmIntrinsics::_getAndSetReference: return inline_unsafe_load_store(T_OBJECT, LS_get_set, Volatile);
461
462 case vmIntrinsics::_loadFence:
463 case vmIntrinsics::_storeFence:
464 case vmIntrinsics::_storeStoreFence:
465 case vmIntrinsics::_fullFence: return inline_unsafe_fence(intrinsic_id());
466
467 case vmIntrinsics::_onSpinWait: return inline_onspinwait();
468
469 case vmIntrinsics::_currentCarrierThread: return inline_native_currentCarrierThread();
470 case vmIntrinsics::_currentThread: return inline_native_currentThread();
471 case vmIntrinsics::_setCurrentThread: return inline_native_setCurrentThread();
472
473 case vmIntrinsics::_scopedValueCache: return inline_native_scopedValueCache();
474 case vmIntrinsics::_setScopedValueCache: return inline_native_setScopedValueCache();
475
476 case vmIntrinsics::_Continuation_pin: return inline_native_Continuation_pinning(false);
477 case vmIntrinsics::_Continuation_unpin: return inline_native_Continuation_pinning(true);
478
479 #if INCLUDE_JVMTI
480 case vmIntrinsics::_notifyJvmtiVThreadStart: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_start()),
481 "notifyJvmtiStart", true, false);
482 case vmIntrinsics::_notifyJvmtiVThreadEnd: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_end()),
483 "notifyJvmtiEnd", false, true);
484 case vmIntrinsics::_notifyJvmtiVThreadMount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_mount()),
485 "notifyJvmtiMount", false, false);
486 case vmIntrinsics::_notifyJvmtiVThreadUnmount: return inline_native_notify_jvmti_funcs(CAST_FROM_FN_PTR(address, OptoRuntime::notify_jvmti_vthread_unmount()),
487 "notifyJvmtiUnmount", false, false);
488 case vmIntrinsics::_notifyJvmtiVThreadDisableSuspend: return inline_native_notify_jvmti_sync();
489 #endif
490
491 #ifdef JFR_HAVE_INTRINSICS
492 case vmIntrinsics::_counterTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, JfrTime::time_function()), "counterTime");
493 case vmIntrinsics::_getEventWriter: return inline_native_getEventWriter();
494 case vmIntrinsics::_jvm_commit: return inline_native_jvm_commit();
495 #endif
496 case vmIntrinsics::_currentTimeMillis: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeMillis), "currentTimeMillis");
497 case vmIntrinsics::_nanoTime: return inline_native_time_funcs(CAST_FROM_FN_PTR(address, os::javaTimeNanos), "nanoTime");
498 case vmIntrinsics::_writeback0: return inline_unsafe_writeback0();
499 case vmIntrinsics::_writebackPreSync0: return inline_unsafe_writebackSync0(true);
500 case vmIntrinsics::_writebackPostSync0: return inline_unsafe_writebackSync0(false);
501 case vmIntrinsics::_allocateInstance: return inline_unsafe_allocate();
502 case vmIntrinsics::_copyMemory: return inline_unsafe_copyMemory();
503 case vmIntrinsics::_setMemory: return inline_unsafe_setMemory();
504 case vmIntrinsics::_getLength: return inline_native_getLength();
505 case vmIntrinsics::_copyOf: return inline_array_copyOf(false);
506 case vmIntrinsics::_copyOfRange: return inline_array_copyOf(true);
507 case vmIntrinsics::_equalsB: return inline_array_equals(StrIntrinsicNode::LL);
508 case vmIntrinsics::_equalsC: return inline_array_equals(StrIntrinsicNode::UU);
509 case vmIntrinsics::_Preconditions_checkIndex: return inline_preconditions_checkIndex(T_INT);
510 case vmIntrinsics::_Preconditions_checkLongIndex: return inline_preconditions_checkIndex(T_LONG);
511 case vmIntrinsics::_clone: return inline_native_clone(intrinsic()->is_virtual());
512
513 case vmIntrinsics::_allocateUninitializedArray: return inline_unsafe_newArray(true);
514 case vmIntrinsics::_newArray: return inline_unsafe_newArray(false);
515
516 case vmIntrinsics::_isAssignableFrom: return inline_native_subtype_check();
517
518 case vmIntrinsics::_isInstance:
519 case vmIntrinsics::_isHidden:
520 case vmIntrinsics::_getSuperclass:
521 case vmIntrinsics::_getClassAccessFlags: return inline_native_Class_query(intrinsic_id());
522
523 case vmIntrinsics::_floatToRawIntBits:
524 case vmIntrinsics::_floatToIntBits:
525 case vmIntrinsics::_intBitsToFloat:
526 case vmIntrinsics::_doubleToRawLongBits:
527 case vmIntrinsics::_doubleToLongBits:
528 case vmIntrinsics::_longBitsToDouble:
529 case vmIntrinsics::_floatToFloat16:
530 case vmIntrinsics::_float16ToFloat: return inline_fp_conversions(intrinsic_id());
531 case vmIntrinsics::_sqrt_float16: return inline_fp16_operations(intrinsic_id(), 1);
532 case vmIntrinsics::_fma_float16: return inline_fp16_operations(intrinsic_id(), 3);
533 case vmIntrinsics::_floatIsFinite:
534 case vmIntrinsics::_floatIsInfinite:
535 case vmIntrinsics::_doubleIsFinite:
536 case vmIntrinsics::_doubleIsInfinite: return inline_fp_range_check(intrinsic_id());
537
538 case vmIntrinsics::_numberOfLeadingZeros_i:
539 case vmIntrinsics::_numberOfLeadingZeros_l:
540 case vmIntrinsics::_numberOfTrailingZeros_i:
541 case vmIntrinsics::_numberOfTrailingZeros_l:
542 case vmIntrinsics::_bitCount_i:
543 case vmIntrinsics::_bitCount_l:
544 case vmIntrinsics::_reverse_i:
545 case vmIntrinsics::_reverse_l:
546 case vmIntrinsics::_reverseBytes_i:
547 case vmIntrinsics::_reverseBytes_l:
548 case vmIntrinsics::_reverseBytes_s:
549 case vmIntrinsics::_reverseBytes_c: return inline_number_methods(intrinsic_id());
550
551 case vmIntrinsics::_compress_i:
552 case vmIntrinsics::_compress_l:
553 case vmIntrinsics::_expand_i:
554 case vmIntrinsics::_expand_l: return inline_bitshuffle_methods(intrinsic_id());
555
556 case vmIntrinsics::_compareUnsigned_i:
557 case vmIntrinsics::_compareUnsigned_l: return inline_compare_unsigned(intrinsic_id());
558
559 case vmIntrinsics::_divideUnsigned_i:
560 case vmIntrinsics::_divideUnsigned_l:
561 case vmIntrinsics::_remainderUnsigned_i:
562 case vmIntrinsics::_remainderUnsigned_l: return inline_divmod_methods(intrinsic_id());
563
564 case vmIntrinsics::_getCallerClass: return inline_native_Reflection_getCallerClass();
565
566 case vmIntrinsics::_Reference_get: return inline_reference_get();
567 case vmIntrinsics::_Reference_refersTo0: return inline_reference_refersTo0(false);
568 case vmIntrinsics::_PhantomReference_refersTo0: return inline_reference_refersTo0(true);
569 case vmIntrinsics::_Reference_clear0: return inline_reference_clear0(false);
570 case vmIntrinsics::_PhantomReference_clear0: return inline_reference_clear0(true);
571
572 case vmIntrinsics::_Class_cast: return inline_Class_cast();
573
574 case vmIntrinsics::_aescrypt_encryptBlock:
575 case vmIntrinsics::_aescrypt_decryptBlock: return inline_aescrypt_Block(intrinsic_id());
576
577 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
578 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
579 return inline_cipherBlockChaining_AESCrypt(intrinsic_id());
580
581 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
582 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
583 return inline_electronicCodeBook_AESCrypt(intrinsic_id());
584
585 case vmIntrinsics::_counterMode_AESCrypt:
586 return inline_counterMode_AESCrypt(intrinsic_id());
587
588 case vmIntrinsics::_galoisCounterMode_AESCrypt:
589 return inline_galoisCounterMode_AESCrypt();
590
591 case vmIntrinsics::_md5_implCompress:
592 case vmIntrinsics::_sha_implCompress:
593 case vmIntrinsics::_sha2_implCompress:
594 case vmIntrinsics::_sha5_implCompress:
595 case vmIntrinsics::_sha3_implCompress:
596 return inline_digestBase_implCompress(intrinsic_id());
597 case vmIntrinsics::_double_keccak:
598 return inline_double_keccak();
599
600 case vmIntrinsics::_digestBase_implCompressMB:
601 return inline_digestBase_implCompressMB(predicate);
602
603 case vmIntrinsics::_multiplyToLen:
604 return inline_multiplyToLen();
605
606 case vmIntrinsics::_squareToLen:
607 return inline_squareToLen();
608
609 case vmIntrinsics::_mulAdd:
610 return inline_mulAdd();
611
612 case vmIntrinsics::_montgomeryMultiply:
613 return inline_montgomeryMultiply();
614 case vmIntrinsics::_montgomerySquare:
615 return inline_montgomerySquare();
616
617 case vmIntrinsics::_bigIntegerRightShiftWorker:
618 return inline_bigIntegerShift(true);
619 case vmIntrinsics::_bigIntegerLeftShiftWorker:
620 return inline_bigIntegerShift(false);
621
622 case vmIntrinsics::_vectorizedMismatch:
623 return inline_vectorizedMismatch();
624
625 case vmIntrinsics::_ghash_processBlocks:
626 return inline_ghash_processBlocks();
627 case vmIntrinsics::_chacha20Block:
628 return inline_chacha20Block();
629 case vmIntrinsics::_kyberNtt:
630 return inline_kyberNtt();
631 case vmIntrinsics::_kyberInverseNtt:
632 return inline_kyberInverseNtt();
633 case vmIntrinsics::_kyberNttMult:
634 return inline_kyberNttMult();
635 case vmIntrinsics::_kyberAddPoly_2:
636 return inline_kyberAddPoly_2();
637 case vmIntrinsics::_kyberAddPoly_3:
638 return inline_kyberAddPoly_3();
639 case vmIntrinsics::_kyber12To16:
640 return inline_kyber12To16();
641 case vmIntrinsics::_kyberBarrettReduce:
642 return inline_kyberBarrettReduce();
643 case vmIntrinsics::_dilithiumAlmostNtt:
644 return inline_dilithiumAlmostNtt();
645 case vmIntrinsics::_dilithiumAlmostInverseNtt:
646 return inline_dilithiumAlmostInverseNtt();
647 case vmIntrinsics::_dilithiumNttMult:
648 return inline_dilithiumNttMult();
649 case vmIntrinsics::_dilithiumMontMulByConstant:
650 return inline_dilithiumMontMulByConstant();
651 case vmIntrinsics::_dilithiumDecomposePoly:
652 return inline_dilithiumDecomposePoly();
653 case vmIntrinsics::_base64_encodeBlock:
654 return inline_base64_encodeBlock();
655 case vmIntrinsics::_base64_decodeBlock:
656 return inline_base64_decodeBlock();
657 case vmIntrinsics::_poly1305_processBlocks:
658 return inline_poly1305_processBlocks();
659 case vmIntrinsics::_intpoly_montgomeryMult_P256:
660 return inline_intpoly_montgomeryMult_P256();
661 case vmIntrinsics::_intpoly_assign:
662 return inline_intpoly_assign();
663 case vmIntrinsics::_encodeISOArray:
664 case vmIntrinsics::_encodeByteISOArray:
665 return inline_encodeISOArray(false);
666 case vmIntrinsics::_encodeAsciiArray:
667 return inline_encodeISOArray(true);
668
669 case vmIntrinsics::_updateCRC32:
670 return inline_updateCRC32();
671 case vmIntrinsics::_updateBytesCRC32:
672 return inline_updateBytesCRC32();
673 case vmIntrinsics::_updateByteBufferCRC32:
674 return inline_updateByteBufferCRC32();
675
676 case vmIntrinsics::_updateBytesCRC32C:
677 return inline_updateBytesCRC32C();
678 case vmIntrinsics::_updateDirectByteBufferCRC32C:
679 return inline_updateDirectByteBufferCRC32C();
680
681 case vmIntrinsics::_updateBytesAdler32:
682 return inline_updateBytesAdler32();
683 case vmIntrinsics::_updateByteBufferAdler32:
684 return inline_updateByteBufferAdler32();
685
686 case vmIntrinsics::_profileBoolean:
687 return inline_profileBoolean();
688 case vmIntrinsics::_isCompileConstant:
689 return inline_isCompileConstant();
690
691 case vmIntrinsics::_countPositives:
692 return inline_countPositives();
693
694 case vmIntrinsics::_fmaD:
695 case vmIntrinsics::_fmaF:
696 return inline_fma(intrinsic_id());
697
698 case vmIntrinsics::_isDigit:
699 case vmIntrinsics::_isLowerCase:
700 case vmIntrinsics::_isUpperCase:
701 case vmIntrinsics::_isWhitespace:
702 return inline_character_compare(intrinsic_id());
703
704 case vmIntrinsics::_min:
705 case vmIntrinsics::_max:
706 case vmIntrinsics::_min_strict:
707 case vmIntrinsics::_max_strict:
708 case vmIntrinsics::_minL:
709 case vmIntrinsics::_maxL:
710 case vmIntrinsics::_minF:
711 case vmIntrinsics::_maxF:
712 case vmIntrinsics::_minD:
713 case vmIntrinsics::_maxD:
714 case vmIntrinsics::_minF_strict:
715 case vmIntrinsics::_maxF_strict:
716 case vmIntrinsics::_minD_strict:
717 case vmIntrinsics::_maxD_strict:
718 return inline_min_max(intrinsic_id());
719
720 case vmIntrinsics::_VectorUnaryOp:
721 return inline_vector_nary_operation(1);
722 case vmIntrinsics::_VectorBinaryOp:
723 return inline_vector_nary_operation(2);
724 case vmIntrinsics::_VectorUnaryLibOp:
725 return inline_vector_call(1);
726 case vmIntrinsics::_VectorBinaryLibOp:
727 return inline_vector_call(2);
728 case vmIntrinsics::_VectorTernaryOp:
729 return inline_vector_nary_operation(3);
730 case vmIntrinsics::_VectorFromBitsCoerced:
731 return inline_vector_frombits_coerced();
732 case vmIntrinsics::_VectorMaskOp:
733 return inline_vector_mask_operation();
734 case vmIntrinsics::_VectorLoadOp:
735 return inline_vector_mem_operation(/*is_store=*/false);
736 case vmIntrinsics::_VectorLoadMaskedOp:
737 return inline_vector_mem_masked_operation(/*is_store*/false);
738 case vmIntrinsics::_VectorStoreOp:
739 return inline_vector_mem_operation(/*is_store=*/true);
740 case vmIntrinsics::_VectorStoreMaskedOp:
741 return inline_vector_mem_masked_operation(/*is_store=*/true);
742 case vmIntrinsics::_VectorGatherOp:
743 return inline_vector_gather_scatter(/*is_scatter*/ false);
744 case vmIntrinsics::_VectorScatterOp:
745 return inline_vector_gather_scatter(/*is_scatter*/ true);
746 case vmIntrinsics::_VectorReductionCoerced:
747 return inline_vector_reduction();
748 case vmIntrinsics::_VectorTest:
749 return inline_vector_test();
750 case vmIntrinsics::_VectorBlend:
751 return inline_vector_blend();
752 case vmIntrinsics::_VectorRearrange:
753 return inline_vector_rearrange();
754 case vmIntrinsics::_VectorSelectFrom:
755 return inline_vector_select_from();
756 case vmIntrinsics::_VectorCompare:
757 return inline_vector_compare();
758 case vmIntrinsics::_VectorBroadcastInt:
759 return inline_vector_broadcast_int();
760 case vmIntrinsics::_VectorConvert:
761 return inline_vector_convert();
762 case vmIntrinsics::_VectorInsert:
763 return inline_vector_insert();
764 case vmIntrinsics::_VectorExtract:
765 return inline_vector_extract();
766 case vmIntrinsics::_VectorCompressExpand:
767 return inline_vector_compress_expand();
768 case vmIntrinsics::_VectorSelectFromTwoVectorOp:
769 return inline_vector_select_from_two_vectors();
770 case vmIntrinsics::_IndexVector:
771 return inline_index_vector();
772 case vmIntrinsics::_IndexPartiallyInUpperRange:
773 return inline_index_partially_in_upper_range();
774
775 case vmIntrinsics::_getObjectSize:
776 return inline_getObjectSize();
777
778 case vmIntrinsics::_blackhole:
779 return inline_blackhole();
780
781 default:
782 // If you get here, it may be that someone has added a new intrinsic
783 // to the list in vmIntrinsics.hpp without implementing it here.
784 #ifndef PRODUCT
785 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
786 tty->print_cr("*** Warning: Unimplemented intrinsic %s(%d)",
787 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
788 }
789 #endif
790 return false;
791 }
792 }
793
794 Node* LibraryCallKit::try_to_predicate(int predicate) {
795 if (!jvms()->has_method()) {
796 // Root JVMState has a null method.
797 assert(map()->memory()->Opcode() == Op_Parm, "");
798 // Insert the memory aliasing node
799 set_all_memory(reset_memory());
800 }
801 assert(merged_memory(), "");
802
803 switch (intrinsic_id()) {
804 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
805 return inline_cipherBlockChaining_AESCrypt_predicate(false);
806 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
807 return inline_cipherBlockChaining_AESCrypt_predicate(true);
808 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
809 return inline_electronicCodeBook_AESCrypt_predicate(false);
810 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
811 return inline_electronicCodeBook_AESCrypt_predicate(true);
812 case vmIntrinsics::_counterMode_AESCrypt:
813 return inline_counterMode_AESCrypt_predicate();
814 case vmIntrinsics::_digestBase_implCompressMB:
815 return inline_digestBase_implCompressMB_predicate(predicate);
816 case vmIntrinsics::_galoisCounterMode_AESCrypt:
817 return inline_galoisCounterMode_AESCrypt_predicate();
818
819 default:
820 // If you get here, it may be that someone has added a new intrinsic
821 // to the list in vmIntrinsics.hpp without implementing it here.
822 #ifndef PRODUCT
823 if ((PrintMiscellaneous && (Verbose || WizardMode)) || PrintOpto) {
824 tty->print_cr("*** Warning: Unimplemented predicate for intrinsic %s(%d)",
825 vmIntrinsics::name_at(intrinsic_id()), vmIntrinsics::as_int(intrinsic_id()));
826 }
827 #endif
828 Node* slow_ctl = control();
829 set_control(top()); // No fast path intrinsic
830 return slow_ctl;
831 }
832 }
833
834 //------------------------------set_result-------------------------------
835 // Helper function for finishing intrinsics.
836 void LibraryCallKit::set_result(RegionNode* region, PhiNode* value) {
837 record_for_igvn(region);
838 set_control(_gvn.transform(region));
839 set_result( _gvn.transform(value));
840 assert(value->type()->basic_type() == result()->bottom_type()->basic_type(), "sanity");
841 }
842
843 //------------------------------generate_guard---------------------------
844 // Helper function for generating guarded fast-slow graph structures.
845 // The given 'test', if true, guards a slow path. If the test fails
846 // then a fast path can be taken. (We generally hope it fails.)
847 // In all cases, GraphKit::control() is updated to the fast path.
848 // The returned value represents the control for the slow path.
849 // The return value is never 'top'; it is either a valid control
850 // or null if it is obvious that the slow path can never be taken.
851 // Also, if region and the slow control are not null, the slow edge
852 // is appended to the region.
853 Node* LibraryCallKit::generate_guard(Node* test, RegionNode* region, float true_prob) {
854 if (stopped()) {
855 // Already short circuited.
856 return nullptr;
857 }
858
859 // Build an if node and its projections.
860 // If test is true we take the slow path, which we assume is uncommon.
861 if (_gvn.type(test) == TypeInt::ZERO) {
862 // The slow branch is never taken. No need to build this guard.
863 return nullptr;
864 }
865
866 IfNode* iff = create_and_map_if(control(), test, true_prob, COUNT_UNKNOWN);
867
868 Node* if_slow = _gvn.transform(new IfTrueNode(iff));
869 if (if_slow == top()) {
870 // The slow branch is never taken. No need to build this guard.
871 return nullptr;
872 }
873
874 if (region != nullptr)
875 region->add_req(if_slow);
876
877 Node* if_fast = _gvn.transform(new IfFalseNode(iff));
878 set_control(if_fast);
879
880 return if_slow;
881 }
882
883 inline Node* LibraryCallKit::generate_slow_guard(Node* test, RegionNode* region) {
884 return generate_guard(test, region, PROB_UNLIKELY_MAG(3));
885 }
886 inline Node* LibraryCallKit::generate_fair_guard(Node* test, RegionNode* region) {
887 return generate_guard(test, region, PROB_FAIR);
888 }
889
890 inline Node* LibraryCallKit::generate_negative_guard(Node* index, RegionNode* region,
891 Node* *pos_index) {
892 if (stopped())
893 return nullptr; // already stopped
894 if (_gvn.type(index)->higher_equal(TypeInt::POS)) // [0,maxint]
895 return nullptr; // index is already adequately typed
896 Node* cmp_lt = _gvn.transform(new CmpINode(index, intcon(0)));
897 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
898 Node* is_neg = generate_guard(bol_lt, region, PROB_MIN);
899 if (is_neg != nullptr && pos_index != nullptr) {
900 // Emulate effect of Parse::adjust_map_after_if.
901 Node* ccast = new CastIINode(control(), index, TypeInt::POS);
902 (*pos_index) = _gvn.transform(ccast);
903 }
904 return is_neg;
905 }
906
907 // Make sure that 'position' is a valid limit index, in [0..length].
908 // There are two equivalent plans for checking this:
909 // A. (offset + copyLength) unsigned<= arrayLength
910 // B. offset <= (arrayLength - copyLength)
911 // We require that all of the values above, except for the sum and
912 // difference, are already known to be non-negative.
913 // Plan A is robust in the face of overflow, if offset and copyLength
914 // are both hugely positive.
915 //
916 // Plan B is less direct and intuitive, but it does not overflow at
917 // all, since the difference of two non-negatives is always
918 // representable. Whenever Java methods must perform the equivalent
919 // check they generally use Plan B instead of Plan A.
920 // For the moment we use Plan A.
921 inline Node* LibraryCallKit::generate_limit_guard(Node* offset,
922 Node* subseq_length,
923 Node* array_length,
924 RegionNode* region) {
925 if (stopped())
926 return nullptr; // already stopped
927 bool zero_offset = _gvn.type(offset) == TypeInt::ZERO;
928 if (zero_offset && subseq_length->eqv_uncast(array_length))
929 return nullptr; // common case of whole-array copy
930 Node* last = subseq_length;
931 if (!zero_offset) // last += offset
932 last = _gvn.transform(new AddINode(last, offset));
933 Node* cmp_lt = _gvn.transform(new CmpUNode(array_length, last));
934 Node* bol_lt = _gvn.transform(new BoolNode(cmp_lt, BoolTest::lt));
935 Node* is_over = generate_guard(bol_lt, region, PROB_MIN);
936 return is_over;
937 }
938
939 // Emit range checks for the given String.value byte array
940 void LibraryCallKit::generate_string_range_check(Node* array, Node* offset, Node* count, bool char_count) {
941 if (stopped()) {
942 return; // already stopped
943 }
944 RegionNode* bailout = new RegionNode(1);
945 record_for_igvn(bailout);
946 if (char_count) {
947 // Convert char count to byte count
948 count = _gvn.transform(new LShiftINode(count, intcon(1)));
949 }
950
951 // Offset and count must not be negative
952 generate_negative_guard(offset, bailout);
953 generate_negative_guard(count, bailout);
954 // Offset + count must not exceed length of array
955 generate_limit_guard(offset, count, load_array_length(array), bailout);
956
957 if (bailout->req() > 1) {
958 PreserveJVMState pjvms(this);
959 set_control(_gvn.transform(bailout));
960 uncommon_trap(Deoptimization::Reason_intrinsic,
961 Deoptimization::Action_maybe_recompile);
962 }
963 }
964
965 Node* LibraryCallKit::current_thread_helper(Node*& tls_output, ByteSize handle_offset,
966 bool is_immutable) {
967 ciKlass* thread_klass = env()->Thread_klass();
968 const Type* thread_type
969 = TypeOopPtr::make_from_klass(thread_klass)->cast_to_ptr_type(TypePtr::NotNull);
970
971 Node* thread = _gvn.transform(new ThreadLocalNode());
972 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(handle_offset));
973 tls_output = thread;
974
975 Node* thread_obj_handle
976 = (is_immutable
977 ? LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
978 TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered)
979 : make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered));
980 thread_obj_handle = _gvn.transform(thread_obj_handle);
981
982 DecoratorSet decorators = IN_NATIVE;
983 if (is_immutable) {
984 decorators |= C2_IMMUTABLE_MEMORY;
985 }
986 return access_load(thread_obj_handle, thread_type, T_OBJECT, decorators);
987 }
988
989 //--------------------------generate_current_thread--------------------
990 Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
991 return current_thread_helper(tls_output, JavaThread::threadObj_offset(),
992 /*is_immutable*/false);
993 }
994
995 //--------------------------generate_virtual_thread--------------------
996 Node* LibraryCallKit::generate_virtual_thread(Node* tls_output) {
997 return current_thread_helper(tls_output, JavaThread::vthread_offset(),
998 !C->method()->changes_current_thread());
999 }
1000
1001 //------------------------------make_string_method_node------------------------
1002 // Helper method for String intrinsic functions. This version is called with
1003 // str1 and str2 pointing to byte[] nodes containing Latin1 or UTF16 encoded
1004 // characters (depending on 'is_byte'). cnt1 and cnt2 are pointing to Int nodes
1005 // containing the lengths of str1 and str2.
1006 Node* LibraryCallKit::make_string_method_node(int opcode, Node* str1_start, Node* cnt1, Node* str2_start, Node* cnt2, StrIntrinsicNode::ArgEnc ae) {
1007 Node* result = nullptr;
1008 switch (opcode) {
1009 case Op_StrIndexOf:
1010 result = new StrIndexOfNode(control(), memory(TypeAryPtr::BYTES),
1011 str1_start, cnt1, str2_start, cnt2, ae);
1012 break;
1013 case Op_StrComp:
1014 result = new StrCompNode(control(), memory(TypeAryPtr::BYTES),
1015 str1_start, cnt1, str2_start, cnt2, ae);
1016 break;
1017 case Op_StrEquals:
1018 // We already know that cnt1 == cnt2 here (checked in 'inline_string_equals').
1019 // Use the constant length if there is one because optimized match rule may exist.
1020 result = new StrEqualsNode(control(), memory(TypeAryPtr::BYTES),
1021 str1_start, str2_start, cnt2->is_Con() ? cnt2 : cnt1, ae);
1022 break;
1023 default:
1024 ShouldNotReachHere();
1025 return nullptr;
1026 }
1027
1028 // All these intrinsics have checks.
1029 C->set_has_split_ifs(true); // Has chance for split-if optimization
1030 clear_upper_avx();
1031
1032 return _gvn.transform(result);
1033 }
1034
1035 //------------------------------inline_string_compareTo------------------------
1036 bool LibraryCallKit::inline_string_compareTo(StrIntrinsicNode::ArgEnc ae) {
1037 Node* arg1 = argument(0);
1038 Node* arg2 = argument(1);
1039
1040 arg1 = must_be_not_null(arg1, true);
1041 arg2 = must_be_not_null(arg2, true);
1042
1043 // Get start addr and length of first argument
1044 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1045 Node* arg1_cnt = load_array_length(arg1);
1046
1047 // Get start addr and length of second argument
1048 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1049 Node* arg2_cnt = load_array_length(arg2);
1050
1051 Node* result = make_string_method_node(Op_StrComp, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1052 set_result(result);
1053 return true;
1054 }
1055
1056 //------------------------------inline_string_equals------------------------
1057 bool LibraryCallKit::inline_string_equals(StrIntrinsicNode::ArgEnc ae) {
1058 Node* arg1 = argument(0);
1059 Node* arg2 = argument(1);
1060
1061 // paths (plus control) merge
1062 RegionNode* region = new RegionNode(3);
1063 Node* phi = new PhiNode(region, TypeInt::BOOL);
1064
1065 if (!stopped()) {
1066
1067 arg1 = must_be_not_null(arg1, true);
1068 arg2 = must_be_not_null(arg2, true);
1069
1070 // Get start addr and length of first argument
1071 Node* arg1_start = array_element_address(arg1, intcon(0), T_BYTE);
1072 Node* arg1_cnt = load_array_length(arg1);
1073
1074 // Get start addr and length of second argument
1075 Node* arg2_start = array_element_address(arg2, intcon(0), T_BYTE);
1076 Node* arg2_cnt = load_array_length(arg2);
1077
1078 // Check for arg1_cnt != arg2_cnt
1079 Node* cmp = _gvn.transform(new CmpINode(arg1_cnt, arg2_cnt));
1080 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
1081 Node* if_ne = generate_slow_guard(bol, nullptr);
1082 if (if_ne != nullptr) {
1083 phi->init_req(2, intcon(0));
1084 region->init_req(2, if_ne);
1085 }
1086
1087 // Check for count == 0 is done by assembler code for StrEquals.
1088
1089 if (!stopped()) {
1090 Node* equals = make_string_method_node(Op_StrEquals, arg1_start, arg1_cnt, arg2_start, arg2_cnt, ae);
1091 phi->init_req(1, equals);
1092 region->init_req(1, control());
1093 }
1094 }
1095
1096 // post merge
1097 set_control(_gvn.transform(region));
1098 record_for_igvn(region);
1099
1100 set_result(_gvn.transform(phi));
1101 return true;
1102 }
1103
1104 //------------------------------inline_array_equals----------------------------
1105 bool LibraryCallKit::inline_array_equals(StrIntrinsicNode::ArgEnc ae) {
1106 assert(ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::LL, "unsupported array types");
1107 Node* arg1 = argument(0);
1108 Node* arg2 = argument(1);
1109
1110 const TypeAryPtr* mtype = (ae == StrIntrinsicNode::UU) ? TypeAryPtr::CHARS : TypeAryPtr::BYTES;
1111 set_result(_gvn.transform(new AryEqNode(control(), memory(mtype), arg1, arg2, ae)));
1112 clear_upper_avx();
1113
1114 return true;
1115 }
1116
1117
1118 //------------------------------inline_countPositives------------------------------
1119 bool LibraryCallKit::inline_countPositives() {
1120 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1121 return false;
1122 }
1123
1124 assert(callee()->signature()->size() == 3, "countPositives has 3 parameters");
1125 // no receiver since it is static method
1126 Node* ba = argument(0);
1127 Node* offset = argument(1);
1128 Node* len = argument(2);
1129
1130 ba = must_be_not_null(ba, true);
1131
1132 // Range checks
1133 generate_string_range_check(ba, offset, len, false);
1134 if (stopped()) {
1135 return true;
1136 }
1137 Node* ba_start = array_element_address(ba, offset, T_BYTE);
1138 Node* result = new CountPositivesNode(control(), memory(TypeAryPtr::BYTES), ba_start, len);
1139 set_result(_gvn.transform(result));
1140 clear_upper_avx();
1141 return true;
1142 }
1143
1144 bool LibraryCallKit::inline_preconditions_checkIndex(BasicType bt) {
1145 Node* index = argument(0);
1146 Node* length = bt == T_INT ? argument(1) : argument(2);
1147 if (too_many_traps(Deoptimization::Reason_intrinsic) || too_many_traps(Deoptimization::Reason_range_check)) {
1148 return false;
1149 }
1150
1151 // check that length is positive
1152 Node* len_pos_cmp = _gvn.transform(CmpNode::make(length, integercon(0, bt), bt));
1153 Node* len_pos_bol = _gvn.transform(new BoolNode(len_pos_cmp, BoolTest::ge));
1154
1155 {
1156 BuildCutout unless(this, len_pos_bol, PROB_MAX);
1157 uncommon_trap(Deoptimization::Reason_intrinsic,
1158 Deoptimization::Action_make_not_entrant);
1159 }
1160
1161 if (stopped()) {
1162 // Length is known to be always negative during compilation and the IR graph so far constructed is good so return success
1163 return true;
1164 }
1165
1166 // length is now known positive, add a cast node to make this explicit
1167 jlong upper_bound = _gvn.type(length)->is_integer(bt)->hi_as_long();
1168 Node* casted_length = ConstraintCastNode::make_cast_for_basic_type(
1169 control(), length, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1170 ConstraintCastNode::RegularDependency, bt);
1171 casted_length = _gvn.transform(casted_length);
1172 replace_in_map(length, casted_length);
1173 length = casted_length;
1174
1175 // Use an unsigned comparison for the range check itself
1176 Node* rc_cmp = _gvn.transform(CmpNode::make(index, length, bt, true));
1177 BoolTest::mask btest = BoolTest::lt;
1178 Node* rc_bool = _gvn.transform(new BoolNode(rc_cmp, btest));
1179 RangeCheckNode* rc = new RangeCheckNode(control(), rc_bool, PROB_MAX, COUNT_UNKNOWN);
1180 _gvn.set_type(rc, rc->Value(&_gvn));
1181 if (!rc_bool->is_Con()) {
1182 record_for_igvn(rc);
1183 }
1184 set_control(_gvn.transform(new IfTrueNode(rc)));
1185 {
1186 PreserveJVMState pjvms(this);
1187 set_control(_gvn.transform(new IfFalseNode(rc)));
1188 uncommon_trap(Deoptimization::Reason_range_check,
1189 Deoptimization::Action_make_not_entrant);
1190 }
1191
1192 if (stopped()) {
1193 // Range check is known to always fail during compilation and the IR graph so far constructed is good so return success
1194 return true;
1195 }
1196
1197 // index is now known to be >= 0 and < length, cast it
1198 Node* result = ConstraintCastNode::make_cast_for_basic_type(
1199 control(), index, TypeInteger::make(0, upper_bound, Type::WidenMax, bt),
1200 ConstraintCastNode::RegularDependency, bt);
1201 result = _gvn.transform(result);
1202 set_result(result);
1203 replace_in_map(index, result);
1204 return true;
1205 }
1206
1207 //------------------------------inline_string_indexOf------------------------
1208 bool LibraryCallKit::inline_string_indexOf(StrIntrinsicNode::ArgEnc ae) {
1209 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1210 return false;
1211 }
1212 Node* src = argument(0);
1213 Node* tgt = argument(1);
1214
1215 // Make the merge point
1216 RegionNode* result_rgn = new RegionNode(4);
1217 Node* result_phi = new PhiNode(result_rgn, TypeInt::INT);
1218
1219 src = must_be_not_null(src, true);
1220 tgt = must_be_not_null(tgt, true);
1221
1222 // Get start addr and length of source string
1223 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
1224 Node* src_count = load_array_length(src);
1225
1226 // Get start addr and length of substring
1227 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1228 Node* tgt_count = load_array_length(tgt);
1229
1230 Node* result = nullptr;
1231 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1232
1233 if (ae == StrIntrinsicNode::UU || ae == StrIntrinsicNode::UL) {
1234 // Divide src size by 2 if String is UTF16 encoded
1235 src_count = _gvn.transform(new RShiftINode(src_count, intcon(1)));
1236 }
1237 if (ae == StrIntrinsicNode::UU) {
1238 // Divide substring size by 2 if String is UTF16 encoded
1239 tgt_count = _gvn.transform(new RShiftINode(tgt_count, intcon(1)));
1240 }
1241
1242 if (call_opt_stub) {
1243 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1244 StubRoutines::_string_indexof_array[ae],
1245 "stringIndexOf", TypePtr::BOTTOM, src_start,
1246 src_count, tgt_start, tgt_count);
1247 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1248 } else {
1249 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1250 result_rgn, result_phi, ae);
1251 }
1252 if (result != nullptr) {
1253 result_phi->init_req(3, result);
1254 result_rgn->init_req(3, control());
1255 }
1256 set_control(_gvn.transform(result_rgn));
1257 record_for_igvn(result_rgn);
1258 set_result(_gvn.transform(result_phi));
1259
1260 return true;
1261 }
1262
1263 //-----------------------------inline_string_indexOfI-----------------------
1264 bool LibraryCallKit::inline_string_indexOfI(StrIntrinsicNode::ArgEnc ae) {
1265 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1266 return false;
1267 }
1268 if (!Matcher::match_rule_supported(Op_StrIndexOf)) {
1269 return false;
1270 }
1271
1272 assert(callee()->signature()->size() == 5, "String.indexOf() has 5 arguments");
1273 Node* src = argument(0); // byte[]
1274 Node* src_count = argument(1); // char count
1275 Node* tgt = argument(2); // byte[]
1276 Node* tgt_count = argument(3); // char count
1277 Node* from_index = argument(4); // char index
1278
1279 src = must_be_not_null(src, true);
1280 tgt = must_be_not_null(tgt, true);
1281
1282 // Multiply byte array index by 2 if String is UTF16 encoded
1283 Node* src_offset = (ae == StrIntrinsicNode::LL) ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1284 src_count = _gvn.transform(new SubINode(src_count, from_index));
1285 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1286 Node* tgt_start = array_element_address(tgt, intcon(0), T_BYTE);
1287
1288 // Range checks
1289 generate_string_range_check(src, src_offset, src_count, ae != StrIntrinsicNode::LL);
1290 generate_string_range_check(tgt, intcon(0), tgt_count, ae == StrIntrinsicNode::UU);
1291 if (stopped()) {
1292 return true;
1293 }
1294
1295 RegionNode* region = new RegionNode(5);
1296 Node* phi = new PhiNode(region, TypeInt::INT);
1297 Node* result = nullptr;
1298
1299 bool call_opt_stub = (StubRoutines::_string_indexof_array[ae] != nullptr);
1300
1301 if (call_opt_stub) {
1302 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::string_IndexOf_Type(),
1303 StubRoutines::_string_indexof_array[ae],
1304 "stringIndexOf", TypePtr::BOTTOM, src_start,
1305 src_count, tgt_start, tgt_count);
1306 result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
1307 } else {
1308 result = make_indexOf_node(src_start, src_count, tgt_start, tgt_count,
1309 region, phi, ae);
1310 }
1311 if (result != nullptr) {
1312 // The result is index relative to from_index if substring was found, -1 otherwise.
1313 // Generate code which will fold into cmove.
1314 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1315 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1316
1317 Node* if_lt = generate_slow_guard(bol, nullptr);
1318 if (if_lt != nullptr) {
1319 // result == -1
1320 phi->init_req(3, result);
1321 region->init_req(3, if_lt);
1322 }
1323 if (!stopped()) {
1324 result = _gvn.transform(new AddINode(result, from_index));
1325 phi->init_req(4, result);
1326 region->init_req(4, control());
1327 }
1328 }
1329
1330 set_control(_gvn.transform(region));
1331 record_for_igvn(region);
1332 set_result(_gvn.transform(phi));
1333 clear_upper_avx();
1334
1335 return true;
1336 }
1337
1338 // Create StrIndexOfNode with fast path checks
1339 Node* LibraryCallKit::make_indexOf_node(Node* src_start, Node* src_count, Node* tgt_start, Node* tgt_count,
1340 RegionNode* region, Node* phi, StrIntrinsicNode::ArgEnc ae) {
1341 // Check for substr count > string count
1342 Node* cmp = _gvn.transform(new CmpINode(tgt_count, src_count));
1343 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::gt));
1344 Node* if_gt = generate_slow_guard(bol, nullptr);
1345 if (if_gt != nullptr) {
1346 phi->init_req(1, intcon(-1));
1347 region->init_req(1, if_gt);
1348 }
1349 if (!stopped()) {
1350 // Check for substr count == 0
1351 cmp = _gvn.transform(new CmpINode(tgt_count, intcon(0)));
1352 bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
1353 Node* if_zero = generate_slow_guard(bol, nullptr);
1354 if (if_zero != nullptr) {
1355 phi->init_req(2, intcon(0));
1356 region->init_req(2, if_zero);
1357 }
1358 }
1359 if (!stopped()) {
1360 return make_string_method_node(Op_StrIndexOf, src_start, src_count, tgt_start, tgt_count, ae);
1361 }
1362 return nullptr;
1363 }
1364
1365 //-----------------------------inline_string_indexOfChar-----------------------
1366 bool LibraryCallKit::inline_string_indexOfChar(StrIntrinsicNode::ArgEnc ae) {
1367 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1368 return false;
1369 }
1370 if (!Matcher::match_rule_supported(Op_StrIndexOfChar)) {
1371 return false;
1372 }
1373 assert(callee()->signature()->size() == 4, "String.indexOfChar() has 4 arguments");
1374 Node* src = argument(0); // byte[]
1375 Node* int_ch = argument(1);
1376 Node* from_index = argument(2);
1377 Node* max = argument(3);
1378
1379 src = must_be_not_null(src, true);
1380
1381 Node* src_offset = ae == StrIntrinsicNode::L ? from_index : _gvn.transform(new LShiftINode(from_index, intcon(1)));
1382 Node* src_start = array_element_address(src, src_offset, T_BYTE);
1383 Node* src_count = _gvn.transform(new SubINode(max, from_index));
1384
1385 // Range checks
1386 generate_string_range_check(src, src_offset, src_count, ae == StrIntrinsicNode::U);
1387
1388 // Check for int_ch >= 0
1389 Node* int_ch_cmp = _gvn.transform(new CmpINode(int_ch, intcon(0)));
1390 Node* int_ch_bol = _gvn.transform(new BoolNode(int_ch_cmp, BoolTest::ge));
1391 {
1392 BuildCutout unless(this, int_ch_bol, PROB_MAX);
1393 uncommon_trap(Deoptimization::Reason_intrinsic,
1394 Deoptimization::Action_maybe_recompile);
1395 }
1396 if (stopped()) {
1397 return true;
1398 }
1399
1400 RegionNode* region = new RegionNode(3);
1401 Node* phi = new PhiNode(region, TypeInt::INT);
1402
1403 Node* result = new StrIndexOfCharNode(control(), memory(TypeAryPtr::BYTES), src_start, src_count, int_ch, ae);
1404 C->set_has_split_ifs(true); // Has chance for split-if optimization
1405 _gvn.transform(result);
1406
1407 Node* cmp = _gvn.transform(new CmpINode(result, intcon(0)));
1408 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::lt));
1409
1410 Node* if_lt = generate_slow_guard(bol, nullptr);
1411 if (if_lt != nullptr) {
1412 // result == -1
1413 phi->init_req(2, result);
1414 region->init_req(2, if_lt);
1415 }
1416 if (!stopped()) {
1417 result = _gvn.transform(new AddINode(result, from_index));
1418 phi->init_req(1, result);
1419 region->init_req(1, control());
1420 }
1421 set_control(_gvn.transform(region));
1422 record_for_igvn(region);
1423 set_result(_gvn.transform(phi));
1424 clear_upper_avx();
1425
1426 return true;
1427 }
1428 //---------------------------inline_string_copy---------------------
1429 // compressIt == true --> generate a compressed copy operation (compress char[]/byte[] to byte[])
1430 // int StringUTF16.compress(char[] src, int srcOff, byte[] dst, int dstOff, int len)
1431 // int StringUTF16.compress(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1432 // compressIt == false --> generate an inflated copy operation (inflate byte[] to char[]/byte[])
1433 // void StringLatin1.inflate(byte[] src, int srcOff, char[] dst, int dstOff, int len)
1434 // void StringLatin1.inflate(byte[] src, int srcOff, byte[] dst, int dstOff, int len)
1435 bool LibraryCallKit::inline_string_copy(bool compress) {
1436 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1437 return false;
1438 }
1439 int nargs = 5; // 2 oops, 3 ints
1440 assert(callee()->signature()->size() == nargs, "string copy has 5 arguments");
1441
1442 Node* src = argument(0);
1443 Node* src_offset = argument(1);
1444 Node* dst = argument(2);
1445 Node* dst_offset = argument(3);
1446 Node* length = argument(4);
1447
1448 // Check for allocation before we add nodes that would confuse
1449 // tightly_coupled_allocation()
1450 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1451
1452 // Figure out the size and type of the elements we will be copying.
1453 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
1454 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
1455 if (src_type == nullptr || dst_type == nullptr) {
1456 return false;
1457 }
1458 BasicType src_elem = src_type->elem()->array_element_basic_type();
1459 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
1460 assert((compress && dst_elem == T_BYTE && (src_elem == T_BYTE || src_elem == T_CHAR)) ||
1461 (!compress && src_elem == T_BYTE && (dst_elem == T_BYTE || dst_elem == T_CHAR)),
1462 "Unsupported array types for inline_string_copy");
1463
1464 src = must_be_not_null(src, true);
1465 dst = must_be_not_null(dst, true);
1466
1467 // Convert char[] offsets to byte[] offsets
1468 bool convert_src = (compress && src_elem == T_BYTE);
1469 bool convert_dst = (!compress && dst_elem == T_BYTE);
1470 if (convert_src) {
1471 src_offset = _gvn.transform(new LShiftINode(src_offset, intcon(1)));
1472 } else if (convert_dst) {
1473 dst_offset = _gvn.transform(new LShiftINode(dst_offset, intcon(1)));
1474 }
1475
1476 // Range checks
1477 generate_string_range_check(src, src_offset, length, convert_src);
1478 generate_string_range_check(dst, dst_offset, length, convert_dst);
1479 if (stopped()) {
1480 return true;
1481 }
1482
1483 Node* src_start = array_element_address(src, src_offset, src_elem);
1484 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
1485 // 'src_start' points to src array + scaled offset
1486 // 'dst_start' points to dst array + scaled offset
1487 Node* count = nullptr;
1488 if (compress) {
1489 count = compress_string(src_start, TypeAryPtr::get_array_body_type(src_elem), dst_start, length);
1490 } else {
1491 inflate_string(src_start, dst_start, TypeAryPtr::get_array_body_type(dst_elem), length);
1492 }
1493
1494 if (alloc != nullptr) {
1495 if (alloc->maybe_set_complete(&_gvn)) {
1496 // "You break it, you buy it."
1497 InitializeNode* init = alloc->initialization();
1498 assert(init->is_complete(), "we just did this");
1499 init->set_complete_with_arraycopy();
1500 assert(dst->is_CheckCastPP(), "sanity");
1501 assert(dst->in(0)->in(0) == init, "dest pinned");
1502 }
1503 // Do not let stores that initialize this object be reordered with
1504 // a subsequent store that would make this object accessible by
1505 // other threads.
1506 // Record what AllocateNode this StoreStore protects so that
1507 // escape analysis can go from the MemBarStoreStoreNode to the
1508 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1509 // based on the escape status of the AllocateNode.
1510 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1511 }
1512 if (compress) {
1513 set_result(_gvn.transform(count));
1514 }
1515 clear_upper_avx();
1516
1517 return true;
1518 }
1519
1520 #ifdef _LP64
1521 #define XTOP ,top() /*additional argument*/
1522 #else //_LP64
1523 #define XTOP /*no additional argument*/
1524 #endif //_LP64
1525
1526 //------------------------inline_string_toBytesU--------------------------
1527 // public static byte[] StringUTF16.toBytes(char[] value, int off, int len)
1528 bool LibraryCallKit::inline_string_toBytesU() {
1529 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1530 return false;
1531 }
1532 // Get the arguments.
1533 Node* value = argument(0);
1534 Node* offset = argument(1);
1535 Node* length = argument(2);
1536
1537 Node* newcopy = nullptr;
1538
1539 // Set the original stack and the reexecute bit for the interpreter to reexecute
1540 // the bytecode that invokes StringUTF16.toBytes() if deoptimization happens.
1541 { PreserveReexecuteState preexecs(this);
1542 jvms()->set_should_reexecute(true);
1543
1544 // Check if a null path was taken unconditionally.
1545 value = null_check(value);
1546
1547 RegionNode* bailout = new RegionNode(1);
1548 record_for_igvn(bailout);
1549
1550 // Range checks
1551 generate_negative_guard(offset, bailout);
1552 generate_negative_guard(length, bailout);
1553 generate_limit_guard(offset, length, load_array_length(value), bailout);
1554 // Make sure that resulting byte[] length does not overflow Integer.MAX_VALUE
1555 generate_limit_guard(length, intcon(0), intcon(max_jint/2), bailout);
1556
1557 if (bailout->req() > 1) {
1558 PreserveJVMState pjvms(this);
1559 set_control(_gvn.transform(bailout));
1560 uncommon_trap(Deoptimization::Reason_intrinsic,
1561 Deoptimization::Action_maybe_recompile);
1562 }
1563 if (stopped()) {
1564 return true;
1565 }
1566
1567 Node* size = _gvn.transform(new LShiftINode(length, intcon(1)));
1568 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_BYTE)));
1569 newcopy = new_array(klass_node, size, 0); // no arguments to push
1570 AllocateArrayNode* alloc = tightly_coupled_allocation(newcopy);
1571 guarantee(alloc != nullptr, "created above");
1572
1573 // Calculate starting addresses.
1574 Node* src_start = array_element_address(value, offset, T_CHAR);
1575 Node* dst_start = basic_plus_adr(newcopy, arrayOopDesc::base_offset_in_bytes(T_BYTE));
1576
1577 // Check if src array address is aligned to HeapWordSize (dst is always aligned)
1578 const TypeInt* toffset = gvn().type(offset)->is_int();
1579 bool aligned = toffset->is_con() && ((toffset->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1580
1581 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1582 const char* copyfunc_name = "arraycopy";
1583 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1584 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1585 OptoRuntime::fast_arraycopy_Type(),
1586 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1587 src_start, dst_start, ConvI2X(length) XTOP);
1588 // Do not let reads from the cloned object float above the arraycopy.
1589 if (alloc->maybe_set_complete(&_gvn)) {
1590 // "You break it, you buy it."
1591 InitializeNode* init = alloc->initialization();
1592 assert(init->is_complete(), "we just did this");
1593 init->set_complete_with_arraycopy();
1594 assert(newcopy->is_CheckCastPP(), "sanity");
1595 assert(newcopy->in(0)->in(0) == init, "dest pinned");
1596 }
1597 // Do not let stores that initialize this object be reordered with
1598 // a subsequent store that would make this object accessible by
1599 // other threads.
1600 // Record what AllocateNode this StoreStore protects so that
1601 // escape analysis can go from the MemBarStoreStoreNode to the
1602 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1603 // based on the escape status of the AllocateNode.
1604 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1605 } // original reexecute is set back here
1606
1607 C->set_has_split_ifs(true); // Has chance for split-if optimization
1608 if (!stopped()) {
1609 set_result(newcopy);
1610 }
1611 clear_upper_avx();
1612
1613 return true;
1614 }
1615
1616 //------------------------inline_string_getCharsU--------------------------
1617 // public void StringUTF16.getChars(byte[] src, int srcBegin, int srcEnd, char dst[], int dstBegin)
1618 bool LibraryCallKit::inline_string_getCharsU() {
1619 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
1620 return false;
1621 }
1622
1623 // Get the arguments.
1624 Node* src = argument(0);
1625 Node* src_begin = argument(1);
1626 Node* src_end = argument(2); // exclusive offset (i < src_end)
1627 Node* dst = argument(3);
1628 Node* dst_begin = argument(4);
1629
1630 // Check for allocation before we add nodes that would confuse
1631 // tightly_coupled_allocation()
1632 AllocateArrayNode* alloc = tightly_coupled_allocation(dst);
1633
1634 // Check if a null path was taken unconditionally.
1635 src = null_check(src);
1636 dst = null_check(dst);
1637 if (stopped()) {
1638 return true;
1639 }
1640
1641 // Get length and convert char[] offset to byte[] offset
1642 Node* length = _gvn.transform(new SubINode(src_end, src_begin));
1643 src_begin = _gvn.transform(new LShiftINode(src_begin, intcon(1)));
1644
1645 // Range checks
1646 generate_string_range_check(src, src_begin, length, true);
1647 generate_string_range_check(dst, dst_begin, length, false);
1648 if (stopped()) {
1649 return true;
1650 }
1651
1652 if (!stopped()) {
1653 // Calculate starting addresses.
1654 Node* src_start = array_element_address(src, src_begin, T_BYTE);
1655 Node* dst_start = array_element_address(dst, dst_begin, T_CHAR);
1656
1657 // Check if array addresses are aligned to HeapWordSize
1658 const TypeInt* tsrc = gvn().type(src_begin)->is_int();
1659 const TypeInt* tdst = gvn().type(dst_begin)->is_int();
1660 bool aligned = tsrc->is_con() && ((tsrc->get_con() * type2aelembytes(T_BYTE)) % HeapWordSize == 0) &&
1661 tdst->is_con() && ((tdst->get_con() * type2aelembytes(T_CHAR)) % HeapWordSize == 0);
1662
1663 // Figure out which arraycopy runtime method to call (disjoint, uninitialized).
1664 const char* copyfunc_name = "arraycopy";
1665 address copyfunc_addr = StubRoutines::select_arraycopy_function(T_CHAR, aligned, true, copyfunc_name, true);
1666 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
1667 OptoRuntime::fast_arraycopy_Type(),
1668 copyfunc_addr, copyfunc_name, TypeRawPtr::BOTTOM,
1669 src_start, dst_start, ConvI2X(length) XTOP);
1670 // Do not let reads from the cloned object float above the arraycopy.
1671 if (alloc != nullptr) {
1672 if (alloc->maybe_set_complete(&_gvn)) {
1673 // "You break it, you buy it."
1674 InitializeNode* init = alloc->initialization();
1675 assert(init->is_complete(), "we just did this");
1676 init->set_complete_with_arraycopy();
1677 assert(dst->is_CheckCastPP(), "sanity");
1678 assert(dst->in(0)->in(0) == init, "dest pinned");
1679 }
1680 // Do not let stores that initialize this object be reordered with
1681 // a subsequent store that would make this object accessible by
1682 // other threads.
1683 // Record what AllocateNode this StoreStore protects so that
1684 // escape analysis can go from the MemBarStoreStoreNode to the
1685 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
1686 // based on the escape status of the AllocateNode.
1687 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
1688 } else {
1689 insert_mem_bar(Op_MemBarCPUOrder);
1690 }
1691 }
1692
1693 C->set_has_split_ifs(true); // Has chance for split-if optimization
1694 return true;
1695 }
1696
1697 //----------------------inline_string_char_access----------------------------
1698 // Store/Load char to/from byte[] array.
1699 // static void StringUTF16.putChar(byte[] val, int index, int c)
1700 // static char StringUTF16.getChar(byte[] val, int index)
1701 bool LibraryCallKit::inline_string_char_access(bool is_store) {
1702 Node* value = argument(0);
1703 Node* index = argument(1);
1704 Node* ch = is_store ? argument(2) : nullptr;
1705
1706 // This intrinsic accesses byte[] array as char[] array. Computing the offsets
1707 // correctly requires matched array shapes.
1708 assert (arrayOopDesc::base_offset_in_bytes(T_CHAR) == arrayOopDesc::base_offset_in_bytes(T_BYTE),
1709 "sanity: byte[] and char[] bases agree");
1710 assert (type2aelembytes(T_CHAR) == type2aelembytes(T_BYTE)*2,
1711 "sanity: byte[] and char[] scales agree");
1712
1713 // Bail when getChar over constants is requested: constant folding would
1714 // reject folding mismatched char access over byte[]. A normal inlining for getChar
1715 // Java method would constant fold nicely instead.
1716 if (!is_store && value->is_Con() && index->is_Con()) {
1717 return false;
1718 }
1719
1720 // Save state and restore on bailout
1721 uint old_sp = sp();
1722 SafePointNode* old_map = clone_map();
1723
1724 value = must_be_not_null(value, true);
1725
1726 Node* adr = array_element_address(value, index, T_CHAR);
1727 if (adr->is_top()) {
1728 set_map(old_map);
1729 set_sp(old_sp);
1730 return false;
1731 }
1732 destruct_map_clone(old_map);
1733 if (is_store) {
1734 access_store_at(value, adr, TypeAryPtr::BYTES, ch, TypeInt::CHAR, T_CHAR, IN_HEAP | MO_UNORDERED | C2_MISMATCHED);
1735 } else {
1736 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);
1737 set_result(ch);
1738 }
1739 return true;
1740 }
1741
1742
1743 //------------------------------inline_math-----------------------------------
1744 // public static double Math.abs(double)
1745 // public static double Math.sqrt(double)
1746 // public static double Math.log(double)
1747 // public static double Math.log10(double)
1748 // public static double Math.round(double)
1749 bool LibraryCallKit::inline_double_math(vmIntrinsics::ID id) {
1750 Node* arg = argument(0);
1751 Node* n = nullptr;
1752 switch (id) {
1753 case vmIntrinsics::_dabs: n = new AbsDNode( arg); break;
1754 case vmIntrinsics::_dsqrt:
1755 case vmIntrinsics::_dsqrt_strict:
1756 n = new SqrtDNode(C, control(), arg); break;
1757 case vmIntrinsics::_ceil: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_ceil); break;
1758 case vmIntrinsics::_floor: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_floor); break;
1759 case vmIntrinsics::_rint: n = RoundDoubleModeNode::make(_gvn, arg, RoundDoubleModeNode::rmode_rint); break;
1760 case vmIntrinsics::_roundD: n = new RoundDNode(arg); break;
1761 case vmIntrinsics::_dcopySign: n = CopySignDNode::make(_gvn, arg, argument(2)); break;
1762 case vmIntrinsics::_dsignum: n = SignumDNode::make(_gvn, arg); break;
1763 default: fatal_unexpected_iid(id); break;
1764 }
1765 set_result(_gvn.transform(n));
1766 return true;
1767 }
1768
1769 //------------------------------inline_math-----------------------------------
1770 // public static float Math.abs(float)
1771 // public static int Math.abs(int)
1772 // public static long Math.abs(long)
1773 bool LibraryCallKit::inline_math(vmIntrinsics::ID id) {
1774 Node* arg = argument(0);
1775 Node* n = nullptr;
1776 switch (id) {
1777 case vmIntrinsics::_fabs: n = new AbsFNode( arg); break;
1778 case vmIntrinsics::_iabs: n = new AbsINode( arg); break;
1779 case vmIntrinsics::_labs: n = new AbsLNode( arg); break;
1780 case vmIntrinsics::_fcopySign: n = new CopySignFNode(arg, argument(1)); break;
1781 case vmIntrinsics::_fsignum: n = SignumFNode::make(_gvn, arg); break;
1782 case vmIntrinsics::_roundF: n = new RoundFNode(arg); break;
1783 default: fatal_unexpected_iid(id); break;
1784 }
1785 set_result(_gvn.transform(n));
1786 return true;
1787 }
1788
1789 //------------------------------runtime_math-----------------------------
1790 bool LibraryCallKit::runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName) {
1791 assert(call_type == OptoRuntime::Math_DD_D_Type() || call_type == OptoRuntime::Math_D_D_Type(),
1792 "must be (DD)D or (D)D type");
1793
1794 // Inputs
1795 Node* a = argument(0);
1796 Node* b = (call_type == OptoRuntime::Math_DD_D_Type()) ? argument(2) : nullptr;
1797
1798 const TypePtr* no_memory_effects = nullptr;
1799 Node* trig = make_runtime_call(RC_LEAF, call_type, funcAddr, funcName,
1800 no_memory_effects,
1801 a, top(), b, b ? top() : nullptr);
1802 Node* value = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1803 #ifdef ASSERT
1804 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1805 assert(value_top == top(), "second value must be top");
1806 #endif
1807
1808 set_result(value);
1809 return true;
1810 }
1811
1812 //------------------------------inline_math_pow-----------------------------
1813 bool LibraryCallKit::inline_math_pow() {
1814 Node* exp = argument(2);
1815 const TypeD* d = _gvn.type(exp)->isa_double_constant();
1816 if (d != nullptr) {
1817 if (d->getd() == 2.0) {
1818 // Special case: pow(x, 2.0) => x * x
1819 Node* base = argument(0);
1820 set_result(_gvn.transform(new MulDNode(base, base)));
1821 return true;
1822 } else if (d->getd() == 0.5 && Matcher::match_rule_supported(Op_SqrtD)) {
1823 // Special case: pow(x, 0.5) => sqrt(x)
1824 Node* base = argument(0);
1825 Node* zero = _gvn.zerocon(T_DOUBLE);
1826
1827 RegionNode* region = new RegionNode(3);
1828 Node* phi = new PhiNode(region, Type::DOUBLE);
1829
1830 Node* cmp = _gvn.transform(new CmpDNode(base, zero));
1831 // According to the API specs, pow(-0.0, 0.5) = 0.0 and sqrt(-0.0) = -0.0.
1832 // So pow(-0.0, 0.5) shouldn't be replaced with sqrt(-0.0).
1833 // -0.0/+0.0 are both excluded since floating-point comparison doesn't distinguish -0.0 from +0.0.
1834 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::le));
1835
1836 Node* if_pow = generate_slow_guard(test, nullptr);
1837 Node* value_sqrt = _gvn.transform(new SqrtDNode(C, control(), base));
1838 phi->init_req(1, value_sqrt);
1839 region->init_req(1, control());
1840
1841 if (if_pow != nullptr) {
1842 set_control(if_pow);
1843 address target = StubRoutines::dpow() != nullptr ? StubRoutines::dpow() :
1844 CAST_FROM_FN_PTR(address, SharedRuntime::dpow);
1845 const TypePtr* no_memory_effects = nullptr;
1846 Node* trig = make_runtime_call(RC_LEAF, OptoRuntime::Math_DD_D_Type(), target, "POW",
1847 no_memory_effects, base, top(), exp, top());
1848 Node* value_pow = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+0));
1849 #ifdef ASSERT
1850 Node* value_top = _gvn.transform(new ProjNode(trig, TypeFunc::Parms+1));
1851 assert(value_top == top(), "second value must be top");
1852 #endif
1853 phi->init_req(2, value_pow);
1854 region->init_req(2, _gvn.transform(new ProjNode(trig, TypeFunc::Control)));
1855 }
1856
1857 C->set_has_split_ifs(true); // Has chance for split-if optimization
1858 set_control(_gvn.transform(region));
1859 record_for_igvn(region);
1860 set_result(_gvn.transform(phi));
1861
1862 return true;
1863 }
1864 }
1865
1866 return StubRoutines::dpow() != nullptr ?
1867 runtime_math(OptoRuntime::Math_DD_D_Type(), StubRoutines::dpow(), "dpow") :
1868 runtime_math(OptoRuntime::Math_DD_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dpow), "POW");
1869 }
1870
1871 //------------------------------inline_math_native-----------------------------
1872 bool LibraryCallKit::inline_math_native(vmIntrinsics::ID id) {
1873 switch (id) {
1874 case vmIntrinsics::_dsin:
1875 return StubRoutines::dsin() != nullptr ?
1876 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dsin(), "dsin") :
1877 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dsin), "SIN");
1878 case vmIntrinsics::_dcos:
1879 return StubRoutines::dcos() != nullptr ?
1880 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dcos(), "dcos") :
1881 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dcos), "COS");
1882 case vmIntrinsics::_dtan:
1883 return StubRoutines::dtan() != nullptr ?
1884 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtan(), "dtan") :
1885 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dtan), "TAN");
1886 case vmIntrinsics::_dtanh:
1887 return StubRoutines::dtanh() != nullptr ?
1888 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dtanh(), "dtanh") : false;
1889 case vmIntrinsics::_dexp:
1890 return StubRoutines::dexp() != nullptr ?
1891 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dexp(), "dexp") :
1892 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dexp), "EXP");
1893 case vmIntrinsics::_dlog:
1894 return StubRoutines::dlog() != nullptr ?
1895 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog(), "dlog") :
1896 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog), "LOG");
1897 case vmIntrinsics::_dlog10:
1898 return StubRoutines::dlog10() != nullptr ?
1899 runtime_math(OptoRuntime::Math_D_D_Type(), StubRoutines::dlog10(), "dlog10") :
1900 runtime_math(OptoRuntime::Math_D_D_Type(), CAST_FROM_FN_PTR(address, SharedRuntime::dlog10), "LOG10");
1901
1902 case vmIntrinsics::_roundD: return Matcher::match_rule_supported(Op_RoundD) ? inline_double_math(id) : false;
1903 case vmIntrinsics::_ceil:
1904 case vmIntrinsics::_floor:
1905 case vmIntrinsics::_rint: return Matcher::match_rule_supported(Op_RoundDoubleMode) ? inline_double_math(id) : false;
1906
1907 case vmIntrinsics::_dsqrt:
1908 case vmIntrinsics::_dsqrt_strict:
1909 return Matcher::match_rule_supported(Op_SqrtD) ? inline_double_math(id) : false;
1910 case vmIntrinsics::_dabs: return Matcher::has_match_rule(Op_AbsD) ? inline_double_math(id) : false;
1911 case vmIntrinsics::_fabs: return Matcher::match_rule_supported(Op_AbsF) ? inline_math(id) : false;
1912 case vmIntrinsics::_iabs: return Matcher::match_rule_supported(Op_AbsI) ? inline_math(id) : false;
1913 case vmIntrinsics::_labs: return Matcher::match_rule_supported(Op_AbsL) ? inline_math(id) : false;
1914
1915 case vmIntrinsics::_dpow: return inline_math_pow();
1916 case vmIntrinsics::_dcopySign: return inline_double_math(id);
1917 case vmIntrinsics::_fcopySign: return inline_math(id);
1918 case vmIntrinsics::_dsignum: return Matcher::match_rule_supported(Op_SignumD) ? inline_double_math(id) : false;
1919 case vmIntrinsics::_fsignum: return Matcher::match_rule_supported(Op_SignumF) ? inline_math(id) : false;
1920 case vmIntrinsics::_roundF: return Matcher::match_rule_supported(Op_RoundF) ? inline_math(id) : false;
1921
1922 // These intrinsics are not yet correctly implemented
1923 case vmIntrinsics::_datan2:
1924 return false;
1925
1926 default:
1927 fatal_unexpected_iid(id);
1928 return false;
1929 }
1930 }
1931
1932 //----------------------------inline_notify-----------------------------------*
1933 bool LibraryCallKit::inline_notify(vmIntrinsics::ID id) {
1934 const TypeFunc* ftype = OptoRuntime::monitor_notify_Type();
1935 address func;
1936 if (id == vmIntrinsics::_notify) {
1937 func = OptoRuntime::monitor_notify_Java();
1938 } else {
1939 func = OptoRuntime::monitor_notifyAll_Java();
1940 }
1941 Node* call = make_runtime_call(RC_NO_LEAF, ftype, func, nullptr, TypeRawPtr::BOTTOM, argument(0));
1942 make_slow_call_ex(call, env()->Throwable_klass(), false);
1943 return true;
1944 }
1945
1946
1947 //----------------------------inline_min_max-----------------------------------
1948 bool LibraryCallKit::inline_min_max(vmIntrinsics::ID id) {
1949 Node* a = nullptr;
1950 Node* b = nullptr;
1951 Node* n = nullptr;
1952 switch (id) {
1953 case vmIntrinsics::_min:
1954 case vmIntrinsics::_max:
1955 case vmIntrinsics::_minF:
1956 case vmIntrinsics::_maxF:
1957 case vmIntrinsics::_minF_strict:
1958 case vmIntrinsics::_maxF_strict:
1959 case vmIntrinsics::_min_strict:
1960 case vmIntrinsics::_max_strict:
1961 assert(callee()->signature()->size() == 2, "minF/maxF has 2 parameters of size 1 each.");
1962 a = argument(0);
1963 b = argument(1);
1964 break;
1965 case vmIntrinsics::_minD:
1966 case vmIntrinsics::_maxD:
1967 case vmIntrinsics::_minD_strict:
1968 case vmIntrinsics::_maxD_strict:
1969 assert(callee()->signature()->size() == 4, "minD/maxD has 2 parameters of size 2 each.");
1970 a = argument(0);
1971 b = argument(2);
1972 break;
1973 case vmIntrinsics::_minL:
1974 case vmIntrinsics::_maxL:
1975 assert(callee()->signature()->size() == 4, "minL/maxL has 2 parameters of size 2 each.");
1976 a = argument(0);
1977 b = argument(2);
1978 break;
1979 default:
1980 fatal_unexpected_iid(id);
1981 break;
1982 }
1983
1984 switch (id) {
1985 case vmIntrinsics::_min:
1986 case vmIntrinsics::_min_strict:
1987 n = new MinINode(a, b);
1988 break;
1989 case vmIntrinsics::_max:
1990 case vmIntrinsics::_max_strict:
1991 n = new MaxINode(a, b);
1992 break;
1993 case vmIntrinsics::_minF:
1994 case vmIntrinsics::_minF_strict:
1995 n = new MinFNode(a, b);
1996 break;
1997 case vmIntrinsics::_maxF:
1998 case vmIntrinsics::_maxF_strict:
1999 n = new MaxFNode(a, b);
2000 break;
2001 case vmIntrinsics::_minD:
2002 case vmIntrinsics::_minD_strict:
2003 n = new MinDNode(a, b);
2004 break;
2005 case vmIntrinsics::_maxD:
2006 case vmIntrinsics::_maxD_strict:
2007 n = new MaxDNode(a, b);
2008 break;
2009 case vmIntrinsics::_minL:
2010 n = new MinLNode(_gvn.C, a, b);
2011 break;
2012 case vmIntrinsics::_maxL:
2013 n = new MaxLNode(_gvn.C, a, b);
2014 break;
2015 default:
2016 fatal_unexpected_iid(id);
2017 break;
2018 }
2019
2020 set_result(_gvn.transform(n));
2021 return true;
2022 }
2023
2024 bool LibraryCallKit::inline_math_mathExact(Node* math, Node* test) {
2025 if (builtin_throw_too_many_traps(Deoptimization::Reason_intrinsic,
2026 env()->ArithmeticException_instance())) {
2027 // It has been already too many times, but we cannot use builtin_throw (e.g. we care about backtraces),
2028 // so let's bail out intrinsic rather than risking deopting again.
2029 return false;
2030 }
2031
2032 Node* bol = _gvn.transform( new BoolNode(test, BoolTest::overflow) );
2033 IfNode* check = create_and_map_if(control(), bol, PROB_UNLIKELY_MAG(3), COUNT_UNKNOWN);
2034 Node* fast_path = _gvn.transform( new IfFalseNode(check));
2035 Node* slow_path = _gvn.transform( new IfTrueNode(check) );
2036
2037 {
2038 PreserveJVMState pjvms(this);
2039 PreserveReexecuteState preexecs(this);
2040 jvms()->set_should_reexecute(true);
2041
2042 set_control(slow_path);
2043 set_i_o(i_o());
2044
2045 builtin_throw(Deoptimization::Reason_intrinsic,
2046 env()->ArithmeticException_instance(),
2047 /*allow_too_many_traps*/ false);
2048 }
2049
2050 set_control(fast_path);
2051 set_result(math);
2052 return true;
2053 }
2054
2055 template <typename OverflowOp>
2056 bool LibraryCallKit::inline_math_overflow(Node* arg1, Node* arg2) {
2057 typedef typename OverflowOp::MathOp MathOp;
2058
2059 MathOp* mathOp = new MathOp(arg1, arg2);
2060 Node* operation = _gvn.transform( mathOp );
2061 Node* ofcheck = _gvn.transform( new OverflowOp(arg1, arg2) );
2062 return inline_math_mathExact(operation, ofcheck);
2063 }
2064
2065 bool LibraryCallKit::inline_math_addExactI(bool is_increment) {
2066 return inline_math_overflow<OverflowAddINode>(argument(0), is_increment ? intcon(1) : argument(1));
2067 }
2068
2069 bool LibraryCallKit::inline_math_addExactL(bool is_increment) {
2070 return inline_math_overflow<OverflowAddLNode>(argument(0), is_increment ? longcon(1) : argument(2));
2071 }
2072
2073 bool LibraryCallKit::inline_math_subtractExactI(bool is_decrement) {
2074 return inline_math_overflow<OverflowSubINode>(argument(0), is_decrement ? intcon(1) : argument(1));
2075 }
2076
2077 bool LibraryCallKit::inline_math_subtractExactL(bool is_decrement) {
2078 return inline_math_overflow<OverflowSubLNode>(argument(0), is_decrement ? longcon(1) : argument(2));
2079 }
2080
2081 bool LibraryCallKit::inline_math_negateExactI() {
2082 return inline_math_overflow<OverflowSubINode>(intcon(0), argument(0));
2083 }
2084
2085 bool LibraryCallKit::inline_math_negateExactL() {
2086 return inline_math_overflow<OverflowSubLNode>(longcon(0), argument(0));
2087 }
2088
2089 bool LibraryCallKit::inline_math_multiplyExactI() {
2090 return inline_math_overflow<OverflowMulINode>(argument(0), argument(1));
2091 }
2092
2093 bool LibraryCallKit::inline_math_multiplyExactL() {
2094 return inline_math_overflow<OverflowMulLNode>(argument(0), argument(2));
2095 }
2096
2097 bool LibraryCallKit::inline_math_multiplyHigh() {
2098 set_result(_gvn.transform(new MulHiLNode(argument(0), argument(2))));
2099 return true;
2100 }
2101
2102 bool LibraryCallKit::inline_math_unsignedMultiplyHigh() {
2103 set_result(_gvn.transform(new UMulHiLNode(argument(0), argument(2))));
2104 return true;
2105 }
2106
2107 inline int
2108 LibraryCallKit::classify_unsafe_addr(Node* &base, Node* &offset, BasicType type) {
2109 const TypePtr* base_type = TypePtr::NULL_PTR;
2110 if (base != nullptr) base_type = _gvn.type(base)->isa_ptr();
2111 if (base_type == nullptr) {
2112 // Unknown type.
2113 return Type::AnyPtr;
2114 } else if (_gvn.type(base->uncast()) == TypePtr::NULL_PTR) {
2115 // Since this is a null+long form, we have to switch to a rawptr.
2116 base = _gvn.transform(new CastX2PNode(offset));
2117 offset = MakeConX(0);
2118 return Type::RawPtr;
2119 } else if (base_type->base() == Type::RawPtr) {
2120 return Type::RawPtr;
2121 } else if (base_type->isa_oopptr()) {
2122 // Base is never null => always a heap address.
2123 if (!TypePtr::NULL_PTR->higher_equal(base_type)) {
2124 return Type::OopPtr;
2125 }
2126 // Offset is small => always a heap address.
2127 const TypeX* offset_type = _gvn.type(offset)->isa_intptr_t();
2128 if (offset_type != nullptr &&
2129 base_type->offset() == 0 && // (should always be?)
2130 offset_type->_lo >= 0 &&
2131 !MacroAssembler::needs_explicit_null_check(offset_type->_hi)) {
2132 return Type::OopPtr;
2133 } else if (type == T_OBJECT) {
2134 // off heap access to an oop doesn't make any sense. Has to be on
2135 // heap.
2136 return Type::OopPtr;
2137 }
2138 // Otherwise, it might either be oop+off or null+addr.
2139 return Type::AnyPtr;
2140 } else {
2141 // No information:
2142 return Type::AnyPtr;
2143 }
2144 }
2145
2146 Node* LibraryCallKit::make_unsafe_address(Node*& base, Node* offset, BasicType type, bool can_cast) {
2147 Node* uncasted_base = base;
2148 int kind = classify_unsafe_addr(uncasted_base, offset, type);
2149 if (kind == Type::RawPtr) {
2150 return basic_plus_adr(top(), uncasted_base, offset);
2151 } else if (kind == Type::AnyPtr) {
2152 assert(base == uncasted_base, "unexpected base change");
2153 if (can_cast) {
2154 if (!_gvn.type(base)->speculative_maybe_null() &&
2155 !too_many_traps(Deoptimization::Reason_speculate_null_check)) {
2156 // According to profiling, this access is always on
2157 // heap. Casting the base to not null and thus avoiding membars
2158 // around the access should allow better optimizations
2159 Node* null_ctl = top();
2160 base = null_check_oop(base, &null_ctl, true, true, true);
2161 assert(null_ctl->is_top(), "no null control here");
2162 return basic_plus_adr(base, offset);
2163 } else if (_gvn.type(base)->speculative_always_null() &&
2164 !too_many_traps(Deoptimization::Reason_speculate_null_assert)) {
2165 // According to profiling, this access is always off
2166 // heap.
2167 base = null_assert(base);
2168 Node* raw_base = _gvn.transform(new CastX2PNode(offset));
2169 offset = MakeConX(0);
2170 return basic_plus_adr(top(), raw_base, offset);
2171 }
2172 }
2173 // We don't know if it's an on heap or off heap access. Fall back
2174 // to raw memory access.
2175 Node* raw = _gvn.transform(new CheckCastPPNode(control(), base, TypeRawPtr::BOTTOM));
2176 return basic_plus_adr(top(), raw, offset);
2177 } else {
2178 assert(base == uncasted_base, "unexpected base change");
2179 // We know it's an on heap access so base can't be null
2180 if (TypePtr::NULL_PTR->higher_equal(_gvn.type(base))) {
2181 base = must_be_not_null(base, true);
2182 }
2183 return basic_plus_adr(base, offset);
2184 }
2185 }
2186
2187 //--------------------------inline_number_methods-----------------------------
2188 // inline int Integer.numberOfLeadingZeros(int)
2189 // inline int Long.numberOfLeadingZeros(long)
2190 //
2191 // inline int Integer.numberOfTrailingZeros(int)
2192 // inline int Long.numberOfTrailingZeros(long)
2193 //
2194 // inline int Integer.bitCount(int)
2195 // inline int Long.bitCount(long)
2196 //
2197 // inline char Character.reverseBytes(char)
2198 // inline short Short.reverseBytes(short)
2199 // inline int Integer.reverseBytes(int)
2200 // inline long Long.reverseBytes(long)
2201 bool LibraryCallKit::inline_number_methods(vmIntrinsics::ID id) {
2202 Node* arg = argument(0);
2203 Node* n = nullptr;
2204 switch (id) {
2205 case vmIntrinsics::_numberOfLeadingZeros_i: n = new CountLeadingZerosINode( arg); break;
2206 case vmIntrinsics::_numberOfLeadingZeros_l: n = new CountLeadingZerosLNode( arg); break;
2207 case vmIntrinsics::_numberOfTrailingZeros_i: n = new CountTrailingZerosINode(arg); break;
2208 case vmIntrinsics::_numberOfTrailingZeros_l: n = new CountTrailingZerosLNode(arg); break;
2209 case vmIntrinsics::_bitCount_i: n = new PopCountINode( arg); break;
2210 case vmIntrinsics::_bitCount_l: n = new PopCountLNode( arg); break;
2211 case vmIntrinsics::_reverseBytes_c: n = new ReverseBytesUSNode( arg); break;
2212 case vmIntrinsics::_reverseBytes_s: n = new ReverseBytesSNode( arg); break;
2213 case vmIntrinsics::_reverseBytes_i: n = new ReverseBytesINode( arg); break;
2214 case vmIntrinsics::_reverseBytes_l: n = new ReverseBytesLNode( arg); break;
2215 case vmIntrinsics::_reverse_i: n = new ReverseINode( arg); break;
2216 case vmIntrinsics::_reverse_l: n = new ReverseLNode( arg); break;
2217 default: fatal_unexpected_iid(id); break;
2218 }
2219 set_result(_gvn.transform(n));
2220 return true;
2221 }
2222
2223 //--------------------------inline_bitshuffle_methods-----------------------------
2224 // inline int Integer.compress(int, int)
2225 // inline int Integer.expand(int, int)
2226 // inline long Long.compress(long, long)
2227 // inline long Long.expand(long, long)
2228 bool LibraryCallKit::inline_bitshuffle_methods(vmIntrinsics::ID id) {
2229 Node* n = nullptr;
2230 switch (id) {
2231 case vmIntrinsics::_compress_i: n = new CompressBitsNode(argument(0), argument(1), TypeInt::INT); break;
2232 case vmIntrinsics::_expand_i: n = new ExpandBitsNode(argument(0), argument(1), TypeInt::INT); break;
2233 case vmIntrinsics::_compress_l: n = new CompressBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2234 case vmIntrinsics::_expand_l: n = new ExpandBitsNode(argument(0), argument(2), TypeLong::LONG); break;
2235 default: fatal_unexpected_iid(id); break;
2236 }
2237 set_result(_gvn.transform(n));
2238 return true;
2239 }
2240
2241 //--------------------------inline_number_methods-----------------------------
2242 // inline int Integer.compareUnsigned(int, int)
2243 // inline int Long.compareUnsigned(long, long)
2244 bool LibraryCallKit::inline_compare_unsigned(vmIntrinsics::ID id) {
2245 Node* arg1 = argument(0);
2246 Node* arg2 = (id == vmIntrinsics::_compareUnsigned_l) ? argument(2) : argument(1);
2247 Node* n = nullptr;
2248 switch (id) {
2249 case vmIntrinsics::_compareUnsigned_i: n = new CmpU3Node(arg1, arg2); break;
2250 case vmIntrinsics::_compareUnsigned_l: n = new CmpUL3Node(arg1, arg2); break;
2251 default: fatal_unexpected_iid(id); break;
2252 }
2253 set_result(_gvn.transform(n));
2254 return true;
2255 }
2256
2257 //--------------------------inline_unsigned_divmod_methods-----------------------------
2258 // inline int Integer.divideUnsigned(int, int)
2259 // inline int Integer.remainderUnsigned(int, int)
2260 // inline long Long.divideUnsigned(long, long)
2261 // inline long Long.remainderUnsigned(long, long)
2262 bool LibraryCallKit::inline_divmod_methods(vmIntrinsics::ID id) {
2263 Node* n = nullptr;
2264 switch (id) {
2265 case vmIntrinsics::_divideUnsigned_i: {
2266 zero_check_int(argument(1));
2267 // Compile-time detect of null-exception
2268 if (stopped()) {
2269 return true; // keep the graph constructed so far
2270 }
2271 n = new UDivINode(control(), argument(0), argument(1));
2272 break;
2273 }
2274 case vmIntrinsics::_divideUnsigned_l: {
2275 zero_check_long(argument(2));
2276 // Compile-time detect of null-exception
2277 if (stopped()) {
2278 return true; // keep the graph constructed so far
2279 }
2280 n = new UDivLNode(control(), argument(0), argument(2));
2281 break;
2282 }
2283 case vmIntrinsics::_remainderUnsigned_i: {
2284 zero_check_int(argument(1));
2285 // Compile-time detect of null-exception
2286 if (stopped()) {
2287 return true; // keep the graph constructed so far
2288 }
2289 n = new UModINode(control(), argument(0), argument(1));
2290 break;
2291 }
2292 case vmIntrinsics::_remainderUnsigned_l: {
2293 zero_check_long(argument(2));
2294 // Compile-time detect of null-exception
2295 if (stopped()) {
2296 return true; // keep the graph constructed so far
2297 }
2298 n = new UModLNode(control(), argument(0), argument(2));
2299 break;
2300 }
2301 default: fatal_unexpected_iid(id); break;
2302 }
2303 set_result(_gvn.transform(n));
2304 return true;
2305 }
2306
2307 //----------------------------inline_unsafe_access----------------------------
2308
2309 const TypeOopPtr* LibraryCallKit::sharpen_unsafe_type(Compile::AliasType* alias_type, const TypePtr *adr_type) {
2310 // Attempt to infer a sharper value type from the offset and base type.
2311 ciKlass* sharpened_klass = nullptr;
2312
2313 // See if it is an instance field, with an object type.
2314 if (alias_type->field() != nullptr) {
2315 if (alias_type->field()->type()->is_klass()) {
2316 sharpened_klass = alias_type->field()->type()->as_klass();
2317 }
2318 }
2319
2320 const TypeOopPtr* result = nullptr;
2321 // See if it is a narrow oop array.
2322 if (adr_type->isa_aryptr()) {
2323 if (adr_type->offset() >= objArrayOopDesc::base_offset_in_bytes()) {
2324 const TypeOopPtr* elem_type = adr_type->is_aryptr()->elem()->make_oopptr();
2325 if (elem_type != nullptr && elem_type->is_loaded()) {
2326 // Sharpen the value type.
2327 result = elem_type;
2328 }
2329 }
2330 }
2331
2332 // The sharpened class might be unloaded if there is no class loader
2333 // contraint in place.
2334 if (result == nullptr && sharpened_klass != nullptr && sharpened_klass->is_loaded()) {
2335 // Sharpen the value type.
2336 result = TypeOopPtr::make_from_klass(sharpened_klass);
2337 }
2338 if (result != nullptr) {
2339 #ifndef PRODUCT
2340 if (C->print_intrinsics() || C->print_inlining()) {
2341 tty->print(" from base type: "); adr_type->dump(); tty->cr();
2342 tty->print(" sharpened value: "); result->dump(); tty->cr();
2343 }
2344 #endif
2345 }
2346 return result;
2347 }
2348
2349 DecoratorSet LibraryCallKit::mo_decorator_for_access_kind(AccessKind kind) {
2350 switch (kind) {
2351 case Relaxed:
2352 return MO_UNORDERED;
2353 case Opaque:
2354 return MO_RELAXED;
2355 case Acquire:
2356 return MO_ACQUIRE;
2357 case Release:
2358 return MO_RELEASE;
2359 case Volatile:
2360 return MO_SEQ_CST;
2361 default:
2362 ShouldNotReachHere();
2363 return 0;
2364 }
2365 }
2366
2367 bool LibraryCallKit::inline_unsafe_access(bool is_store, const BasicType type, const AccessKind kind, const bool unaligned) {
2368 if (callee()->is_static()) return false; // caller must have the capability!
2369 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2370 guarantee(!is_store || kind != Acquire, "Acquire accesses can be produced only for loads");
2371 guarantee( is_store || kind != Release, "Release accesses can be produced only for stores");
2372 assert(type != T_OBJECT || !unaligned, "unaligned access not supported with object type");
2373
2374 if (is_reference_type(type)) {
2375 decorators |= ON_UNKNOWN_OOP_REF;
2376 }
2377
2378 if (unaligned) {
2379 decorators |= C2_UNALIGNED;
2380 }
2381
2382 #ifndef PRODUCT
2383 {
2384 ResourceMark rm;
2385 // Check the signatures.
2386 ciSignature* sig = callee()->signature();
2387 #ifdef ASSERT
2388 if (!is_store) {
2389 // Object getReference(Object base, int/long offset), etc.
2390 BasicType rtype = sig->return_type()->basic_type();
2391 assert(rtype == type, "getter must return the expected value");
2392 assert(sig->count() == 2, "oop getter has 2 arguments");
2393 assert(sig->type_at(0)->basic_type() == T_OBJECT, "getter base is object");
2394 assert(sig->type_at(1)->basic_type() == T_LONG, "getter offset is correct");
2395 } else {
2396 // void putReference(Object base, int/long offset, Object x), etc.
2397 assert(sig->return_type()->basic_type() == T_VOID, "putter must not return a value");
2398 assert(sig->count() == 3, "oop putter has 3 arguments");
2399 assert(sig->type_at(0)->basic_type() == T_OBJECT, "putter base is object");
2400 assert(sig->type_at(1)->basic_type() == T_LONG, "putter offset is correct");
2401 BasicType vtype = sig->type_at(sig->count()-1)->basic_type();
2402 assert(vtype == type, "putter must accept the expected value");
2403 }
2404 #endif // ASSERT
2405 }
2406 #endif //PRODUCT
2407
2408 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2409
2410 Node* receiver = argument(0); // type: oop
2411
2412 // Build address expression.
2413 Node* heap_base_oop = top();
2414
2415 // The base is either a Java object or a value produced by Unsafe.staticFieldBase
2416 Node* base = argument(1); // type: oop
2417 // The offset is a value produced by Unsafe.staticFieldOffset or Unsafe.objectFieldOffset
2418 Node* offset = argument(2); // type: long
2419 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2420 // to be plain byte offsets, which are also the same as those accepted
2421 // by oopDesc::field_addr.
2422 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
2423 "fieldOffset must be byte-scaled");
2424 // 32-bit machines ignore the high half!
2425 offset = ConvL2X(offset);
2426
2427 // Save state and restore on bailout
2428 uint old_sp = sp();
2429 SafePointNode* old_map = clone_map();
2430
2431 Node* adr = make_unsafe_address(base, offset, type, kind == Relaxed);
2432 assert(!stopped(), "Inlining of unsafe access failed: address construction stopped unexpectedly");
2433
2434 if (_gvn.type(base->uncast())->isa_ptr() == TypePtr::NULL_PTR) {
2435 if (type != T_OBJECT) {
2436 decorators |= IN_NATIVE; // off-heap primitive access
2437 } else {
2438 set_map(old_map);
2439 set_sp(old_sp);
2440 return false; // off-heap oop accesses are not supported
2441 }
2442 } else {
2443 heap_base_oop = base; // on-heap or mixed access
2444 }
2445
2446 // Can base be null? Otherwise, always on-heap access.
2447 bool can_access_non_heap = TypePtr::NULL_PTR->higher_equal(_gvn.type(base));
2448
2449 if (!can_access_non_heap) {
2450 decorators |= IN_HEAP;
2451 }
2452
2453 Node* val = is_store ? argument(4) : nullptr;
2454
2455 const TypePtr* adr_type = _gvn.type(adr)->isa_ptr();
2456 if (adr_type == TypePtr::NULL_PTR) {
2457 set_map(old_map);
2458 set_sp(old_sp);
2459 return false; // off-heap access with zero address
2460 }
2461
2462 // Try to categorize the address.
2463 Compile::AliasType* alias_type = C->alias_type(adr_type);
2464 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2465
2466 if (alias_type->adr_type() == TypeInstPtr::KLASS ||
2467 alias_type->adr_type() == TypeAryPtr::RANGE) {
2468 set_map(old_map);
2469 set_sp(old_sp);
2470 return false; // not supported
2471 }
2472
2473 bool mismatched = false;
2474 BasicType bt = alias_type->basic_type();
2475 if (bt != T_ILLEGAL) {
2476 assert(alias_type->adr_type()->is_oopptr(), "should be on-heap access");
2477 if (bt == T_BYTE && adr_type->isa_aryptr()) {
2478 // Alias type doesn't differentiate between byte[] and boolean[]).
2479 // Use address type to get the element type.
2480 bt = adr_type->is_aryptr()->elem()->array_element_basic_type();
2481 }
2482 if (is_reference_type(bt, true)) {
2483 // accessing an array field with getReference is not a mismatch
2484 bt = T_OBJECT;
2485 }
2486 if ((bt == T_OBJECT) != (type == T_OBJECT)) {
2487 // Don't intrinsify mismatched object accesses
2488 set_map(old_map);
2489 set_sp(old_sp);
2490 return false;
2491 }
2492 mismatched = (bt != type);
2493 } else if (alias_type->adr_type()->isa_oopptr()) {
2494 mismatched = true; // conservatively mark all "wide" on-heap accesses as mismatched
2495 }
2496
2497 destruct_map_clone(old_map);
2498 assert(!mismatched || alias_type->adr_type()->is_oopptr(), "off-heap access can't be mismatched");
2499
2500 if (mismatched) {
2501 decorators |= C2_MISMATCHED;
2502 }
2503
2504 // First guess at the value type.
2505 const Type *value_type = Type::get_const_basic_type(type);
2506
2507 // Figure out the memory ordering.
2508 decorators |= mo_decorator_for_access_kind(kind);
2509
2510 if (!is_store && type == T_OBJECT) {
2511 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2512 if (tjp != nullptr) {
2513 value_type = tjp;
2514 }
2515 }
2516
2517 receiver = null_check(receiver);
2518 if (stopped()) {
2519 return true;
2520 }
2521 // Heap pointers get a null-check from the interpreter,
2522 // as a courtesy. However, this is not guaranteed by Unsafe,
2523 // and it is not possible to fully distinguish unintended nulls
2524 // from intended ones in this API.
2525
2526 if (!is_store) {
2527 Node* p = nullptr;
2528 // Try to constant fold a load from a constant field
2529 ciField* field = alias_type->field();
2530 if (heap_base_oop != top() && field != nullptr && field->is_constant() && !mismatched) {
2531 // final or stable field
2532 p = make_constant_from_field(field, heap_base_oop);
2533 }
2534
2535 if (p == nullptr) { // Could not constant fold the load
2536 p = access_load_at(heap_base_oop, adr, adr_type, value_type, type, decorators);
2537 // Normalize the value returned by getBoolean in the following cases
2538 if (type == T_BOOLEAN &&
2539 (mismatched ||
2540 heap_base_oop == top() || // - heap_base_oop is null or
2541 (can_access_non_heap && field == nullptr)) // - heap_base_oop is potentially null
2542 // and the unsafe access is made to large offset
2543 // (i.e., larger than the maximum offset necessary for any
2544 // field access)
2545 ) {
2546 IdealKit ideal = IdealKit(this);
2547 #define __ ideal.
2548 IdealVariable normalized_result(ideal);
2549 __ declarations_done();
2550 __ set(normalized_result, p);
2551 __ if_then(p, BoolTest::ne, ideal.ConI(0));
2552 __ set(normalized_result, ideal.ConI(1));
2553 ideal.end_if();
2554 final_sync(ideal);
2555 p = __ value(normalized_result);
2556 #undef __
2557 }
2558 }
2559 if (type == T_ADDRESS) {
2560 p = gvn().transform(new CastP2XNode(nullptr, p));
2561 p = ConvX2UL(p);
2562 }
2563 // The load node has the control of the preceding MemBarCPUOrder. All
2564 // following nodes will have the control of the MemBarCPUOrder inserted at
2565 // the end of this method. So, pushing the load onto the stack at a later
2566 // point is fine.
2567 set_result(p);
2568 } else {
2569 if (bt == T_ADDRESS) {
2570 // Repackage the long as a pointer.
2571 val = ConvL2X(val);
2572 val = gvn().transform(new CastX2PNode(val));
2573 }
2574 access_store_at(heap_base_oop, adr, adr_type, val, value_type, type, decorators);
2575 }
2576
2577 return true;
2578 }
2579
2580 //----------------------------inline_unsafe_load_store----------------------------
2581 // This method serves a couple of different customers (depending on LoadStoreKind):
2582 //
2583 // LS_cmp_swap:
2584 //
2585 // boolean compareAndSetReference(Object o, long offset, Object expected, Object x);
2586 // boolean compareAndSetInt( Object o, long offset, int expected, int x);
2587 // boolean compareAndSetLong( Object o, long offset, long expected, long x);
2588 //
2589 // LS_cmp_swap_weak:
2590 //
2591 // boolean weakCompareAndSetReference( Object o, long offset, Object expected, Object x);
2592 // boolean weakCompareAndSetReferencePlain( Object o, long offset, Object expected, Object x);
2593 // boolean weakCompareAndSetReferenceAcquire(Object o, long offset, Object expected, Object x);
2594 // boolean weakCompareAndSetReferenceRelease(Object o, long offset, Object expected, Object x);
2595 //
2596 // boolean weakCompareAndSetInt( Object o, long offset, int expected, int x);
2597 // boolean weakCompareAndSetIntPlain( Object o, long offset, int expected, int x);
2598 // boolean weakCompareAndSetIntAcquire( Object o, long offset, int expected, int x);
2599 // boolean weakCompareAndSetIntRelease( Object o, long offset, int expected, int x);
2600 //
2601 // boolean weakCompareAndSetLong( Object o, long offset, long expected, long x);
2602 // boolean weakCompareAndSetLongPlain( Object o, long offset, long expected, long x);
2603 // boolean weakCompareAndSetLongAcquire( Object o, long offset, long expected, long x);
2604 // boolean weakCompareAndSetLongRelease( Object o, long offset, long expected, long x);
2605 //
2606 // LS_cmp_exchange:
2607 //
2608 // Object compareAndExchangeReferenceVolatile(Object o, long offset, Object expected, Object x);
2609 // Object compareAndExchangeReferenceAcquire( Object o, long offset, Object expected, Object x);
2610 // Object compareAndExchangeReferenceRelease( Object o, long offset, Object expected, Object x);
2611 //
2612 // Object compareAndExchangeIntVolatile( Object o, long offset, Object expected, Object x);
2613 // Object compareAndExchangeIntAcquire( Object o, long offset, Object expected, Object x);
2614 // Object compareAndExchangeIntRelease( Object o, long offset, Object expected, Object x);
2615 //
2616 // Object compareAndExchangeLongVolatile( Object o, long offset, Object expected, Object x);
2617 // Object compareAndExchangeLongAcquire( Object o, long offset, Object expected, Object x);
2618 // Object compareAndExchangeLongRelease( Object o, long offset, Object expected, Object x);
2619 //
2620 // LS_get_add:
2621 //
2622 // int getAndAddInt( Object o, long offset, int delta)
2623 // long getAndAddLong(Object o, long offset, long delta)
2624 //
2625 // LS_get_set:
2626 //
2627 // int getAndSet(Object o, long offset, int newValue)
2628 // long getAndSet(Object o, long offset, long newValue)
2629 // Object getAndSet(Object o, long offset, Object newValue)
2630 //
2631 bool LibraryCallKit::inline_unsafe_load_store(const BasicType type, const LoadStoreKind kind, const AccessKind access_kind) {
2632 // This basic scheme here is the same as inline_unsafe_access, but
2633 // differs in enough details that combining them would make the code
2634 // overly confusing. (This is a true fact! I originally combined
2635 // them, but even I was confused by it!) As much code/comments as
2636 // possible are retained from inline_unsafe_access though to make
2637 // the correspondences clearer. - dl
2638
2639 if (callee()->is_static()) return false; // caller must have the capability!
2640
2641 DecoratorSet decorators = C2_UNSAFE_ACCESS;
2642 decorators |= mo_decorator_for_access_kind(access_kind);
2643
2644 #ifndef PRODUCT
2645 BasicType rtype;
2646 {
2647 ResourceMark rm;
2648 // Check the signatures.
2649 ciSignature* sig = callee()->signature();
2650 rtype = sig->return_type()->basic_type();
2651 switch(kind) {
2652 case LS_get_add:
2653 case LS_get_set: {
2654 // Check the signatures.
2655 #ifdef ASSERT
2656 assert(rtype == type, "get and set must return the expected type");
2657 assert(sig->count() == 3, "get and set has 3 arguments");
2658 assert(sig->type_at(0)->basic_type() == T_OBJECT, "get and set base is object");
2659 assert(sig->type_at(1)->basic_type() == T_LONG, "get and set offset is long");
2660 assert(sig->type_at(2)->basic_type() == type, "get and set must take expected type as new value/delta");
2661 assert(access_kind == Volatile, "mo is not passed to intrinsic nodes in current implementation");
2662 #endif // ASSERT
2663 break;
2664 }
2665 case LS_cmp_swap:
2666 case LS_cmp_swap_weak: {
2667 // Check the signatures.
2668 #ifdef ASSERT
2669 assert(rtype == T_BOOLEAN, "CAS must return boolean");
2670 assert(sig->count() == 4, "CAS has 4 arguments");
2671 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2672 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2673 #endif // ASSERT
2674 break;
2675 }
2676 case LS_cmp_exchange: {
2677 // Check the signatures.
2678 #ifdef ASSERT
2679 assert(rtype == type, "CAS must return the expected type");
2680 assert(sig->count() == 4, "CAS has 4 arguments");
2681 assert(sig->type_at(0)->basic_type() == T_OBJECT, "CAS base is object");
2682 assert(sig->type_at(1)->basic_type() == T_LONG, "CAS offset is long");
2683 #endif // ASSERT
2684 break;
2685 }
2686 default:
2687 ShouldNotReachHere();
2688 }
2689 }
2690 #endif //PRODUCT
2691
2692 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
2693
2694 // Get arguments:
2695 Node* receiver = nullptr;
2696 Node* base = nullptr;
2697 Node* offset = nullptr;
2698 Node* oldval = nullptr;
2699 Node* newval = nullptr;
2700 switch(kind) {
2701 case LS_cmp_swap:
2702 case LS_cmp_swap_weak:
2703 case LS_cmp_exchange: {
2704 const bool two_slot_type = type2size[type] == 2;
2705 receiver = argument(0); // type: oop
2706 base = argument(1); // type: oop
2707 offset = argument(2); // type: long
2708 oldval = argument(4); // type: oop, int, or long
2709 newval = argument(two_slot_type ? 6 : 5); // type: oop, int, or long
2710 break;
2711 }
2712 case LS_get_add:
2713 case LS_get_set: {
2714 receiver = argument(0); // type: oop
2715 base = argument(1); // type: oop
2716 offset = argument(2); // type: long
2717 oldval = nullptr;
2718 newval = argument(4); // type: oop, int, or long
2719 break;
2720 }
2721 default:
2722 ShouldNotReachHere();
2723 }
2724
2725 // Build field offset expression.
2726 // We currently rely on the cookies produced by Unsafe.xxxFieldOffset
2727 // to be plain byte offsets, which are also the same as those accepted
2728 // by oopDesc::field_addr.
2729 assert(Unsafe_field_offset_to_byte_offset(11) == 11, "fieldOffset must be byte-scaled");
2730 // 32-bit machines ignore the high half of long offsets
2731 offset = ConvL2X(offset);
2732 // Save state and restore on bailout
2733 uint old_sp = sp();
2734 SafePointNode* old_map = clone_map();
2735 Node* adr = make_unsafe_address(base, offset,type, false);
2736 const TypePtr *adr_type = _gvn.type(adr)->isa_ptr();
2737
2738 Compile::AliasType* alias_type = C->alias_type(adr_type);
2739 BasicType bt = alias_type->basic_type();
2740 if (bt != T_ILLEGAL &&
2741 (is_reference_type(bt) != (type == T_OBJECT))) {
2742 // Don't intrinsify mismatched object accesses.
2743 set_map(old_map);
2744 set_sp(old_sp);
2745 return false;
2746 }
2747
2748 destruct_map_clone(old_map);
2749
2750 // For CAS, unlike inline_unsafe_access, there seems no point in
2751 // trying to refine types. Just use the coarse types here.
2752 assert(alias_type->index() != Compile::AliasIdxBot, "no bare pointers here");
2753 const Type *value_type = Type::get_const_basic_type(type);
2754
2755 switch (kind) {
2756 case LS_get_set:
2757 case LS_cmp_exchange: {
2758 if (type == T_OBJECT) {
2759 const TypeOopPtr* tjp = sharpen_unsafe_type(alias_type, adr_type);
2760 if (tjp != nullptr) {
2761 value_type = tjp;
2762 }
2763 }
2764 break;
2765 }
2766 case LS_cmp_swap:
2767 case LS_cmp_swap_weak:
2768 case LS_get_add:
2769 break;
2770 default:
2771 ShouldNotReachHere();
2772 }
2773
2774 // Null check receiver.
2775 receiver = null_check(receiver);
2776 if (stopped()) {
2777 return true;
2778 }
2779
2780 int alias_idx = C->get_alias_index(adr_type);
2781
2782 if (is_reference_type(type)) {
2783 decorators |= IN_HEAP | ON_UNKNOWN_OOP_REF;
2784
2785 // Transformation of a value which could be null pointer (CastPP #null)
2786 // could be delayed during Parse (for example, in adjust_map_after_if()).
2787 // Execute transformation here to avoid barrier generation in such case.
2788 if (_gvn.type(newval) == TypePtr::NULL_PTR)
2789 newval = _gvn.makecon(TypePtr::NULL_PTR);
2790
2791 if (oldval != nullptr && _gvn.type(oldval) == TypePtr::NULL_PTR) {
2792 // Refine the value to a null constant, when it is known to be null
2793 oldval = _gvn.makecon(TypePtr::NULL_PTR);
2794 }
2795 }
2796
2797 Node* result = nullptr;
2798 switch (kind) {
2799 case LS_cmp_exchange: {
2800 result = access_atomic_cmpxchg_val_at(base, adr, adr_type, alias_idx,
2801 oldval, newval, value_type, type, decorators);
2802 break;
2803 }
2804 case LS_cmp_swap_weak:
2805 decorators |= C2_WEAK_CMPXCHG;
2806 case LS_cmp_swap: {
2807 result = access_atomic_cmpxchg_bool_at(base, adr, adr_type, alias_idx,
2808 oldval, newval, value_type, type, decorators);
2809 break;
2810 }
2811 case LS_get_set: {
2812 result = access_atomic_xchg_at(base, adr, adr_type, alias_idx,
2813 newval, value_type, type, decorators);
2814 break;
2815 }
2816 case LS_get_add: {
2817 result = access_atomic_add_at(base, adr, adr_type, alias_idx,
2818 newval, value_type, type, decorators);
2819 break;
2820 }
2821 default:
2822 ShouldNotReachHere();
2823 }
2824
2825 assert(type2size[result->bottom_type()->basic_type()] == type2size[rtype], "result type should match");
2826 set_result(result);
2827 return true;
2828 }
2829
2830 bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
2831 // Regardless of form, don't allow previous ld/st to move down,
2832 // then issue acquire, release, or volatile mem_bar.
2833 insert_mem_bar(Op_MemBarCPUOrder);
2834 switch(id) {
2835 case vmIntrinsics::_loadFence:
2836 insert_mem_bar(Op_LoadFence);
2837 return true;
2838 case vmIntrinsics::_storeFence:
2839 insert_mem_bar(Op_StoreFence);
2840 return true;
2841 case vmIntrinsics::_storeStoreFence:
2842 insert_mem_bar(Op_StoreStoreFence);
2843 return true;
2844 case vmIntrinsics::_fullFence:
2845 insert_mem_bar(Op_MemBarVolatile);
2846 return true;
2847 default:
2848 fatal_unexpected_iid(id);
2849 return false;
2850 }
2851 }
2852
2853 bool LibraryCallKit::inline_onspinwait() {
2854 insert_mem_bar(Op_OnSpinWait);
2855 return true;
2856 }
2857
2858 bool LibraryCallKit::klass_needs_init_guard(Node* kls) {
2859 if (!kls->is_Con()) {
2860 return true;
2861 }
2862 const TypeInstKlassPtr* klsptr = kls->bottom_type()->isa_instklassptr();
2863 if (klsptr == nullptr) {
2864 return true;
2865 }
2866 ciInstanceKlass* ik = klsptr->instance_klass();
2867 // don't need a guard for a klass that is already initialized
2868 return !ik->is_initialized();
2869 }
2870
2871 //----------------------------inline_unsafe_writeback0-------------------------
2872 // public native void Unsafe.writeback0(long address)
2873 bool LibraryCallKit::inline_unsafe_writeback0() {
2874 if (!Matcher::has_match_rule(Op_CacheWB)) {
2875 return false;
2876 }
2877 #ifndef PRODUCT
2878 assert(Matcher::has_match_rule(Op_CacheWBPreSync), "found match rule for CacheWB but not CacheWBPreSync");
2879 assert(Matcher::has_match_rule(Op_CacheWBPostSync), "found match rule for CacheWB but not CacheWBPostSync");
2880 ciSignature* sig = callee()->signature();
2881 assert(sig->type_at(0)->basic_type() == T_LONG, "Unsafe_writeback0 address is long!");
2882 #endif
2883 null_check_receiver(); // null-check, then ignore
2884 Node *addr = argument(1);
2885 addr = new CastX2PNode(addr);
2886 addr = _gvn.transform(addr);
2887 Node *flush = new CacheWBNode(control(), memory(TypeRawPtr::BOTTOM), addr);
2888 flush = _gvn.transform(flush);
2889 set_memory(flush, TypeRawPtr::BOTTOM);
2890 return true;
2891 }
2892
2893 //----------------------------inline_unsafe_writeback0-------------------------
2894 // public native void Unsafe.writeback0(long address)
2895 bool LibraryCallKit::inline_unsafe_writebackSync0(bool is_pre) {
2896 if (is_pre && !Matcher::has_match_rule(Op_CacheWBPreSync)) {
2897 return false;
2898 }
2899 if (!is_pre && !Matcher::has_match_rule(Op_CacheWBPostSync)) {
2900 return false;
2901 }
2902 #ifndef PRODUCT
2903 assert(Matcher::has_match_rule(Op_CacheWB),
2904 (is_pre ? "found match rule for CacheWBPreSync but not CacheWB"
2905 : "found match rule for CacheWBPostSync but not CacheWB"));
2906
2907 #endif
2908 null_check_receiver(); // null-check, then ignore
2909 Node *sync;
2910 if (is_pre) {
2911 sync = new CacheWBPreSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2912 } else {
2913 sync = new CacheWBPostSyncNode(control(), memory(TypeRawPtr::BOTTOM));
2914 }
2915 sync = _gvn.transform(sync);
2916 set_memory(sync, TypeRawPtr::BOTTOM);
2917 return true;
2918 }
2919
2920 //----------------------------inline_unsafe_allocate---------------------------
2921 // public native Object Unsafe.allocateInstance(Class<?> cls);
2922 bool LibraryCallKit::inline_unsafe_allocate() {
2923
2924 #if INCLUDE_JVMTI
2925 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
2926 return false;
2927 }
2928 #endif //INCLUDE_JVMTI
2929
2930 if (callee()->is_static()) return false; // caller must have the capability!
2931
2932 null_check_receiver(); // null-check, then ignore
2933 Node* cls = null_check(argument(1));
2934 if (stopped()) return true;
2935
2936 Node* kls = load_klass_from_mirror(cls, false, nullptr, 0);
2937 kls = null_check(kls);
2938 if (stopped()) return true; // argument was like int.class
2939
2940 #if INCLUDE_JVMTI
2941 // Don't try to access new allocated obj in the intrinsic.
2942 // It causes perfomance issues even when jvmti event VmObjectAlloc is disabled.
2943 // Deoptimize and allocate in interpreter instead.
2944 Node* addr = makecon(TypeRawPtr::make((address) &JvmtiExport::_should_notify_object_alloc));
2945 Node* should_post_vm_object_alloc = make_load(this->control(), addr, TypeInt::INT, T_INT, MemNode::unordered);
2946 Node* chk = _gvn.transform(new CmpINode(should_post_vm_object_alloc, intcon(0)));
2947 Node* tst = _gvn.transform(new BoolNode(chk, BoolTest::eq));
2948 {
2949 BuildCutout unless(this, tst, PROB_MAX);
2950 uncommon_trap(Deoptimization::Reason_intrinsic,
2951 Deoptimization::Action_make_not_entrant);
2952 }
2953 if (stopped()) {
2954 return true;
2955 }
2956 #endif //INCLUDE_JVMTI
2957
2958 Node* test = nullptr;
2959 if (LibraryCallKit::klass_needs_init_guard(kls)) {
2960 // Note: The argument might still be an illegal value like
2961 // Serializable.class or Object[].class. The runtime will handle it.
2962 // But we must make an explicit check for initialization.
2963 Node* insp = basic_plus_adr(kls, in_bytes(InstanceKlass::init_state_offset()));
2964 // Use T_BOOLEAN for InstanceKlass::_init_state so the compiler
2965 // can generate code to load it as unsigned byte.
2966 Node* inst = make_load(nullptr, insp, TypeInt::UBYTE, T_BOOLEAN, MemNode::acquire);
2967 Node* bits = intcon(InstanceKlass::fully_initialized);
2968 test = _gvn.transform(new SubINode(inst, bits));
2969 // The 'test' is non-zero if we need to take a slow path.
2970 }
2971
2972 Node* obj = new_instance(kls, test);
2973 set_result(obj);
2974 return true;
2975 }
2976
2977 //------------------------inline_native_time_funcs--------------
2978 // inline code for System.currentTimeMillis() and System.nanoTime()
2979 // these have the same type and signature
2980 bool LibraryCallKit::inline_native_time_funcs(address funcAddr, const char* funcName) {
2981 const TypeFunc* tf = OptoRuntime::void_long_Type();
2982 const TypePtr* no_memory_effects = nullptr;
2983 Node* time = make_runtime_call(RC_LEAF, tf, funcAddr, funcName, no_memory_effects);
2984 Node* value = _gvn.transform(new ProjNode(time, TypeFunc::Parms+0));
2985 #ifdef ASSERT
2986 Node* value_top = _gvn.transform(new ProjNode(time, TypeFunc::Parms+1));
2987 assert(value_top == top(), "second value must be top");
2988 #endif
2989 set_result(value);
2990 return true;
2991 }
2992
2993
2994 #if INCLUDE_JVMTI
2995
2996 // When notifications are disabled then just update the VTMS transition bit and return.
2997 // Otherwise, the bit is updated in the given function call implementing JVMTI notification protocol.
2998 bool LibraryCallKit::inline_native_notify_jvmti_funcs(address funcAddr, const char* funcName, bool is_start, bool is_end) {
2999 if (!DoJVMTIVirtualThreadTransitions) {
3000 return true;
3001 }
3002 Node* vt_oop = _gvn.transform(must_be_not_null(argument(0), true)); // VirtualThread this argument
3003 IdealKit ideal(this);
3004
3005 Node* ONE = ideal.ConI(1);
3006 Node* hide = is_start ? ideal.ConI(0) : (is_end ? ideal.ConI(1) : _gvn.transform(argument(1)));
3007 Node* addr = makecon(TypeRawPtr::make((address)&JvmtiVTMSTransitionDisabler::_VTMS_notify_jvmti_events));
3008 Node* notify_jvmti_enabled = ideal.load(ideal.ctrl(), addr, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3009
3010 ideal.if_then(notify_jvmti_enabled, BoolTest::eq, ONE); {
3011 sync_kit(ideal);
3012 // if notifyJvmti enabled then make a call to the given SharedRuntime function
3013 const TypeFunc* tf = OptoRuntime::notify_jvmti_vthread_Type();
3014 make_runtime_call(RC_NO_LEAF, tf, funcAddr, funcName, TypePtr::BOTTOM, vt_oop, hide);
3015 ideal.sync_kit(this);
3016 } ideal.else_(); {
3017 // set hide value to the VTMS transition bit in current JavaThread and VirtualThread object
3018 Node* thread = ideal.thread();
3019 Node* jt_addr = basic_plus_adr(thread, in_bytes(JavaThread::is_in_VTMS_transition_offset()));
3020 Node* vt_addr = basic_plus_adr(vt_oop, java_lang_Thread::is_in_VTMS_transition_offset());
3021
3022 sync_kit(ideal);
3023 access_store_at(nullptr, jt_addr, _gvn.type(jt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3024 access_store_at(nullptr, vt_addr, _gvn.type(vt_addr)->is_ptr(), hide, _gvn.type(hide), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3025
3026 ideal.sync_kit(this);
3027 } ideal.end_if();
3028 final_sync(ideal);
3029
3030 return true;
3031 }
3032
3033 // Always update the is_disable_suspend bit.
3034 bool LibraryCallKit::inline_native_notify_jvmti_sync() {
3035 if (!DoJVMTIVirtualThreadTransitions) {
3036 return true;
3037 }
3038 IdealKit ideal(this);
3039
3040 {
3041 // unconditionally update the is_disable_suspend bit in current JavaThread
3042 Node* thread = ideal.thread();
3043 Node* arg = _gvn.transform(argument(0)); // argument for notification
3044 Node* addr = basic_plus_adr(thread, in_bytes(JavaThread::is_disable_suspend_offset()));
3045 const TypePtr *addr_type = _gvn.type(addr)->isa_ptr();
3046
3047 sync_kit(ideal);
3048 access_store_at(nullptr, addr, addr_type, arg, _gvn.type(arg), T_BOOLEAN, IN_NATIVE | MO_UNORDERED);
3049 ideal.sync_kit(this);
3050 }
3051 final_sync(ideal);
3052
3053 return true;
3054 }
3055
3056 #endif // INCLUDE_JVMTI
3057
3058 #ifdef JFR_HAVE_INTRINSICS
3059
3060 /**
3061 * if oop->klass != null
3062 * // normal class
3063 * epoch = _epoch_state ? 2 : 1
3064 * if oop->klass->trace_id & ((epoch << META_SHIFT) | epoch)) != epoch {
3065 * ... // enter slow path when the klass is first recorded or the epoch of JFR shifts
3066 * }
3067 * id = oop->klass->trace_id >> TRACE_ID_SHIFT // normal class path
3068 * else
3069 * // primitive class
3070 * if oop->array_klass != null
3071 * id = (oop->array_klass->trace_id >> TRACE_ID_SHIFT) + 1 // primitive class path
3072 * else
3073 * id = LAST_TYPE_ID + 1 // void class path
3074 * if (!signaled)
3075 * signaled = true
3076 */
3077 bool LibraryCallKit::inline_native_classID() {
3078 Node* cls = argument(0);
3079
3080 IdealKit ideal(this);
3081 #define __ ideal.
3082 IdealVariable result(ideal); __ declarations_done();
3083 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3084 basic_plus_adr(cls, java_lang_Class::klass_offset()),
3085 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3086
3087
3088 __ if_then(kls, BoolTest::ne, null()); {
3089 Node* kls_trace_id_addr = basic_plus_adr(kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3090 Node* kls_trace_id_raw = ideal.load(ideal.ctrl(), kls_trace_id_addr,TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3091
3092 Node* epoch_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_address()));
3093 Node* epoch = ideal.load(ideal.ctrl(), epoch_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw);
3094 epoch = _gvn.transform(new LShiftLNode(longcon(1), epoch));
3095 Node* mask = _gvn.transform(new LShiftLNode(epoch, intcon(META_SHIFT)));
3096 mask = _gvn.transform(new OrLNode(mask, epoch));
3097 Node* kls_trace_id_raw_and_mask = _gvn.transform(new AndLNode(kls_trace_id_raw, mask));
3098
3099 float unlikely = PROB_UNLIKELY(0.999);
3100 __ if_then(kls_trace_id_raw_and_mask, BoolTest::ne, epoch, unlikely); {
3101 sync_kit(ideal);
3102 make_runtime_call(RC_LEAF,
3103 OptoRuntime::class_id_load_barrier_Type(),
3104 CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::load_barrier),
3105 "class id load barrier",
3106 TypePtr::BOTTOM,
3107 kls);
3108 ideal.sync_kit(this);
3109 } __ end_if();
3110
3111 ideal.set(result, _gvn.transform(new URShiftLNode(kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT))));
3112 } __ else_(); {
3113 Node* array_kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(),
3114 basic_plus_adr(cls, java_lang_Class::array_klass_offset()),
3115 TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
3116 __ if_then(array_kls, BoolTest::ne, null()); {
3117 Node* array_kls_trace_id_addr = basic_plus_adr(array_kls, in_bytes(KLASS_TRACE_ID_OFFSET));
3118 Node* array_kls_trace_id_raw = ideal.load(ideal.ctrl(), array_kls_trace_id_addr, TypeLong::LONG, T_LONG, Compile::AliasIdxRaw);
3119 Node* array_kls_trace_id = _gvn.transform(new URShiftLNode(array_kls_trace_id_raw, ideal.ConI(TRACE_ID_SHIFT)));
3120 ideal.set(result, _gvn.transform(new AddLNode(array_kls_trace_id, longcon(1))));
3121 } __ else_(); {
3122 // void class case
3123 ideal.set(result, _gvn.transform(longcon(LAST_TYPE_ID + 1)));
3124 } __ end_if();
3125
3126 Node* signaled_flag_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::signal_address()));
3127 Node* signaled = ideal.load(ideal.ctrl(), signaled_flag_address, TypeInt::BOOL, T_BOOLEAN, Compile::AliasIdxRaw, true, MemNode::acquire);
3128 __ if_then(signaled, BoolTest::ne, ideal.ConI(1)); {
3129 ideal.store(ideal.ctrl(), signaled_flag_address, ideal.ConI(1), T_BOOLEAN, Compile::AliasIdxRaw, MemNode::release, true);
3130 } __ end_if();
3131 } __ end_if();
3132
3133 final_sync(ideal);
3134 set_result(ideal.value(result));
3135 #undef __
3136 return true;
3137 }
3138
3139 //------------------------inline_native_jvm_commit------------------
3140 bool LibraryCallKit::inline_native_jvm_commit() {
3141 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3142
3143 // Save input memory and i_o state.
3144 Node* input_memory_state = reset_memory();
3145 set_all_memory(input_memory_state);
3146 Node* input_io_state = i_o();
3147
3148 // TLS.
3149 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3150 // Jfr java buffer.
3151 Node* java_buffer_offset = _gvn.transform(new AddPNode(top(), tls_ptr, _gvn.transform(MakeConX(in_bytes(JAVA_BUFFER_OFFSET_JFR)))));
3152 Node* java_buffer = _gvn.transform(new LoadPNode(control(), input_memory_state, java_buffer_offset, TypePtr::BOTTOM, TypeRawPtr::NOTNULL, MemNode::unordered));
3153 Node* java_buffer_pos_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_POS_OFFSET)))));
3154
3155 // Load the current value of the notified field in the JfrThreadLocal.
3156 Node* notified_offset = basic_plus_adr(top(), tls_ptr, in_bytes(NOTIFY_OFFSET_JFR));
3157 Node* notified = make_load(control(), notified_offset, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3158
3159 // Test for notification.
3160 Node* notified_cmp = _gvn.transform(new CmpINode(notified, _gvn.intcon(1)));
3161 Node* test_notified = _gvn.transform(new BoolNode(notified_cmp, BoolTest::eq));
3162 IfNode* iff_notified = create_and_map_if(control(), test_notified, PROB_MIN, COUNT_UNKNOWN);
3163
3164 // True branch, is notified.
3165 Node* is_notified = _gvn.transform(new IfTrueNode(iff_notified));
3166 set_control(is_notified);
3167
3168 // Reset notified state.
3169 store_to_memory(control(), notified_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::unordered);
3170 Node* notified_reset_memory = reset_memory();
3171
3172 // 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.
3173 Node* current_pos_X = _gvn.transform(new LoadXNode(control(), input_memory_state, java_buffer_pos_offset, TypeRawPtr::NOTNULL, TypeX_X, MemNode::unordered));
3174 // Convert the machine-word to a long.
3175 Node* current_pos = _gvn.transform(ConvX2L(current_pos_X));
3176
3177 // False branch, not notified.
3178 Node* not_notified = _gvn.transform(new IfFalseNode(iff_notified));
3179 set_control(not_notified);
3180 set_all_memory(input_memory_state);
3181
3182 // Arg is the next position as a long.
3183 Node* arg = argument(0);
3184 // Convert long to machine-word.
3185 Node* next_pos_X = _gvn.transform(ConvL2X(arg));
3186
3187 // Store the next_position to the underlying jfr java buffer.
3188 store_to_memory(control(), java_buffer_pos_offset, next_pos_X, LP64_ONLY(T_LONG) NOT_LP64(T_INT), MemNode::release);
3189
3190 Node* commit_memory = reset_memory();
3191 set_all_memory(commit_memory);
3192
3193 // 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.
3194 Node* java_buffer_flags_offset = _gvn.transform(new AddPNode(top(), java_buffer, _gvn.transform(MakeConX(in_bytes(JFR_BUFFER_FLAGS_OFFSET)))));
3195 Node* flags = make_load(control(), java_buffer_flags_offset, TypeInt::UBYTE, T_BYTE, MemNode::unordered);
3196 Node* lease_constant = _gvn.transform(_gvn.intcon(4));
3197
3198 // And flags with lease constant.
3199 Node* lease = _gvn.transform(new AndINode(flags, lease_constant));
3200
3201 // Branch on lease to conditionalize returning the leased java buffer.
3202 Node* lease_cmp = _gvn.transform(new CmpINode(lease, lease_constant));
3203 Node* test_lease = _gvn.transform(new BoolNode(lease_cmp, BoolTest::eq));
3204 IfNode* iff_lease = create_and_map_if(control(), test_lease, PROB_MIN, COUNT_UNKNOWN);
3205
3206 // False branch, not a lease.
3207 Node* not_lease = _gvn.transform(new IfFalseNode(iff_lease));
3208
3209 // True branch, is lease.
3210 Node* is_lease = _gvn.transform(new IfTrueNode(iff_lease));
3211 set_control(is_lease);
3212
3213 // Make a runtime call, which can safepoint, to return the leased buffer. This updates both the JfrThreadLocal and the Java event writer oop.
3214 Node* call_return_lease = make_runtime_call(RC_NO_LEAF,
3215 OptoRuntime::void_void_Type(),
3216 SharedRuntime::jfr_return_lease(),
3217 "return_lease", TypePtr::BOTTOM);
3218 Node* call_return_lease_control = _gvn.transform(new ProjNode(call_return_lease, TypeFunc::Control));
3219
3220 RegionNode* lease_compare_rgn = new RegionNode(PATH_LIMIT);
3221 record_for_igvn(lease_compare_rgn);
3222 PhiNode* lease_compare_mem = new PhiNode(lease_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3223 record_for_igvn(lease_compare_mem);
3224 PhiNode* lease_compare_io = new PhiNode(lease_compare_rgn, Type::ABIO);
3225 record_for_igvn(lease_compare_io);
3226 PhiNode* lease_result_value = new PhiNode(lease_compare_rgn, TypeLong::LONG);
3227 record_for_igvn(lease_result_value);
3228
3229 // Update control and phi nodes.
3230 lease_compare_rgn->init_req(_true_path, call_return_lease_control);
3231 lease_compare_rgn->init_req(_false_path, not_lease);
3232
3233 lease_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3234 lease_compare_mem->init_req(_false_path, commit_memory);
3235
3236 lease_compare_io->init_req(_true_path, i_o());
3237 lease_compare_io->init_req(_false_path, input_io_state);
3238
3239 lease_result_value->init_req(_true_path, null()); // if the lease was returned, return 0.
3240 lease_result_value->init_req(_false_path, arg); // if not lease, return new updated position.
3241
3242 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3243 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3244 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3245 PhiNode* result_value = new PhiNode(result_rgn, TypeLong::LONG);
3246
3247 // Update control and phi nodes.
3248 result_rgn->init_req(_true_path, is_notified);
3249 result_rgn->init_req(_false_path, _gvn.transform(lease_compare_rgn));
3250
3251 result_mem->init_req(_true_path, notified_reset_memory);
3252 result_mem->init_req(_false_path, _gvn.transform(lease_compare_mem));
3253
3254 result_io->init_req(_true_path, input_io_state);
3255 result_io->init_req(_false_path, _gvn.transform(lease_compare_io));
3256
3257 result_value->init_req(_true_path, current_pos);
3258 result_value->init_req(_false_path, _gvn.transform(lease_result_value));
3259
3260 // Set output state.
3261 set_control(_gvn.transform(result_rgn));
3262 set_all_memory(_gvn.transform(result_mem));
3263 set_i_o(_gvn.transform(result_io));
3264 set_result(result_rgn, result_value);
3265 return true;
3266 }
3267
3268 /*
3269 * The intrinsic is a model of this pseudo-code:
3270 *
3271 * JfrThreadLocal* const tl = Thread::jfr_thread_local()
3272 * jobject h_event_writer = tl->java_event_writer();
3273 * if (h_event_writer == nullptr) {
3274 * return nullptr;
3275 * }
3276 * oop threadObj = Thread::threadObj();
3277 * oop vthread = java_lang_Thread::vthread(threadObj);
3278 * traceid tid;
3279 * bool pinVirtualThread;
3280 * bool excluded;
3281 * if (vthread != threadObj) { // i.e. current thread is virtual
3282 * tid = java_lang_Thread::tid(vthread);
3283 * u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(vthread);
3284 * pinVirtualThread = VMContinuations;
3285 * excluded = vthread_epoch_raw & excluded_mask;
3286 * if (!excluded) {
3287 * traceid current_epoch = JfrTraceIdEpoch::current_generation();
3288 * u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3289 * if (vthread_epoch != current_epoch) {
3290 * write_checkpoint();
3291 * }
3292 * }
3293 * } else {
3294 * tid = java_lang_Thread::tid(threadObj);
3295 * u2 thread_epoch_raw = java_lang_Thread::jfr_epoch(threadObj);
3296 * pinVirtualThread = false;
3297 * excluded = thread_epoch_raw & excluded_mask;
3298 * }
3299 * oop event_writer = JNIHandles::resolve_non_null(h_event_writer);
3300 * traceid tid_in_event_writer = getField(event_writer, "threadID");
3301 * if (tid_in_event_writer != tid) {
3302 * setField(event_writer, "pinVirtualThread", pinVirtualThread);
3303 * setField(event_writer, "excluded", excluded);
3304 * setField(event_writer, "threadID", tid);
3305 * }
3306 * return event_writer
3307 */
3308 bool LibraryCallKit::inline_native_getEventWriter() {
3309 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3310
3311 // Save input memory and i_o state.
3312 Node* input_memory_state = reset_memory();
3313 set_all_memory(input_memory_state);
3314 Node* input_io_state = i_o();
3315
3316 // The most significant bit of the u2 is used to denote thread exclusion
3317 Node* excluded_shift = _gvn.intcon(15);
3318 Node* excluded_mask = _gvn.intcon(1 << 15);
3319 // The epoch generation is the range [1-32767]
3320 Node* epoch_mask = _gvn.intcon(32767);
3321
3322 // TLS
3323 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3324
3325 // Load the address of java event writer jobject handle from the jfr_thread_local structure.
3326 Node* jobj_ptr = basic_plus_adr(top(), tls_ptr, in_bytes(THREAD_LOCAL_WRITER_OFFSET_JFR));
3327
3328 // Load the eventwriter jobject handle.
3329 Node* jobj = make_load(control(), jobj_ptr, TypeRawPtr::BOTTOM, T_ADDRESS, MemNode::unordered);
3330
3331 // Null check the jobject handle.
3332 Node* jobj_cmp_null = _gvn.transform(new CmpPNode(jobj, null()));
3333 Node* test_jobj_not_equal_null = _gvn.transform(new BoolNode(jobj_cmp_null, BoolTest::ne));
3334 IfNode* iff_jobj_not_equal_null = create_and_map_if(control(), test_jobj_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3335
3336 // False path, jobj is null.
3337 Node* jobj_is_null = _gvn.transform(new IfFalseNode(iff_jobj_not_equal_null));
3338
3339 // True path, jobj is not null.
3340 Node* jobj_is_not_null = _gvn.transform(new IfTrueNode(iff_jobj_not_equal_null));
3341
3342 set_control(jobj_is_not_null);
3343
3344 // Load the threadObj for the CarrierThread.
3345 Node* threadObj = generate_current_thread(tls_ptr);
3346
3347 // Load the vthread.
3348 Node* vthread = generate_virtual_thread(tls_ptr);
3349
3350 // If vthread != threadObj, this is a virtual thread.
3351 Node* vthread_cmp_threadObj = _gvn.transform(new CmpPNode(vthread, threadObj));
3352 Node* test_vthread_not_equal_threadObj = _gvn.transform(new BoolNode(vthread_cmp_threadObj, BoolTest::ne));
3353 IfNode* iff_vthread_not_equal_threadObj =
3354 create_and_map_if(jobj_is_not_null, test_vthread_not_equal_threadObj, PROB_FAIR, COUNT_UNKNOWN);
3355
3356 // False branch, fallback to threadObj.
3357 Node* vthread_equal_threadObj = _gvn.transform(new IfFalseNode(iff_vthread_not_equal_threadObj));
3358 set_control(vthread_equal_threadObj);
3359
3360 // Load the tid field from the vthread object.
3361 Node* thread_obj_tid = load_field_from_object(threadObj, "tid", "J");
3362
3363 // Load the raw epoch value from the threadObj.
3364 Node* threadObj_epoch_offset = basic_plus_adr(threadObj, java_lang_Thread::jfr_epoch_offset());
3365 Node* threadObj_epoch_raw = access_load_at(threadObj, threadObj_epoch_offset,
3366 _gvn.type(threadObj_epoch_offset)->isa_ptr(),
3367 TypeInt::CHAR, T_CHAR,
3368 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3369
3370 // Mask off the excluded information from the epoch.
3371 Node * threadObj_is_excluded = _gvn.transform(new AndINode(threadObj_epoch_raw, excluded_mask));
3372
3373 // True branch, this is a virtual thread.
3374 Node* vthread_not_equal_threadObj = _gvn.transform(new IfTrueNode(iff_vthread_not_equal_threadObj));
3375 set_control(vthread_not_equal_threadObj);
3376
3377 // Load the tid field from the vthread object.
3378 Node* vthread_tid = load_field_from_object(vthread, "tid", "J");
3379
3380 // Continuation support determines if a virtual thread should be pinned.
3381 Node* global_addr = makecon(TypeRawPtr::make((address)&VMContinuations));
3382 Node* continuation_support = make_load(control(), global_addr, TypeInt::BOOL, T_BOOLEAN, MemNode::unordered);
3383
3384 // Load the raw epoch value from the vthread.
3385 Node* vthread_epoch_offset = basic_plus_adr(vthread, java_lang_Thread::jfr_epoch_offset());
3386 Node* vthread_epoch_raw = access_load_at(vthread, vthread_epoch_offset, _gvn.type(vthread_epoch_offset)->is_ptr(),
3387 TypeInt::CHAR, T_CHAR,
3388 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3389
3390 // Mask off the excluded information from the epoch.
3391 Node * vthread_is_excluded = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(excluded_mask)));
3392
3393 // Branch on excluded to conditionalize updating the epoch for the virtual thread.
3394 Node* is_excluded_cmp = _gvn.transform(new CmpINode(vthread_is_excluded, _gvn.transform(excluded_mask)));
3395 Node* test_not_excluded = _gvn.transform(new BoolNode(is_excluded_cmp, BoolTest::ne));
3396 IfNode* iff_not_excluded = create_and_map_if(control(), test_not_excluded, PROB_MAX, COUNT_UNKNOWN);
3397
3398 // False branch, vthread is excluded, no need to write epoch info.
3399 Node* excluded = _gvn.transform(new IfFalseNode(iff_not_excluded));
3400
3401 // True branch, vthread is included, update epoch info.
3402 Node* included = _gvn.transform(new IfTrueNode(iff_not_excluded));
3403 set_control(included);
3404
3405 // Get epoch value.
3406 Node* epoch = _gvn.transform(new AndINode(vthread_epoch_raw, _gvn.transform(epoch_mask)));
3407
3408 // Load the current epoch generation. The value is unsigned 16-bit, so we type it as T_CHAR.
3409 Node* epoch_generation_address = makecon(TypeRawPtr::make(JfrIntrinsicSupport::epoch_generation_address()));
3410 Node* current_epoch_generation = make_load(control(), epoch_generation_address, TypeInt::CHAR, T_CHAR, MemNode::unordered);
3411
3412 // Compare the epoch in the vthread to the current epoch generation.
3413 Node* const epoch_cmp = _gvn.transform(new CmpUNode(current_epoch_generation, epoch));
3414 Node* test_epoch_not_equal = _gvn.transform(new BoolNode(epoch_cmp, BoolTest::ne));
3415 IfNode* iff_epoch_not_equal = create_and_map_if(control(), test_epoch_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3416
3417 // False path, epoch is equal, checkpoint information is valid.
3418 Node* epoch_is_equal = _gvn.transform(new IfFalseNode(iff_epoch_not_equal));
3419
3420 // True path, epoch is not equal, write a checkpoint for the vthread.
3421 Node* epoch_is_not_equal = _gvn.transform(new IfTrueNode(iff_epoch_not_equal));
3422
3423 set_control(epoch_is_not_equal);
3424
3425 // Make a runtime call, which can safepoint, to write a checkpoint for the vthread for this epoch.
3426 // The call also updates the native thread local thread id and the vthread with the current epoch.
3427 Node* call_write_checkpoint = make_runtime_call(RC_NO_LEAF,
3428 OptoRuntime::jfr_write_checkpoint_Type(),
3429 SharedRuntime::jfr_write_checkpoint(),
3430 "write_checkpoint", TypePtr::BOTTOM);
3431 Node* call_write_checkpoint_control = _gvn.transform(new ProjNode(call_write_checkpoint, TypeFunc::Control));
3432
3433 // vthread epoch != current epoch
3434 RegionNode* epoch_compare_rgn = new RegionNode(PATH_LIMIT);
3435 record_for_igvn(epoch_compare_rgn);
3436 PhiNode* epoch_compare_mem = new PhiNode(epoch_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3437 record_for_igvn(epoch_compare_mem);
3438 PhiNode* epoch_compare_io = new PhiNode(epoch_compare_rgn, Type::ABIO);
3439 record_for_igvn(epoch_compare_io);
3440
3441 // Update control and phi nodes.
3442 epoch_compare_rgn->init_req(_true_path, call_write_checkpoint_control);
3443 epoch_compare_rgn->init_req(_false_path, epoch_is_equal);
3444 epoch_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3445 epoch_compare_mem->init_req(_false_path, input_memory_state);
3446 epoch_compare_io->init_req(_true_path, i_o());
3447 epoch_compare_io->init_req(_false_path, input_io_state);
3448
3449 // excluded != true
3450 RegionNode* exclude_compare_rgn = new RegionNode(PATH_LIMIT);
3451 record_for_igvn(exclude_compare_rgn);
3452 PhiNode* exclude_compare_mem = new PhiNode(exclude_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3453 record_for_igvn(exclude_compare_mem);
3454 PhiNode* exclude_compare_io = new PhiNode(exclude_compare_rgn, Type::ABIO);
3455 record_for_igvn(exclude_compare_io);
3456
3457 // Update control and phi nodes.
3458 exclude_compare_rgn->init_req(_true_path, _gvn.transform(epoch_compare_rgn));
3459 exclude_compare_rgn->init_req(_false_path, excluded);
3460 exclude_compare_mem->init_req(_true_path, _gvn.transform(epoch_compare_mem));
3461 exclude_compare_mem->init_req(_false_path, input_memory_state);
3462 exclude_compare_io->init_req(_true_path, _gvn.transform(epoch_compare_io));
3463 exclude_compare_io->init_req(_false_path, input_io_state);
3464
3465 // vthread != threadObj
3466 RegionNode* vthread_compare_rgn = new RegionNode(PATH_LIMIT);
3467 record_for_igvn(vthread_compare_rgn);
3468 PhiNode* vthread_compare_mem = new PhiNode(vthread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3469 PhiNode* vthread_compare_io = new PhiNode(vthread_compare_rgn, Type::ABIO);
3470 record_for_igvn(vthread_compare_io);
3471 PhiNode* tid = new PhiNode(vthread_compare_rgn, TypeLong::LONG);
3472 record_for_igvn(tid);
3473 PhiNode* exclusion = new PhiNode(vthread_compare_rgn, TypeInt::CHAR);
3474 record_for_igvn(exclusion);
3475 PhiNode* pinVirtualThread = new PhiNode(vthread_compare_rgn, TypeInt::BOOL);
3476 record_for_igvn(pinVirtualThread);
3477
3478 // Update control and phi nodes.
3479 vthread_compare_rgn->init_req(_true_path, _gvn.transform(exclude_compare_rgn));
3480 vthread_compare_rgn->init_req(_false_path, vthread_equal_threadObj);
3481 vthread_compare_mem->init_req(_true_path, _gvn.transform(exclude_compare_mem));
3482 vthread_compare_mem->init_req(_false_path, input_memory_state);
3483 vthread_compare_io->init_req(_true_path, _gvn.transform(exclude_compare_io));
3484 vthread_compare_io->init_req(_false_path, input_io_state);
3485 tid->init_req(_true_path, _gvn.transform(vthread_tid));
3486 tid->init_req(_false_path, _gvn.transform(thread_obj_tid));
3487 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3488 exclusion->init_req(_false_path, _gvn.transform(threadObj_is_excluded));
3489 pinVirtualThread->init_req(_true_path, _gvn.transform(continuation_support));
3490 pinVirtualThread->init_req(_false_path, _gvn.intcon(0));
3491
3492 // Update branch state.
3493 set_control(_gvn.transform(vthread_compare_rgn));
3494 set_all_memory(_gvn.transform(vthread_compare_mem));
3495 set_i_o(_gvn.transform(vthread_compare_io));
3496
3497 // Load the event writer oop by dereferencing the jobject handle.
3498 ciKlass* klass_EventWriter = env()->find_system_klass(ciSymbol::make("jdk/jfr/internal/event/EventWriter"));
3499 assert(klass_EventWriter->is_loaded(), "invariant");
3500 ciInstanceKlass* const instklass_EventWriter = klass_EventWriter->as_instance_klass();
3501 const TypeKlassPtr* const aklass = TypeKlassPtr::make(instklass_EventWriter);
3502 const TypeOopPtr* const xtype = aklass->as_instance_type();
3503 Node* jobj_untagged = _gvn.transform(new AddPNode(top(), jobj, _gvn.MakeConX(-JNIHandles::TypeTag::global)));
3504 Node* event_writer = access_load(jobj_untagged, xtype, T_OBJECT, IN_NATIVE | C2_CONTROL_DEPENDENT_LOAD);
3505
3506 // Load the current thread id from the event writer object.
3507 Node* const event_writer_tid = load_field_from_object(event_writer, "threadID", "J");
3508 // Get the field offset to, conditionally, store an updated tid value later.
3509 Node* const event_writer_tid_field = field_address_from_object(event_writer, "threadID", "J", false);
3510 // Get the field offset to, conditionally, store an updated exclusion value later.
3511 Node* const event_writer_excluded_field = field_address_from_object(event_writer, "excluded", "Z", false);
3512 // Get the field offset to, conditionally, store an updated pinVirtualThread value later.
3513 Node* const event_writer_pin_field = field_address_from_object(event_writer, "pinVirtualThread", "Z", false);
3514
3515 RegionNode* event_writer_tid_compare_rgn = new RegionNode(PATH_LIMIT);
3516 record_for_igvn(event_writer_tid_compare_rgn);
3517 PhiNode* event_writer_tid_compare_mem = new PhiNode(event_writer_tid_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3518 record_for_igvn(event_writer_tid_compare_mem);
3519 PhiNode* event_writer_tid_compare_io = new PhiNode(event_writer_tid_compare_rgn, Type::ABIO);
3520 record_for_igvn(event_writer_tid_compare_io);
3521
3522 // Compare the current tid from the thread object to what is currently stored in the event writer object.
3523 Node* const tid_cmp = _gvn.transform(new CmpLNode(event_writer_tid, _gvn.transform(tid)));
3524 Node* test_tid_not_equal = _gvn.transform(new BoolNode(tid_cmp, BoolTest::ne));
3525 IfNode* iff_tid_not_equal = create_and_map_if(_gvn.transform(vthread_compare_rgn), test_tid_not_equal, PROB_FAIR, COUNT_UNKNOWN);
3526
3527 // False path, tids are the same.
3528 Node* tid_is_equal = _gvn.transform(new IfFalseNode(iff_tid_not_equal));
3529
3530 // True path, tid is not equal, need to update the tid in the event writer.
3531 Node* tid_is_not_equal = _gvn.transform(new IfTrueNode(iff_tid_not_equal));
3532 record_for_igvn(tid_is_not_equal);
3533
3534 // Store the pin state to the event writer.
3535 store_to_memory(tid_is_not_equal, event_writer_pin_field, _gvn.transform(pinVirtualThread), T_BOOLEAN, MemNode::unordered);
3536
3537 // Store the exclusion state to the event writer.
3538 Node* excluded_bool = _gvn.transform(new URShiftINode(_gvn.transform(exclusion), excluded_shift));
3539 store_to_memory(tid_is_not_equal, event_writer_excluded_field, excluded_bool, T_BOOLEAN, MemNode::unordered);
3540
3541 // Store the tid to the event writer.
3542 store_to_memory(tid_is_not_equal, event_writer_tid_field, tid, T_LONG, MemNode::unordered);
3543
3544 // Update control and phi nodes.
3545 event_writer_tid_compare_rgn->init_req(_true_path, tid_is_not_equal);
3546 event_writer_tid_compare_rgn->init_req(_false_path, tid_is_equal);
3547 event_writer_tid_compare_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3548 event_writer_tid_compare_mem->init_req(_false_path, _gvn.transform(vthread_compare_mem));
3549 event_writer_tid_compare_io->init_req(_true_path, _gvn.transform(i_o()));
3550 event_writer_tid_compare_io->init_req(_false_path, _gvn.transform(vthread_compare_io));
3551
3552 // Result of top level CFG, Memory, IO and Value.
3553 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3554 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3555 PhiNode* result_io = new PhiNode(result_rgn, Type::ABIO);
3556 PhiNode* result_value = new PhiNode(result_rgn, TypeInstPtr::BOTTOM);
3557
3558 // Result control.
3559 result_rgn->init_req(_true_path, _gvn.transform(event_writer_tid_compare_rgn));
3560 result_rgn->init_req(_false_path, jobj_is_null);
3561
3562 // Result memory.
3563 result_mem->init_req(_true_path, _gvn.transform(event_writer_tid_compare_mem));
3564 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3565
3566 // Result IO.
3567 result_io->init_req(_true_path, _gvn.transform(event_writer_tid_compare_io));
3568 result_io->init_req(_false_path, _gvn.transform(input_io_state));
3569
3570 // Result value.
3571 result_value->init_req(_true_path, _gvn.transform(event_writer)); // return event writer oop
3572 result_value->init_req(_false_path, null()); // return null
3573
3574 // Set output state.
3575 set_control(_gvn.transform(result_rgn));
3576 set_all_memory(_gvn.transform(result_mem));
3577 set_i_o(_gvn.transform(result_io));
3578 set_result(result_rgn, result_value);
3579 return true;
3580 }
3581
3582 /*
3583 * The intrinsic is a model of this pseudo-code:
3584 *
3585 * JfrThreadLocal* const tl = thread->jfr_thread_local();
3586 * if (carrierThread != thread) { // is virtual thread
3587 * const u2 vthread_epoch_raw = java_lang_Thread::jfr_epoch(thread);
3588 * bool excluded = vthread_epoch_raw & excluded_mask;
3589 * Atomic::store(&tl->_contextual_tid, java_lang_Thread::tid(thread));
3590 * Atomic::store(&tl->_contextual_thread_excluded, is_excluded);
3591 * if (!excluded) {
3592 * const u2 vthread_epoch = vthread_epoch_raw & epoch_mask;
3593 * Atomic::store(&tl->_vthread_epoch, vthread_epoch);
3594 * }
3595 * Atomic::release_store(&tl->_vthread, true);
3596 * return;
3597 * }
3598 * Atomic::release_store(&tl->_vthread, false);
3599 */
3600 void LibraryCallKit::extend_setCurrentThread(Node* jt, Node* thread) {
3601 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3602
3603 Node* input_memory_state = reset_memory();
3604 set_all_memory(input_memory_state);
3605
3606 // The most significant bit of the u2 is used to denote thread exclusion
3607 Node* excluded_mask = _gvn.intcon(1 << 15);
3608 // The epoch generation is the range [1-32767]
3609 Node* epoch_mask = _gvn.intcon(32767);
3610
3611 Node* const carrierThread = generate_current_thread(jt);
3612 // If thread != carrierThread, this is a virtual thread.
3613 Node* thread_cmp_carrierThread = _gvn.transform(new CmpPNode(thread, carrierThread));
3614 Node* test_thread_not_equal_carrierThread = _gvn.transform(new BoolNode(thread_cmp_carrierThread, BoolTest::ne));
3615 IfNode* iff_thread_not_equal_carrierThread =
3616 create_and_map_if(control(), test_thread_not_equal_carrierThread, PROB_FAIR, COUNT_UNKNOWN);
3617
3618 Node* vthread_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_OFFSET_JFR));
3619
3620 // False branch, is carrierThread.
3621 Node* thread_equal_carrierThread = _gvn.transform(new IfFalseNode(iff_thread_not_equal_carrierThread));
3622 // Store release
3623 Node* vthread_false_memory = store_to_memory(thread_equal_carrierThread, vthread_offset, _gvn.intcon(0), T_BOOLEAN, MemNode::release, true);
3624
3625 set_all_memory(input_memory_state);
3626
3627 // True branch, is virtual thread.
3628 Node* thread_not_equal_carrierThread = _gvn.transform(new IfTrueNode(iff_thread_not_equal_carrierThread));
3629 set_control(thread_not_equal_carrierThread);
3630
3631 // Load the raw epoch value from the vthread.
3632 Node* epoch_offset = basic_plus_adr(thread, java_lang_Thread::jfr_epoch_offset());
3633 Node* epoch_raw = access_load_at(thread, epoch_offset, _gvn.type(epoch_offset)->is_ptr(), TypeInt::CHAR, T_CHAR,
3634 IN_HEAP | MO_UNORDERED | C2_MISMATCHED | C2_CONTROL_DEPENDENT_LOAD);
3635
3636 // Mask off the excluded information from the epoch.
3637 Node * const is_excluded = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(excluded_mask)));
3638
3639 // Load the tid field from the thread.
3640 Node* tid = load_field_from_object(thread, "tid", "J");
3641
3642 // Store the vthread tid to the jfr thread local.
3643 Node* thread_id_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_ID_OFFSET_JFR));
3644 Node* tid_memory = store_to_memory(control(), thread_id_offset, tid, T_LONG, MemNode::unordered, true);
3645
3646 // Branch is_excluded to conditionalize updating the epoch .
3647 Node* excluded_cmp = _gvn.transform(new CmpINode(is_excluded, _gvn.transform(excluded_mask)));
3648 Node* test_excluded = _gvn.transform(new BoolNode(excluded_cmp, BoolTest::eq));
3649 IfNode* iff_excluded = create_and_map_if(control(), test_excluded, PROB_MIN, COUNT_UNKNOWN);
3650
3651 // True branch, vthread is excluded, no need to write epoch info.
3652 Node* excluded = _gvn.transform(new IfTrueNode(iff_excluded));
3653 set_control(excluded);
3654 Node* vthread_is_excluded = _gvn.intcon(1);
3655
3656 // False branch, vthread is included, update epoch info.
3657 Node* included = _gvn.transform(new IfFalseNode(iff_excluded));
3658 set_control(included);
3659 Node* vthread_is_included = _gvn.intcon(0);
3660
3661 // Get epoch value.
3662 Node* epoch = _gvn.transform(new AndINode(epoch_raw, _gvn.transform(epoch_mask)));
3663
3664 // Store the vthread epoch to the jfr thread local.
3665 Node* vthread_epoch_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EPOCH_OFFSET_JFR));
3666 Node* included_memory = store_to_memory(control(), vthread_epoch_offset, epoch, T_CHAR, MemNode::unordered, true);
3667
3668 RegionNode* excluded_rgn = new RegionNode(PATH_LIMIT);
3669 record_for_igvn(excluded_rgn);
3670 PhiNode* excluded_mem = new PhiNode(excluded_rgn, Type::MEMORY, TypePtr::BOTTOM);
3671 record_for_igvn(excluded_mem);
3672 PhiNode* exclusion = new PhiNode(excluded_rgn, TypeInt::BOOL);
3673 record_for_igvn(exclusion);
3674
3675 // Merge the excluded control and memory.
3676 excluded_rgn->init_req(_true_path, excluded);
3677 excluded_rgn->init_req(_false_path, included);
3678 excluded_mem->init_req(_true_path, tid_memory);
3679 excluded_mem->init_req(_false_path, included_memory);
3680 exclusion->init_req(_true_path, _gvn.transform(vthread_is_excluded));
3681 exclusion->init_req(_false_path, _gvn.transform(vthread_is_included));
3682
3683 // Set intermediate state.
3684 set_control(_gvn.transform(excluded_rgn));
3685 set_all_memory(excluded_mem);
3686
3687 // Store the vthread exclusion state to the jfr thread local.
3688 Node* thread_local_excluded_offset = basic_plus_adr(jt, in_bytes(THREAD_LOCAL_OFFSET_JFR + VTHREAD_EXCLUDED_OFFSET_JFR));
3689 store_to_memory(control(), thread_local_excluded_offset, _gvn.transform(exclusion), T_BOOLEAN, MemNode::unordered, true);
3690
3691 // Store release
3692 Node * vthread_true_memory = store_to_memory(control(), vthread_offset, _gvn.intcon(1), T_BOOLEAN, MemNode::release, true);
3693
3694 RegionNode* thread_compare_rgn = new RegionNode(PATH_LIMIT);
3695 record_for_igvn(thread_compare_rgn);
3696 PhiNode* thread_compare_mem = new PhiNode(thread_compare_rgn, Type::MEMORY, TypePtr::BOTTOM);
3697 record_for_igvn(thread_compare_mem);
3698 PhiNode* vthread = new PhiNode(thread_compare_rgn, TypeInt::BOOL);
3699 record_for_igvn(vthread);
3700
3701 // Merge the thread_compare control and memory.
3702 thread_compare_rgn->init_req(_true_path, control());
3703 thread_compare_rgn->init_req(_false_path, thread_equal_carrierThread);
3704 thread_compare_mem->init_req(_true_path, vthread_true_memory);
3705 thread_compare_mem->init_req(_false_path, vthread_false_memory);
3706
3707 // Set output state.
3708 set_control(_gvn.transform(thread_compare_rgn));
3709 set_all_memory(_gvn.transform(thread_compare_mem));
3710 }
3711
3712 #endif // JFR_HAVE_INTRINSICS
3713
3714 //------------------------inline_native_currentCarrierThread------------------
3715 bool LibraryCallKit::inline_native_currentCarrierThread() {
3716 Node* junk = nullptr;
3717 set_result(generate_current_thread(junk));
3718 return true;
3719 }
3720
3721 //------------------------inline_native_currentThread------------------
3722 bool LibraryCallKit::inline_native_currentThread() {
3723 Node* junk = nullptr;
3724 set_result(generate_virtual_thread(junk));
3725 return true;
3726 }
3727
3728 //------------------------inline_native_setVthread------------------
3729 bool LibraryCallKit::inline_native_setCurrentThread() {
3730 assert(C->method()->changes_current_thread(),
3731 "method changes current Thread but is not annotated ChangesCurrentThread");
3732 Node* arr = argument(1);
3733 Node* thread = _gvn.transform(new ThreadLocalNode());
3734 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::vthread_offset()));
3735 Node* thread_obj_handle
3736 = make_load(nullptr, p, p->bottom_type()->is_ptr(), T_OBJECT, MemNode::unordered);
3737 thread_obj_handle = _gvn.transform(thread_obj_handle);
3738 const TypePtr *adr_type = _gvn.type(thread_obj_handle)->isa_ptr();
3739 access_store_at(nullptr, thread_obj_handle, adr_type, arr, _gvn.type(arr), T_OBJECT, IN_NATIVE | MO_UNORDERED);
3740
3741 // Change the _monitor_owner_id of the JavaThread
3742 Node* tid = load_field_from_object(arr, "tid", "J");
3743 Node* monitor_owner_id_offset = basic_plus_adr(thread, in_bytes(JavaThread::monitor_owner_id_offset()));
3744 store_to_memory(control(), monitor_owner_id_offset, tid, T_LONG, MemNode::unordered, true);
3745
3746 JFR_ONLY(extend_setCurrentThread(thread, arr);)
3747 return true;
3748 }
3749
3750 const Type* LibraryCallKit::scopedValueCache_type() {
3751 ciKlass* objects_klass = ciObjArrayKlass::make(env()->Object_klass());
3752 const TypeOopPtr* etype = TypeOopPtr::make_from_klass(env()->Object_klass());
3753 const TypeAry* arr0 = TypeAry::make(etype, TypeInt::POS);
3754
3755 // Because we create the scopedValue cache lazily we have to make the
3756 // type of the result BotPTR.
3757 bool xk = etype->klass_is_exact();
3758 const Type* objects_type = TypeAryPtr::make(TypePtr::BotPTR, arr0, objects_klass, xk, 0);
3759 return objects_type;
3760 }
3761
3762 Node* LibraryCallKit::scopedValueCache_helper() {
3763 Node* thread = _gvn.transform(new ThreadLocalNode());
3764 Node* p = basic_plus_adr(top()/*!oop*/, thread, in_bytes(JavaThread::scopedValueCache_offset()));
3765 // We cannot use immutable_memory() because we might flip onto a
3766 // different carrier thread, at which point we'll need to use that
3767 // carrier thread's cache.
3768 // return _gvn.transform(LoadNode::make(_gvn, nullptr, immutable_memory(), p, p->bottom_type()->is_ptr(),
3769 // TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered));
3770 return make_load(nullptr, p, p->bottom_type()->is_ptr(), T_ADDRESS, MemNode::unordered);
3771 }
3772
3773 //------------------------inline_native_scopedValueCache------------------
3774 bool LibraryCallKit::inline_native_scopedValueCache() {
3775 Node* cache_obj_handle = scopedValueCache_helper();
3776 const Type* objects_type = scopedValueCache_type();
3777 set_result(access_load(cache_obj_handle, objects_type, T_OBJECT, IN_NATIVE));
3778
3779 return true;
3780 }
3781
3782 //------------------------inline_native_setScopedValueCache------------------
3783 bool LibraryCallKit::inline_native_setScopedValueCache() {
3784 Node* arr = argument(0);
3785 Node* cache_obj_handle = scopedValueCache_helper();
3786 const Type* objects_type = scopedValueCache_type();
3787
3788 const TypePtr *adr_type = _gvn.type(cache_obj_handle)->isa_ptr();
3789 access_store_at(nullptr, cache_obj_handle, adr_type, arr, objects_type, T_OBJECT, IN_NATIVE | MO_UNORDERED);
3790
3791 return true;
3792 }
3793
3794 //------------------------inline_native_Continuation_pin and unpin-----------
3795
3796 // Shared implementation routine for both pin and unpin.
3797 bool LibraryCallKit::inline_native_Continuation_pinning(bool unpin) {
3798 enum { _true_path = 1, _false_path = 2, PATH_LIMIT };
3799
3800 // Save input memory.
3801 Node* input_memory_state = reset_memory();
3802 set_all_memory(input_memory_state);
3803
3804 // TLS
3805 Node* tls_ptr = _gvn.transform(new ThreadLocalNode());
3806 Node* last_continuation_offset = basic_plus_adr(top(), tls_ptr, in_bytes(JavaThread::cont_entry_offset()));
3807 Node* last_continuation = make_load(control(), last_continuation_offset, last_continuation_offset->get_ptr_type(), T_ADDRESS, MemNode::unordered);
3808
3809 // Null check the last continuation object.
3810 Node* continuation_cmp_null = _gvn.transform(new CmpPNode(last_continuation, null()));
3811 Node* test_continuation_not_equal_null = _gvn.transform(new BoolNode(continuation_cmp_null, BoolTest::ne));
3812 IfNode* iff_continuation_not_equal_null = create_and_map_if(control(), test_continuation_not_equal_null, PROB_MAX, COUNT_UNKNOWN);
3813
3814 // False path, last continuation is null.
3815 Node* continuation_is_null = _gvn.transform(new IfFalseNode(iff_continuation_not_equal_null));
3816
3817 // True path, last continuation is not null.
3818 Node* continuation_is_not_null = _gvn.transform(new IfTrueNode(iff_continuation_not_equal_null));
3819
3820 set_control(continuation_is_not_null);
3821
3822 // Load the pin count from the last continuation.
3823 Node* pin_count_offset = basic_plus_adr(top(), last_continuation, in_bytes(ContinuationEntry::pin_count_offset()));
3824 Node* pin_count = make_load(control(), pin_count_offset, TypeInt::INT, T_INT, MemNode::unordered);
3825
3826 // The loaded pin count is compared against a context specific rhs for over/underflow detection.
3827 Node* pin_count_rhs;
3828 if (unpin) {
3829 pin_count_rhs = _gvn.intcon(0);
3830 } else {
3831 pin_count_rhs = _gvn.intcon(UINT32_MAX);
3832 }
3833 Node* pin_count_cmp = _gvn.transform(new CmpUNode(_gvn.transform(pin_count), pin_count_rhs));
3834 Node* test_pin_count_over_underflow = _gvn.transform(new BoolNode(pin_count_cmp, BoolTest::eq));
3835 IfNode* iff_pin_count_over_underflow = create_and_map_if(control(), test_pin_count_over_underflow, PROB_MIN, COUNT_UNKNOWN);
3836
3837 // True branch, pin count over/underflow.
3838 Node* pin_count_over_underflow = _gvn.transform(new IfTrueNode(iff_pin_count_over_underflow));
3839 {
3840 // Trap (but not deoptimize (Action_none)) and continue in the interpreter
3841 // which will throw IllegalStateException for pin count over/underflow.
3842 // No memory changed so far - we can use memory create by reset_memory()
3843 // at the beginning of this intrinsic. No need to call reset_memory() again.
3844 PreserveJVMState pjvms(this);
3845 set_control(pin_count_over_underflow);
3846 uncommon_trap(Deoptimization::Reason_intrinsic,
3847 Deoptimization::Action_none);
3848 assert(stopped(), "invariant");
3849 }
3850
3851 // False branch, no pin count over/underflow. Increment or decrement pin count and store back.
3852 Node* valid_pin_count = _gvn.transform(new IfFalseNode(iff_pin_count_over_underflow));
3853 set_control(valid_pin_count);
3854
3855 Node* next_pin_count;
3856 if (unpin) {
3857 next_pin_count = _gvn.transform(new SubINode(pin_count, _gvn.intcon(1)));
3858 } else {
3859 next_pin_count = _gvn.transform(new AddINode(pin_count, _gvn.intcon(1)));
3860 }
3861
3862 store_to_memory(control(), pin_count_offset, next_pin_count, T_INT, MemNode::unordered);
3863
3864 // Result of top level CFG and Memory.
3865 RegionNode* result_rgn = new RegionNode(PATH_LIMIT);
3866 record_for_igvn(result_rgn);
3867 PhiNode* result_mem = new PhiNode(result_rgn, Type::MEMORY, TypePtr::BOTTOM);
3868 record_for_igvn(result_mem);
3869
3870 result_rgn->init_req(_true_path, _gvn.transform(valid_pin_count));
3871 result_rgn->init_req(_false_path, _gvn.transform(continuation_is_null));
3872 result_mem->init_req(_true_path, _gvn.transform(reset_memory()));
3873 result_mem->init_req(_false_path, _gvn.transform(input_memory_state));
3874
3875 // Set output state.
3876 set_control(_gvn.transform(result_rgn));
3877 set_all_memory(_gvn.transform(result_mem));
3878
3879 return true;
3880 }
3881
3882 //---------------------------load_mirror_from_klass----------------------------
3883 // Given a klass oop, load its java mirror (a java.lang.Class oop).
3884 Node* LibraryCallKit::load_mirror_from_klass(Node* klass) {
3885 Node* p = basic_plus_adr(klass, in_bytes(Klass::java_mirror_offset()));
3886 Node* load = make_load(nullptr, p, TypeRawPtr::NOTNULL, T_ADDRESS, MemNode::unordered);
3887 // mirror = ((OopHandle)mirror)->resolve();
3888 return access_load(load, TypeInstPtr::MIRROR, T_OBJECT, IN_NATIVE);
3889 }
3890
3891 //-----------------------load_klass_from_mirror_common-------------------------
3892 // Given a java mirror (a java.lang.Class oop), load its corresponding klass oop.
3893 // Test the klass oop for null (signifying a primitive Class like Integer.TYPE),
3894 // and branch to the given path on the region.
3895 // If never_see_null, take an uncommon trap on null, so we can optimistically
3896 // compile for the non-null case.
3897 // If the region is null, force never_see_null = true.
3898 Node* LibraryCallKit::load_klass_from_mirror_common(Node* mirror,
3899 bool never_see_null,
3900 RegionNode* region,
3901 int null_path,
3902 int offset) {
3903 if (region == nullptr) never_see_null = true;
3904 Node* p = basic_plus_adr(mirror, offset);
3905 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
3906 Node* kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, kls_type));
3907 Node* null_ctl = top();
3908 kls = null_check_oop(kls, &null_ctl, never_see_null);
3909 if (region != nullptr) {
3910 // Set region->in(null_path) if the mirror is a primitive (e.g, int.class).
3911 region->init_req(null_path, null_ctl);
3912 } else {
3913 assert(null_ctl == top(), "no loose ends");
3914 }
3915 return kls;
3916 }
3917
3918 //--------------------(inline_native_Class_query helpers)---------------------
3919 // Use this for JVM_ACC_INTERFACE.
3920 // Fall through if (mods & mask) == bits, take the guard otherwise.
3921 Node* LibraryCallKit::generate_klass_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region,
3922 ByteSize offset, const Type* type, BasicType bt) {
3923 // Branch around if the given klass has the given modifier bit set.
3924 // Like generate_guard, adds a new path onto the region.
3925 Node* modp = basic_plus_adr(kls, in_bytes(offset));
3926 Node* mods = make_load(nullptr, modp, type, bt, MemNode::unordered);
3927 Node* mask = intcon(modifier_mask);
3928 Node* bits = intcon(modifier_bits);
3929 Node* mbit = _gvn.transform(new AndINode(mods, mask));
3930 Node* cmp = _gvn.transform(new CmpINode(mbit, bits));
3931 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::ne));
3932 return generate_fair_guard(bol, region);
3933 }
3934 Node* LibraryCallKit::generate_interface_guard(Node* kls, RegionNode* region) {
3935 return generate_klass_flags_guard(kls, JVM_ACC_INTERFACE, 0, region,
3936 Klass::access_flags_offset(), TypeInt::CHAR, T_CHAR);
3937 }
3938
3939 // Use this for testing if Klass is_hidden, has_finalizer, and is_cloneable_fast.
3940 Node* LibraryCallKit::generate_misc_flags_guard(Node* kls, int modifier_mask, int modifier_bits, RegionNode* region) {
3941 return generate_klass_flags_guard(kls, modifier_mask, modifier_bits, region,
3942 Klass::misc_flags_offset(), TypeInt::UBYTE, T_BOOLEAN);
3943 }
3944
3945 Node* LibraryCallKit::generate_hidden_class_guard(Node* kls, RegionNode* region) {
3946 return generate_misc_flags_guard(kls, KlassFlags::_misc_is_hidden_class, 0, region);
3947 }
3948
3949 //-------------------------inline_native_Class_query-------------------
3950 bool LibraryCallKit::inline_native_Class_query(vmIntrinsics::ID id) {
3951 const Type* return_type = TypeInt::BOOL;
3952 Node* prim_return_value = top(); // what happens if it's a primitive class?
3953 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
3954 bool expect_prim = false; // most of these guys expect to work on refs
3955
3956 enum { _normal_path = 1, _prim_path = 2, PATH_LIMIT };
3957
3958 Node* mirror = argument(0);
3959 Node* obj = top();
3960
3961 switch (id) {
3962 case vmIntrinsics::_isInstance:
3963 // nothing is an instance of a primitive type
3964 prim_return_value = intcon(0);
3965 obj = argument(1);
3966 break;
3967 case vmIntrinsics::_isHidden:
3968 prim_return_value = intcon(0);
3969 break;
3970 case vmIntrinsics::_getSuperclass:
3971 prim_return_value = null();
3972 return_type = TypeInstPtr::MIRROR->cast_to_ptr_type(TypePtr::BotPTR);
3973 break;
3974 case vmIntrinsics::_getClassAccessFlags:
3975 prim_return_value = intcon(JVM_ACC_ABSTRACT | JVM_ACC_FINAL | JVM_ACC_PUBLIC);
3976 return_type = TypeInt::CHAR;
3977 break;
3978 default:
3979 fatal_unexpected_iid(id);
3980 break;
3981 }
3982
3983 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
3984 if (mirror_con == nullptr) return false; // cannot happen?
3985
3986 #ifndef PRODUCT
3987 if (C->print_intrinsics() || C->print_inlining()) {
3988 ciType* k = mirror_con->java_mirror_type();
3989 if (k) {
3990 tty->print("Inlining %s on constant Class ", vmIntrinsics::name_at(intrinsic_id()));
3991 k->print_name();
3992 tty->cr();
3993 }
3994 }
3995 #endif
3996
3997 // Null-check the mirror, and the mirror's klass ptr (in case it is a primitive).
3998 RegionNode* region = new RegionNode(PATH_LIMIT);
3999 record_for_igvn(region);
4000 PhiNode* phi = new PhiNode(region, return_type);
4001
4002 // The mirror will never be null of Reflection.getClassAccessFlags, however
4003 // it may be null for Class.isInstance or Class.getModifiers. Throw a NPE
4004 // if it is. See bug 4774291.
4005
4006 // For Reflection.getClassAccessFlags(), the null check occurs in
4007 // the wrong place; see inline_unsafe_access(), above, for a similar
4008 // situation.
4009 mirror = null_check(mirror);
4010 // If mirror or obj is dead, only null-path is taken.
4011 if (stopped()) return true;
4012
4013 if (expect_prim) never_see_null = false; // expect nulls (meaning prims)
4014
4015 // Now load the mirror's klass metaobject, and null-check it.
4016 // Side-effects region with the control path if the klass is null.
4017 Node* kls = load_klass_from_mirror(mirror, never_see_null, region, _prim_path);
4018 // If kls is null, we have a primitive mirror.
4019 phi->init_req(_prim_path, prim_return_value);
4020 if (stopped()) { set_result(region, phi); return true; }
4021 bool safe_for_replace = (region->in(_prim_path) == top());
4022
4023 Node* p; // handy temp
4024 Node* null_ctl;
4025
4026 // Now that we have the non-null klass, we can perform the real query.
4027 // For constant classes, the query will constant-fold in LoadNode::Value.
4028 Node* query_value = top();
4029 switch (id) {
4030 case vmIntrinsics::_isInstance:
4031 // nothing is an instance of a primitive type
4032 query_value = gen_instanceof(obj, kls, safe_for_replace);
4033 break;
4034
4035 case vmIntrinsics::_isHidden:
4036 // (To verify this code sequence, check the asserts in JVM_IsHiddenClass.)
4037 if (generate_hidden_class_guard(kls, region) != nullptr)
4038 // A guard was added. If the guard is taken, it was an hidden class.
4039 phi->add_req(intcon(1));
4040 // If we fall through, it's a plain class.
4041 query_value = intcon(0);
4042 break;
4043
4044
4045 case vmIntrinsics::_getSuperclass:
4046 // The rules here are somewhat unfortunate, but we can still do better
4047 // with random logic than with a JNI call.
4048 // Interfaces store null or Object as _super, but must report null.
4049 // Arrays store an intermediate super as _super, but must report Object.
4050 // Other types can report the actual _super.
4051 // (To verify this code sequence, check the asserts in JVM_IsInterface.)
4052 if (generate_interface_guard(kls, region) != nullptr)
4053 // A guard was added. If the guard is taken, it was an interface.
4054 phi->add_req(null());
4055 if (generate_array_guard(kls, region) != nullptr)
4056 // A guard was added. If the guard is taken, it was an array.
4057 phi->add_req(makecon(TypeInstPtr::make(env()->Object_klass()->java_mirror())));
4058 // If we fall through, it's a plain class. Get its _super.
4059 p = basic_plus_adr(kls, in_bytes(Klass::super_offset()));
4060 kls = _gvn.transform(LoadKlassNode::make(_gvn, immutable_memory(), p, TypeRawPtr::BOTTOM, TypeInstKlassPtr::OBJECT_OR_NULL));
4061 null_ctl = top();
4062 kls = null_check_oop(kls, &null_ctl);
4063 if (null_ctl != top()) {
4064 // If the guard is taken, Object.superClass is null (both klass and mirror).
4065 region->add_req(null_ctl);
4066 phi ->add_req(null());
4067 }
4068 if (!stopped()) {
4069 query_value = load_mirror_from_klass(kls);
4070 }
4071 break;
4072
4073 case vmIntrinsics::_getClassAccessFlags:
4074 p = basic_plus_adr(kls, in_bytes(Klass::access_flags_offset()));
4075 query_value = make_load(nullptr, p, TypeInt::CHAR, T_CHAR, MemNode::unordered);
4076 break;
4077
4078 default:
4079 fatal_unexpected_iid(id);
4080 break;
4081 }
4082
4083 // Fall-through is the normal case of a query to a real class.
4084 phi->init_req(1, query_value);
4085 region->init_req(1, control());
4086
4087 C->set_has_split_ifs(true); // Has chance for split-if optimization
4088 set_result(region, phi);
4089 return true;
4090 }
4091
4092 //-------------------------inline_Class_cast-------------------
4093 bool LibraryCallKit::inline_Class_cast() {
4094 Node* mirror = argument(0); // Class
4095 Node* obj = argument(1);
4096 const TypeInstPtr* mirror_con = _gvn.type(mirror)->isa_instptr();
4097 if (mirror_con == nullptr) {
4098 return false; // dead path (mirror->is_top()).
4099 }
4100 if (obj == nullptr || obj->is_top()) {
4101 return false; // dead path
4102 }
4103 const TypeOopPtr* tp = _gvn.type(obj)->isa_oopptr();
4104
4105 // First, see if Class.cast() can be folded statically.
4106 // java_mirror_type() returns non-null for compile-time Class constants.
4107 ciType* tm = mirror_con->java_mirror_type();
4108 if (tm != nullptr && tm->is_klass() &&
4109 tp != nullptr) {
4110 if (!tp->is_loaded()) {
4111 // Don't use intrinsic when class is not loaded.
4112 return false;
4113 } else {
4114 int static_res = C->static_subtype_check(TypeKlassPtr::make(tm->as_klass(), Type::trust_interfaces), tp->as_klass_type());
4115 if (static_res == Compile::SSC_always_true) {
4116 // isInstance() is true - fold the code.
4117 set_result(obj);
4118 return true;
4119 } else if (static_res == Compile::SSC_always_false) {
4120 // Don't use intrinsic, have to throw ClassCastException.
4121 // If the reference is null, the non-intrinsic bytecode will
4122 // be optimized appropriately.
4123 return false;
4124 }
4125 }
4126 }
4127
4128 // Bailout intrinsic and do normal inlining if exception path is frequent.
4129 if (too_many_traps(Deoptimization::Reason_intrinsic)) {
4130 return false;
4131 }
4132
4133 // Generate dynamic checks.
4134 // Class.cast() is java implementation of _checkcast bytecode.
4135 // Do checkcast (Parse::do_checkcast()) optimizations here.
4136
4137 mirror = null_check(mirror);
4138 // If mirror is dead, only null-path is taken.
4139 if (stopped()) {
4140 return true;
4141 }
4142
4143 // Not-subtype or the mirror's klass ptr is null (in case it is a primitive).
4144 enum { _bad_type_path = 1, _prim_path = 2, PATH_LIMIT };
4145 RegionNode* region = new RegionNode(PATH_LIMIT);
4146 record_for_igvn(region);
4147
4148 // Now load the mirror's klass metaobject, and null-check it.
4149 // If kls is null, we have a primitive mirror and
4150 // nothing is an instance of a primitive type.
4151 Node* kls = load_klass_from_mirror(mirror, false, region, _prim_path);
4152
4153 Node* res = top();
4154 if (!stopped()) {
4155 Node* bad_type_ctrl = top();
4156 // Do checkcast optimizations.
4157 res = gen_checkcast(obj, kls, &bad_type_ctrl);
4158 region->init_req(_bad_type_path, bad_type_ctrl);
4159 }
4160 if (region->in(_prim_path) != top() ||
4161 region->in(_bad_type_path) != top()) {
4162 // Let Interpreter throw ClassCastException.
4163 PreserveJVMState pjvms(this);
4164 set_control(_gvn.transform(region));
4165 uncommon_trap(Deoptimization::Reason_intrinsic,
4166 Deoptimization::Action_maybe_recompile);
4167 }
4168 if (!stopped()) {
4169 set_result(res);
4170 }
4171 return true;
4172 }
4173
4174
4175 //--------------------------inline_native_subtype_check------------------------
4176 // This intrinsic takes the JNI calls out of the heart of
4177 // UnsafeFieldAccessorImpl.set, which improves Field.set, readObject, etc.
4178 bool LibraryCallKit::inline_native_subtype_check() {
4179 // Pull both arguments off the stack.
4180 Node* args[2]; // two java.lang.Class mirrors: superc, subc
4181 args[0] = argument(0);
4182 args[1] = argument(1);
4183 Node* klasses[2]; // corresponding Klasses: superk, subk
4184 klasses[0] = klasses[1] = top();
4185
4186 enum {
4187 // A full decision tree on {superc is prim, subc is prim}:
4188 _prim_0_path = 1, // {P,N} => false
4189 // {P,P} & superc!=subc => false
4190 _prim_same_path, // {P,P} & superc==subc => true
4191 _prim_1_path, // {N,P} => false
4192 _ref_subtype_path, // {N,N} & subtype check wins => true
4193 _both_ref_path, // {N,N} & subtype check loses => false
4194 PATH_LIMIT
4195 };
4196
4197 RegionNode* region = new RegionNode(PATH_LIMIT);
4198 Node* phi = new PhiNode(region, TypeInt::BOOL);
4199 record_for_igvn(region);
4200
4201 const TypePtr* adr_type = TypeRawPtr::BOTTOM; // memory type of loads
4202 const TypeKlassPtr* kls_type = TypeInstKlassPtr::OBJECT_OR_NULL;
4203 int class_klass_offset = java_lang_Class::klass_offset();
4204
4205 // First null-check both mirrors and load each mirror's klass metaobject.
4206 int which_arg;
4207 for (which_arg = 0; which_arg <= 1; which_arg++) {
4208 Node* arg = args[which_arg];
4209 arg = null_check(arg);
4210 if (stopped()) break;
4211 args[which_arg] = arg;
4212
4213 Node* p = basic_plus_adr(arg, class_klass_offset);
4214 Node* kls = LoadKlassNode::make(_gvn, immutable_memory(), p, adr_type, kls_type);
4215 klasses[which_arg] = _gvn.transform(kls);
4216 }
4217
4218 // Having loaded both klasses, test each for null.
4219 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4220 for (which_arg = 0; which_arg <= 1; which_arg++) {
4221 Node* kls = klasses[which_arg];
4222 Node* null_ctl = top();
4223 kls = null_check_oop(kls, &null_ctl, never_see_null);
4224 int prim_path = (which_arg == 0 ? _prim_0_path : _prim_1_path);
4225 region->init_req(prim_path, null_ctl);
4226 if (stopped()) break;
4227 klasses[which_arg] = kls;
4228 }
4229
4230 if (!stopped()) {
4231 // now we have two reference types, in klasses[0..1]
4232 Node* subk = klasses[1]; // the argument to isAssignableFrom
4233 Node* superk = klasses[0]; // the receiver
4234 region->set_req(_both_ref_path, gen_subtype_check(subk, superk));
4235 // now we have a successful reference subtype check
4236 region->set_req(_ref_subtype_path, control());
4237 }
4238
4239 // If both operands are primitive (both klasses null), then
4240 // we must return true when they are identical primitives.
4241 // It is convenient to test this after the first null klass check.
4242 set_control(region->in(_prim_0_path)); // go back to first null check
4243 if (!stopped()) {
4244 // Since superc is primitive, make a guard for the superc==subc case.
4245 Node* cmp_eq = _gvn.transform(new CmpPNode(args[0], args[1]));
4246 Node* bol_eq = _gvn.transform(new BoolNode(cmp_eq, BoolTest::eq));
4247 generate_guard(bol_eq, region, PROB_FAIR);
4248 if (region->req() == PATH_LIMIT+1) {
4249 // A guard was added. If the added guard is taken, superc==subc.
4250 region->swap_edges(PATH_LIMIT, _prim_same_path);
4251 region->del_req(PATH_LIMIT);
4252 }
4253 region->set_req(_prim_0_path, control()); // Not equal after all.
4254 }
4255
4256 // these are the only paths that produce 'true':
4257 phi->set_req(_prim_same_path, intcon(1));
4258 phi->set_req(_ref_subtype_path, intcon(1));
4259
4260 // pull together the cases:
4261 assert(region->req() == PATH_LIMIT, "sane region");
4262 for (uint i = 1; i < region->req(); i++) {
4263 Node* ctl = region->in(i);
4264 if (ctl == nullptr || ctl == top()) {
4265 region->set_req(i, top());
4266 phi ->set_req(i, top());
4267 } else if (phi->in(i) == nullptr) {
4268 phi->set_req(i, intcon(0)); // all other paths produce 'false'
4269 }
4270 }
4271
4272 set_control(_gvn.transform(region));
4273 set_result(_gvn.transform(phi));
4274 return true;
4275 }
4276
4277 //---------------------generate_array_guard_common------------------------
4278 Node* LibraryCallKit::generate_array_guard_common(Node* kls, RegionNode* region,
4279 bool obj_array, bool not_array, Node** obj) {
4280
4281 if (stopped()) {
4282 return nullptr;
4283 }
4284
4285 // If obj_array/non_array==false/false:
4286 // Branch around if the given klass is in fact an array (either obj or prim).
4287 // If obj_array/non_array==false/true:
4288 // Branch around if the given klass is not an array klass of any kind.
4289 // If obj_array/non_array==true/true:
4290 // Branch around if the kls is not an oop array (kls is int[], String, etc.)
4291 // If obj_array/non_array==true/false:
4292 // Branch around if the kls is an oop array (Object[] or subtype)
4293 //
4294 // Like generate_guard, adds a new path onto the region.
4295 jint layout_con = 0;
4296 Node* layout_val = get_layout_helper(kls, layout_con);
4297 if (layout_val == nullptr) {
4298 bool query = (obj_array
4299 ? Klass::layout_helper_is_objArray(layout_con)
4300 : Klass::layout_helper_is_array(layout_con));
4301 if (query == not_array) {
4302 return nullptr; // never a branch
4303 } else { // always a branch
4304 Node* always_branch = control();
4305 if (region != nullptr)
4306 region->add_req(always_branch);
4307 set_control(top());
4308 return always_branch;
4309 }
4310 }
4311 // Now test the correct condition.
4312 jint nval = (obj_array
4313 ? (jint)(Klass::_lh_array_tag_type_value
4314 << Klass::_lh_array_tag_shift)
4315 : Klass::_lh_neutral_value);
4316 Node* cmp = _gvn.transform(new CmpINode(layout_val, intcon(nval)));
4317 BoolTest::mask btest = BoolTest::lt; // correct for testing is_[obj]array
4318 // invert the test if we are looking for a non-array
4319 if (not_array) btest = BoolTest(btest).negate();
4320 Node* bol = _gvn.transform(new BoolNode(cmp, btest));
4321 Node* ctrl = generate_fair_guard(bol, region);
4322 Node* is_array_ctrl = not_array ? control() : ctrl;
4323 if (obj != nullptr && is_array_ctrl != nullptr && is_array_ctrl != top()) {
4324 // Keep track of the fact that 'obj' is an array to prevent
4325 // array specific accesses from floating above the guard.
4326 *obj = _gvn.transform(new CastPPNode(is_array_ctrl, *obj, TypeAryPtr::BOTTOM));
4327 }
4328 return ctrl;
4329 }
4330
4331
4332 //-----------------------inline_native_newArray--------------------------
4333 // private static native Object java.lang.reflect.newArray(Class<?> componentType, int length);
4334 // private native Object Unsafe.allocateUninitializedArray0(Class<?> cls, int size);
4335 bool LibraryCallKit::inline_unsafe_newArray(bool uninitialized) {
4336 Node* mirror;
4337 Node* count_val;
4338 if (uninitialized) {
4339 null_check_receiver();
4340 mirror = argument(1);
4341 count_val = argument(2);
4342 } else {
4343 mirror = argument(0);
4344 count_val = argument(1);
4345 }
4346
4347 mirror = null_check(mirror);
4348 // If mirror or obj is dead, only null-path is taken.
4349 if (stopped()) return true;
4350
4351 enum { _normal_path = 1, _slow_path = 2, PATH_LIMIT };
4352 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4353 PhiNode* result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
4354 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4355 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4356
4357 bool never_see_null = !too_many_traps(Deoptimization::Reason_null_check);
4358 Node* klass_node = load_array_klass_from_mirror(mirror, never_see_null,
4359 result_reg, _slow_path);
4360 Node* normal_ctl = control();
4361 Node* no_array_ctl = result_reg->in(_slow_path);
4362
4363 // Generate code for the slow case. We make a call to newArray().
4364 set_control(no_array_ctl);
4365 if (!stopped()) {
4366 // Either the input type is void.class, or else the
4367 // array klass has not yet been cached. Either the
4368 // ensuing call will throw an exception, or else it
4369 // will cache the array klass for next time.
4370 PreserveJVMState pjvms(this);
4371 CallJavaNode* slow_call = nullptr;
4372 if (uninitialized) {
4373 // Generate optimized virtual call (holder class 'Unsafe' is final)
4374 slow_call = generate_method_call(vmIntrinsics::_allocateUninitializedArray, false, false, true);
4375 } else {
4376 slow_call = generate_method_call_static(vmIntrinsics::_newArray, true);
4377 }
4378 Node* slow_result = set_results_for_java_call(slow_call);
4379 // this->control() comes from set_results_for_java_call
4380 result_reg->set_req(_slow_path, control());
4381 result_val->set_req(_slow_path, slow_result);
4382 result_io ->set_req(_slow_path, i_o());
4383 result_mem->set_req(_slow_path, reset_memory());
4384 }
4385
4386 set_control(normal_ctl);
4387 if (!stopped()) {
4388 // Normal case: The array type has been cached in the java.lang.Class.
4389 // The following call works fine even if the array type is polymorphic.
4390 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4391 Node* obj = new_array(klass_node, count_val, 0); // no arguments to push
4392 result_reg->init_req(_normal_path, control());
4393 result_val->init_req(_normal_path, obj);
4394 result_io ->init_req(_normal_path, i_o());
4395 result_mem->init_req(_normal_path, reset_memory());
4396
4397 if (uninitialized) {
4398 // Mark the allocation so that zeroing is skipped
4399 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(obj);
4400 alloc->maybe_set_complete(&_gvn);
4401 }
4402 }
4403
4404 // Return the combined state.
4405 set_i_o( _gvn.transform(result_io) );
4406 set_all_memory( _gvn.transform(result_mem));
4407
4408 C->set_has_split_ifs(true); // Has chance for split-if optimization
4409 set_result(result_reg, result_val);
4410 return true;
4411 }
4412
4413 //----------------------inline_native_getLength--------------------------
4414 // public static native int java.lang.reflect.Array.getLength(Object array);
4415 bool LibraryCallKit::inline_native_getLength() {
4416 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4417
4418 Node* array = null_check(argument(0));
4419 // If array is dead, only null-path is taken.
4420 if (stopped()) return true;
4421
4422 // Deoptimize if it is a non-array.
4423 Node* non_array = generate_non_array_guard(load_object_klass(array), nullptr, &array);
4424
4425 if (non_array != nullptr) {
4426 PreserveJVMState pjvms(this);
4427 set_control(non_array);
4428 uncommon_trap(Deoptimization::Reason_intrinsic,
4429 Deoptimization::Action_maybe_recompile);
4430 }
4431
4432 // If control is dead, only non-array-path is taken.
4433 if (stopped()) return true;
4434
4435 // The works fine even if the array type is polymorphic.
4436 // It could be a dynamic mix of int[], boolean[], Object[], etc.
4437 Node* result = load_array_length(array);
4438
4439 C->set_has_split_ifs(true); // Has chance for split-if optimization
4440 set_result(result);
4441 return true;
4442 }
4443
4444 //------------------------inline_array_copyOf----------------------------
4445 // public static <T,U> T[] java.util.Arrays.copyOf( U[] original, int newLength, Class<? extends T[]> newType);
4446 // public static <T,U> T[] java.util.Arrays.copyOfRange(U[] original, int from, int to, Class<? extends T[]> newType);
4447 bool LibraryCallKit::inline_array_copyOf(bool is_copyOfRange) {
4448 if (too_many_traps(Deoptimization::Reason_intrinsic)) return false;
4449
4450 // Get the arguments.
4451 Node* original = argument(0);
4452 Node* start = is_copyOfRange? argument(1): intcon(0);
4453 Node* end = is_copyOfRange? argument(2): argument(1);
4454 Node* array_type_mirror = is_copyOfRange? argument(3): argument(2);
4455
4456 Node* newcopy = nullptr;
4457
4458 // Set the original stack and the reexecute bit for the interpreter to reexecute
4459 // the bytecode that invokes Arrays.copyOf if deoptimization happens.
4460 { PreserveReexecuteState preexecs(this);
4461 jvms()->set_should_reexecute(true);
4462
4463 array_type_mirror = null_check(array_type_mirror);
4464 original = null_check(original);
4465
4466 // Check if a null path was taken unconditionally.
4467 if (stopped()) return true;
4468
4469 Node* orig_length = load_array_length(original);
4470
4471 Node* klass_node = load_klass_from_mirror(array_type_mirror, false, nullptr, 0);
4472 klass_node = null_check(klass_node);
4473
4474 RegionNode* bailout = new RegionNode(1);
4475 record_for_igvn(bailout);
4476
4477 // Despite the generic type of Arrays.copyOf, the mirror might be int, int[], etc.
4478 // Bail out if that is so.
4479 Node* not_objArray = generate_non_objArray_guard(klass_node, bailout);
4480 if (not_objArray != nullptr) {
4481 // Improve the klass node's type from the new optimistic assumption:
4482 ciKlass* ak = ciArrayKlass::make(env()->Object_klass());
4483 const Type* akls = TypeKlassPtr::make(TypePtr::NotNull, ak, 0/*offset*/);
4484 Node* cast = new CastPPNode(control(), klass_node, akls);
4485 klass_node = _gvn.transform(cast);
4486 }
4487
4488 // Bail out if either start or end is negative.
4489 generate_negative_guard(start, bailout, &start);
4490 generate_negative_guard(end, bailout, &end);
4491
4492 Node* length = end;
4493 if (_gvn.type(start) != TypeInt::ZERO) {
4494 length = _gvn.transform(new SubINode(end, start));
4495 }
4496
4497 // Bail out if length is negative (i.e., if start > end).
4498 // Without this the new_array would throw
4499 // NegativeArraySizeException but IllegalArgumentException is what
4500 // should be thrown
4501 generate_negative_guard(length, bailout, &length);
4502
4503 // Bail out if start is larger than the original length
4504 Node* orig_tail = _gvn.transform(new SubINode(orig_length, start));
4505 generate_negative_guard(orig_tail, bailout, &orig_tail);
4506
4507 if (bailout->req() > 1) {
4508 PreserveJVMState pjvms(this);
4509 set_control(_gvn.transform(bailout));
4510 uncommon_trap(Deoptimization::Reason_intrinsic,
4511 Deoptimization::Action_maybe_recompile);
4512 }
4513
4514 if (!stopped()) {
4515 // How many elements will we copy from the original?
4516 // The answer is MinI(orig_tail, length).
4517 Node* moved = _gvn.transform(new MinINode(orig_tail, length));
4518
4519 // Generate a direct call to the right arraycopy function(s).
4520 // We know the copy is disjoint but we might not know if the
4521 // oop stores need checking.
4522 // Extreme case: Arrays.copyOf((Integer[])x, 10, String[].class).
4523 // This will fail a store-check if x contains any non-nulls.
4524
4525 // ArrayCopyNode:Ideal may transform the ArrayCopyNode to
4526 // loads/stores but it is legal only if we're sure the
4527 // Arrays.copyOf would succeed. So we need all input arguments
4528 // to the copyOf to be validated, including that the copy to the
4529 // new array won't trigger an ArrayStoreException. That subtype
4530 // check can be optimized if we know something on the type of
4531 // the input array from type speculation.
4532 if (_gvn.type(klass_node)->singleton()) {
4533 const TypeKlassPtr* subk = _gvn.type(load_object_klass(original))->is_klassptr();
4534 const TypeKlassPtr* superk = _gvn.type(klass_node)->is_klassptr();
4535
4536 int test = C->static_subtype_check(superk, subk);
4537 if (test != Compile::SSC_always_true && test != Compile::SSC_always_false) {
4538 const TypeOopPtr* t_original = _gvn.type(original)->is_oopptr();
4539 if (t_original->speculative_type() != nullptr) {
4540 original = maybe_cast_profiled_obj(original, t_original->speculative_type(), true);
4541 }
4542 }
4543 }
4544
4545 bool validated = false;
4546 // Reason_class_check rather than Reason_intrinsic because we
4547 // want to intrinsify even if this traps.
4548 if (!too_many_traps(Deoptimization::Reason_class_check)) {
4549 Node* not_subtype_ctrl = gen_subtype_check(original, klass_node);
4550
4551 if (not_subtype_ctrl != top()) {
4552 PreserveJVMState pjvms(this);
4553 set_control(not_subtype_ctrl);
4554 uncommon_trap(Deoptimization::Reason_class_check,
4555 Deoptimization::Action_make_not_entrant);
4556 assert(stopped(), "Should be stopped");
4557 }
4558 validated = true;
4559 }
4560
4561 if (!stopped()) {
4562 newcopy = new_array(klass_node, length, 0); // no arguments to push
4563
4564 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, original, start, newcopy, intcon(0), moved, true, true,
4565 load_object_klass(original), klass_node);
4566 if (!is_copyOfRange) {
4567 ac->set_copyof(validated);
4568 } else {
4569 ac->set_copyofrange(validated);
4570 }
4571 Node* n = _gvn.transform(ac);
4572 if (n == ac) {
4573 ac->connect_outputs(this);
4574 } else {
4575 assert(validated, "shouldn't transform if all arguments not validated");
4576 set_all_memory(n);
4577 }
4578 }
4579 }
4580 } // original reexecute is set back here
4581
4582 C->set_has_split_ifs(true); // Has chance for split-if optimization
4583 if (!stopped()) {
4584 set_result(newcopy);
4585 }
4586 return true;
4587 }
4588
4589
4590 //----------------------generate_virtual_guard---------------------------
4591 // Helper for hashCode and clone. Peeks inside the vtable to avoid a call.
4592 Node* LibraryCallKit::generate_virtual_guard(Node* obj_klass,
4593 RegionNode* slow_region) {
4594 ciMethod* method = callee();
4595 int vtable_index = method->vtable_index();
4596 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4597 "bad index %d", vtable_index);
4598 // Get the Method* out of the appropriate vtable entry.
4599 int entry_offset = in_bytes(Klass::vtable_start_offset()) +
4600 vtable_index*vtableEntry::size_in_bytes() +
4601 in_bytes(vtableEntry::method_offset());
4602 Node* entry_addr = basic_plus_adr(obj_klass, entry_offset);
4603 Node* target_call = make_load(nullptr, entry_addr, TypePtr::NOTNULL, T_ADDRESS, MemNode::unordered);
4604
4605 // Compare the target method with the expected method (e.g., Object.hashCode).
4606 const TypePtr* native_call_addr = TypeMetadataPtr::make(method);
4607
4608 Node* native_call = makecon(native_call_addr);
4609 Node* chk_native = _gvn.transform(new CmpPNode(target_call, native_call));
4610 Node* test_native = _gvn.transform(new BoolNode(chk_native, BoolTest::ne));
4611
4612 return generate_slow_guard(test_native, slow_region);
4613 }
4614
4615 //-----------------------generate_method_call----------------------------
4616 // Use generate_method_call to make a slow-call to the real
4617 // method if the fast path fails. An alternative would be to
4618 // use a stub like OptoRuntime::slow_arraycopy_Java.
4619 // This only works for expanding the current library call,
4620 // not another intrinsic. (E.g., don't use this for making an
4621 // arraycopy call inside of the copyOf intrinsic.)
4622 CallJavaNode*
4623 LibraryCallKit::generate_method_call(vmIntrinsicID method_id, bool is_virtual, bool is_static, bool res_not_null) {
4624 // When compiling the intrinsic method itself, do not use this technique.
4625 guarantee(callee() != C->method(), "cannot make slow-call to self");
4626
4627 ciMethod* method = callee();
4628 // ensure the JVMS we have will be correct for this call
4629 guarantee(method_id == method->intrinsic_id(), "must match");
4630
4631 const TypeFunc* tf = TypeFunc::make(method);
4632 if (res_not_null) {
4633 assert(tf->return_type() == T_OBJECT, "");
4634 const TypeTuple* range = tf->range();
4635 const Type** fields = TypeTuple::fields(range->cnt());
4636 fields[TypeFunc::Parms] = range->field_at(TypeFunc::Parms)->filter_speculative(TypePtr::NOTNULL);
4637 const TypeTuple* new_range = TypeTuple::make(range->cnt(), fields);
4638 tf = TypeFunc::make(tf->domain(), new_range);
4639 }
4640 CallJavaNode* slow_call;
4641 if (is_static) {
4642 assert(!is_virtual, "");
4643 slow_call = new CallStaticJavaNode(C, tf,
4644 SharedRuntime::get_resolve_static_call_stub(), method);
4645 } else if (is_virtual) {
4646 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4647 int vtable_index = Method::invalid_vtable_index;
4648 if (UseInlineCaches) {
4649 // Suppress the vtable call
4650 } else {
4651 // hashCode and clone are not a miranda methods,
4652 // so the vtable index is fixed.
4653 // No need to use the linkResolver to get it.
4654 vtable_index = method->vtable_index();
4655 assert(vtable_index >= 0 || vtable_index == Method::nonvirtual_vtable_index,
4656 "bad index %d", vtable_index);
4657 }
4658 slow_call = new CallDynamicJavaNode(tf,
4659 SharedRuntime::get_resolve_virtual_call_stub(),
4660 method, vtable_index);
4661 } else { // neither virtual nor static: opt_virtual
4662 assert(!gvn().type(argument(0))->maybe_null(), "should not be null");
4663 slow_call = new CallStaticJavaNode(C, tf,
4664 SharedRuntime::get_resolve_opt_virtual_call_stub(), method);
4665 slow_call->set_optimized_virtual(true);
4666 }
4667 if (CallGenerator::is_inlined_method_handle_intrinsic(this->method(), bci(), callee())) {
4668 // To be able to issue a direct call (optimized virtual or virtual)
4669 // and skip a call to MH.linkTo*/invokeBasic adapter, additional information
4670 // about the method being invoked should be attached to the call site to
4671 // make resolution logic work (see SharedRuntime::resolve_{virtual,opt_virtual}_call_C).
4672 slow_call->set_override_symbolic_info(true);
4673 }
4674 set_arguments_for_java_call(slow_call);
4675 set_edges_for_java_call(slow_call);
4676 return slow_call;
4677 }
4678
4679
4680 /**
4681 * Build special case code for calls to hashCode on an object. This call may
4682 * be virtual (invokevirtual) or bound (invokespecial). For each case we generate
4683 * slightly different code.
4684 */
4685 bool LibraryCallKit::inline_native_hashcode(bool is_virtual, bool is_static) {
4686 assert(is_static == callee()->is_static(), "correct intrinsic selection");
4687 assert(!(is_virtual && is_static), "either virtual, special, or static");
4688
4689 enum { _slow_path = 1, _null_path, _fast_path, _fast_path2, PATH_LIMIT };
4690
4691 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
4692 PhiNode* result_val = new PhiNode(result_reg, TypeInt::INT);
4693 PhiNode* result_io = new PhiNode(result_reg, Type::ABIO);
4694 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
4695 Node* obj = nullptr;
4696 if (!is_static) {
4697 // Check for hashing null object
4698 obj = null_check_receiver();
4699 if (stopped()) return true; // unconditionally null
4700 result_reg->init_req(_null_path, top());
4701 result_val->init_req(_null_path, top());
4702 } else {
4703 // Do a null check, and return zero if null.
4704 // System.identityHashCode(null) == 0
4705 obj = argument(0);
4706 Node* null_ctl = top();
4707 obj = null_check_oop(obj, &null_ctl);
4708 result_reg->init_req(_null_path, null_ctl);
4709 result_val->init_req(_null_path, _gvn.intcon(0));
4710 }
4711
4712 // Unconditionally null? Then return right away.
4713 if (stopped()) {
4714 set_control( result_reg->in(_null_path));
4715 if (!stopped())
4716 set_result(result_val->in(_null_path));
4717 return true;
4718 }
4719
4720 // We only go to the fast case code if we pass a number of guards. The
4721 // paths which do not pass are accumulated in the slow_region.
4722 RegionNode* slow_region = new RegionNode(1);
4723 record_for_igvn(slow_region);
4724
4725 // If this is a virtual call, we generate a funny guard. We pull out
4726 // the vtable entry corresponding to hashCode() from the target object.
4727 // If the target method which we are calling happens to be the native
4728 // Object hashCode() method, we pass the guard. We do not need this
4729 // guard for non-virtual calls -- the caller is known to be the native
4730 // Object hashCode().
4731 if (is_virtual) {
4732 // After null check, get the object's klass.
4733 Node* obj_klass = load_object_klass(obj);
4734 generate_virtual_guard(obj_klass, slow_region);
4735 }
4736
4737 if (UseCompactObjectHeaders) {
4738 // Get the header out of the object.
4739 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4740 // The control of the load must be null. Otherwise, the load can move before
4741 // the null check after castPP removal.
4742 Node* no_ctrl = nullptr;
4743 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4744
4745 // Test the header to see if the object is in hashed or copied state.
4746 Node* hashctrl_mask = _gvn.MakeConX(markWord::hashctrl_mask_in_place);
4747 Node* masked_header = _gvn.transform(new AndXNode(header, hashctrl_mask));
4748
4749 // Take slow-path when the object has not been hashed.
4750 Node* not_hashed_val = _gvn.MakeConX(0);
4751 Node* chk_hashed = _gvn.transform(new CmpXNode(masked_header, not_hashed_val));
4752 Node* test_hashed = _gvn.transform(new BoolNode(chk_hashed, BoolTest::eq));
4753
4754 generate_slow_guard(test_hashed, slow_region);
4755
4756 // Test whether the object is hashed or hashed&copied.
4757 Node* hashed_copied = _gvn.MakeConX(markWord::hashctrl_expanded_mask_in_place | markWord::hashctrl_hashed_mask_in_place);
4758 Node* chk_copied = _gvn.transform(new CmpXNode(masked_header, hashed_copied));
4759 // If true, then object has been hashed&copied, otherwise it's only hashed.
4760 Node* test_copied = _gvn.transform(new BoolNode(chk_copied, BoolTest::eq));
4761 IfNode* if_copied = create_and_map_if(control(), test_copied, PROB_FAIR, COUNT_UNKNOWN);
4762 Node* if_true = _gvn.transform(new IfTrueNode(if_copied));
4763 Node* if_false = _gvn.transform(new IfFalseNode(if_copied));
4764
4765 // Hashed&Copied path: read hash-code out of the object.
4766 set_control(if_true);
4767 // result_val->del_req(_fast_path2);
4768 // result_reg->del_req(_fast_path2);
4769 // result_io->del_req(_fast_path2);
4770 // result_mem->del_req(_fast_path2);
4771
4772 Node* obj_klass = load_object_klass(obj);
4773 Node* hash_addr;
4774 const TypeKlassPtr* klass_t = _gvn.type(obj_klass)->isa_klassptr();
4775 bool load_offset_runtime = true;
4776
4777 if (klass_t != nullptr) {
4778 if (klass_t->klass_is_exact() && klass_t->isa_instklassptr()) {
4779 ciInstanceKlass* ciKlass = reinterpret_cast<ciInstanceKlass*>(klass_t->is_instklassptr()->exact_klass());
4780 if (!ciKlass->is_mirror_instance_klass() && !ciKlass->is_reference_instance_klass()) {
4781 // We know the InstanceKlass, load hash_offset from there at compile-time.
4782 int hash_offset = ciKlass->hash_offset_in_bytes();
4783 hash_addr = basic_plus_adr(obj, hash_offset);
4784 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4785 result_val->init_req(_fast_path2, loaded_hash);
4786 result_reg->init_req(_fast_path2, control());
4787 load_offset_runtime = false;
4788 }
4789 }
4790 }
4791
4792 //tty->print_cr("Load hash-offset at runtime: %s", BOOL_TO_STR(load_offset_runtime));
4793
4794 if (load_offset_runtime) {
4795 // We don't know if it is an array or an exact type, figure it out at run-time.
4796 // If not an ordinary instance, then we need to take slow-path.
4797 Node* kind_addr = basic_plus_adr(obj_klass, Klass::kind_offset_in_bytes());
4798 Node* kind = make_load(control(), kind_addr, TypeInt::INT, T_INT, MemNode::unordered);
4799 Node* instance_val = _gvn.intcon(Klass::InstanceKlassKind);
4800 Node* chk_inst = _gvn.transform(new CmpINode(kind, instance_val));
4801 Node* test_inst = _gvn.transform(new BoolNode(chk_inst, BoolTest::ne));
4802 generate_slow_guard(test_inst, slow_region);
4803
4804 // Otherwise it's an instance and we can read the hash_offset from the InstanceKlass.
4805 Node* hash_offset_addr = basic_plus_adr(obj_klass, InstanceKlass::hash_offset_offset_in_bytes());
4806 Node* hash_offset = make_load(control(), hash_offset_addr, TypeInt::INT, T_INT, MemNode::unordered);
4807 // hash_offset->dump();
4808 Node* hash_addr = basic_plus_adr(obj, ConvI2X(hash_offset));
4809 Compile::current()->set_has_unsafe_access(true);
4810 Node* loaded_hash = make_load(control(), hash_addr, TypeInt::INT, T_INT, MemNode::unordered);
4811 result_val->init_req(_fast_path2, loaded_hash);
4812 result_reg->init_req(_fast_path2, control());
4813 }
4814
4815 // Hashed-only path: recompute hash-code from object address.
4816 set_control(if_false);
4817 // Our constants.
4818 Node* M = _gvn.intcon(0x337954D5);
4819 Node* A = _gvn.intcon(0xAAAAAAAA);
4820 // Split object address into lo and hi 32 bits.
4821 Node* obj_addr = _gvn.transform(new CastP2XNode(nullptr, obj));
4822 Node* x = _gvn.transform(new ConvL2INode(obj_addr));
4823 Node* upper_addr = _gvn.transform(new URShiftLNode(obj_addr, _gvn.intcon(32)));
4824 Node* y = _gvn.transform(new ConvL2INode(upper_addr));
4825
4826 Node* H0 = _gvn.transform(new XorINode(x, y));
4827 Node* L0 = _gvn.transform(new XorINode(x, A));
4828
4829 // Full multiplication of two 32 bit values L0 and M into a hi/lo result in two 32 bit values V0 and U0.
4830 Node* L0_64 = _gvn.transform(new ConvI2LNode(L0));
4831 L0_64 = _gvn.transform(new AndLNode(L0_64, _gvn.longcon(0xFFFFFFFF)));
4832 Node* M_64 = _gvn.transform(new ConvI2LNode(M));
4833 // M_64 = _gvn.transform(new AndLNode(M_64, _gvn.longcon(0xFFFFFFFF)));
4834 Node* prod64 = _gvn.transform(new MulLNode(L0_64, M_64));
4835 Node* V0 = _gvn.transform(new ConvL2INode(prod64));
4836 Node* prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4837 Node* U0 = _gvn.transform(new ConvL2INode(prod_upper));
4838
4839 Node* Q0 = _gvn.transform(new MulINode(H0, M));
4840 Node* L1 = _gvn.transform(new XorINode(Q0, U0));
4841
4842 // Full multiplication of two 32 bit values L1 and M into a hi/lo result in two 32 bit values V1 and U1.
4843 Node* L1_64 = _gvn.transform(new ConvI2LNode(L1));
4844 L1_64 = _gvn.transform(new AndLNode(L1_64, _gvn.longcon(0xFFFFFFFF)));
4845 prod64 = _gvn.transform(new MulLNode(L1_64, M_64));
4846 Node* V1 = _gvn.transform(new ConvL2INode(prod64));
4847 prod_upper = _gvn.transform(new URShiftLNode(prod64, _gvn.intcon(32)));
4848 Node* U1 = _gvn.transform(new ConvL2INode(prod_upper));
4849
4850 Node* P1 = _gvn.transform(new XorINode(V0, M));
4851
4852 // Right rotate P1 by distance L1.
4853 Node* distance = _gvn.transform(new AndINode(L1, _gvn.intcon(32 - 1)));
4854 Node* inverse_distance = _gvn.transform(new SubINode(_gvn.intcon(32), distance));
4855 Node* ror_part1 = _gvn.transform(new URShiftINode(P1, distance));
4856 Node* ror_part2 = _gvn.transform(new LShiftINode(P1, inverse_distance));
4857 Node* Q1 = _gvn.transform(new OrINode(ror_part1, ror_part2));
4858
4859 Node* L2 = _gvn.transform(new XorINode(Q1, U1));
4860 Node* hash = _gvn.transform(new XorINode(V1, L2));
4861 Node* hash_truncated = _gvn.transform(new AndINode(hash, _gvn.intcon(markWord::hash_mask)));
4862
4863 // TODO: We could generate a fast case here under the following conditions:
4864 // - The hashctrl is set to hash_is_copied (see markWord::hash_is_copied())
4865 // - The type of the object is known
4866 // Then we can load the identity hashcode from the int field at Klass::hash_offset_in_bytes() of the object.
4867 result_val->init_req(_fast_path, hash_truncated);
4868 } else {
4869 // Get the header out of the object, use LoadMarkNode when available
4870 Node* header_addr = basic_plus_adr(obj, oopDesc::mark_offset_in_bytes());
4871 // The control of the load must be null. Otherwise, the load can move before
4872 // the null check after castPP removal.
4873 Node* no_ctrl = nullptr;
4874 Node* header = make_load(no_ctrl, header_addr, TypeX_X, TypeX_X->basic_type(), MemNode::unordered);
4875
4876 if (!UseObjectMonitorTable) {
4877 // Test the header to see if it is safe to read w.r.t. locking.
4878 Node *lock_mask = _gvn.MakeConX(markWord::lock_mask_in_place);
4879 Node *lmasked_header = _gvn.transform(new AndXNode(header, lock_mask));
4880 if (LockingMode == LM_LIGHTWEIGHT) {
4881 Node *monitor_val = _gvn.MakeConX(markWord::monitor_value);
4882 Node *chk_monitor = _gvn.transform(new CmpXNode(lmasked_header, monitor_val));
4883 Node *test_monitor = _gvn.transform(new BoolNode(chk_monitor, BoolTest::eq));
4884
4885 generate_slow_guard(test_monitor, slow_region);
4886 } else {
4887 Node *unlocked_val = _gvn.MakeConX(markWord::unlocked_value);
4888 Node *chk_unlocked = _gvn.transform(new CmpXNode(lmasked_header, unlocked_val));
4889 Node *test_not_unlocked = _gvn.transform(new BoolNode(chk_unlocked, BoolTest::ne));
4890
4891 generate_slow_guard(test_not_unlocked, slow_region);
4892 }
4893 }
4894
4895 // Get the hash value and check to see that it has been properly assigned.
4896 // We depend on hash_mask being at most 32 bits and avoid the use of
4897 // hash_mask_in_place because it could be larger than 32 bits in a 64-bit
4898 // vm: see markWord.hpp.
4899 Node *hash_mask = _gvn.intcon(markWord::hash_mask);
4900 Node *hash_shift = _gvn.intcon(markWord::hash_shift);
4901 Node *hshifted_header= _gvn.transform(new URShiftXNode(header, hash_shift));
4902 // This hack lets the hash bits live anywhere in the mark object now, as long
4903 // as the shift drops the relevant bits into the low 32 bits. Note that
4904 // Java spec says that HashCode is an int so there's no point in capturing
4905 // an 'X'-sized hashcode (32 in 32-bit build or 64 in 64-bit build).
4906 hshifted_header = ConvX2I(hshifted_header);
4907 Node *hash_val = _gvn.transform(new AndINode(hshifted_header, hash_mask));
4908
4909 Node *no_hash_val = _gvn.intcon(markWord::no_hash);
4910 Node *chk_assigned = _gvn.transform(new CmpINode( hash_val, no_hash_val));
4911 Node *test_assigned = _gvn.transform(new BoolNode( chk_assigned, BoolTest::eq));
4912
4913 generate_slow_guard(test_assigned, slow_region);
4914
4915 result_val->init_req(_fast_path, hash_val);
4916
4917 // _fast_path2 is not used here.
4918 result_val->del_req(_fast_path2);
4919 result_reg->del_req(_fast_path2);
4920 result_io->del_req(_fast_path2);
4921 result_mem->del_req(_fast_path2);
4922 }
4923
4924 Node* init_mem = reset_memory();
4925 // fill in the rest of the null path:
4926 result_io ->init_req(_null_path, i_o());
4927 result_mem->init_req(_null_path, init_mem);
4928
4929 result_reg->init_req(_fast_path, control());
4930 result_io ->init_req(_fast_path, i_o());
4931 result_mem->init_req(_fast_path, init_mem);
4932
4933 if (UseCompactObjectHeaders) {
4934 result_io->init_req(_fast_path2, i_o());
4935 result_mem->init_req(_fast_path2, init_mem);
4936 }
4937
4938 // Generate code for the slow case. We make a call to hashCode().
4939 assert(slow_region != nullptr, "must have slow_region");
4940 set_control(_gvn.transform(slow_region));
4941 if (!stopped()) {
4942 // No need for PreserveJVMState, because we're using up the present state.
4943 set_all_memory(init_mem);
4944 vmIntrinsics::ID hashCode_id = is_static ? vmIntrinsics::_identityHashCode : vmIntrinsics::_hashCode;
4945 CallJavaNode* slow_call = generate_method_call(hashCode_id, is_virtual, is_static, false);
4946 Node* slow_result = set_results_for_java_call(slow_call);
4947 // this->control() comes from set_results_for_java_call
4948 result_reg->init_req(_slow_path, control());
4949 result_val->init_req(_slow_path, slow_result);
4950 result_io ->set_req(_slow_path, i_o());
4951 result_mem ->set_req(_slow_path, reset_memory());
4952 }
4953
4954 // Return the combined state.
4955 set_i_o( _gvn.transform(result_io) );
4956 set_all_memory( _gvn.transform(result_mem));
4957
4958 set_result(result_reg, result_val);
4959 return true;
4960 }
4961
4962 //---------------------------inline_native_getClass----------------------------
4963 // public final native Class<?> java.lang.Object.getClass();
4964 //
4965 // Build special case code for calls to getClass on an object.
4966 bool LibraryCallKit::inline_native_getClass() {
4967 Node* obj = null_check_receiver();
4968 if (stopped()) return true;
4969 set_result(load_mirror_from_klass(load_object_klass(obj)));
4970 return true;
4971 }
4972
4973 //-----------------inline_native_Reflection_getCallerClass---------------------
4974 // public static native Class<?> sun.reflect.Reflection.getCallerClass();
4975 //
4976 // In the presence of deep enough inlining, getCallerClass() becomes a no-op.
4977 //
4978 // NOTE: This code must perform the same logic as JVM_GetCallerClass
4979 // in that it must skip particular security frames and checks for
4980 // caller sensitive methods.
4981 bool LibraryCallKit::inline_native_Reflection_getCallerClass() {
4982 #ifndef PRODUCT
4983 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4984 tty->print_cr("Attempting to inline sun.reflect.Reflection.getCallerClass");
4985 }
4986 #endif
4987
4988 if (!jvms()->has_method()) {
4989 #ifndef PRODUCT
4990 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
4991 tty->print_cr(" Bailing out because intrinsic was inlined at top level");
4992 }
4993 #endif
4994 return false;
4995 }
4996
4997 // Walk back up the JVM state to find the caller at the required
4998 // depth.
4999 JVMState* caller_jvms = jvms();
5000
5001 // Cf. JVM_GetCallerClass
5002 // NOTE: Start the loop at depth 1 because the current JVM state does
5003 // not include the Reflection.getCallerClass() frame.
5004 for (int n = 1; caller_jvms != nullptr; caller_jvms = caller_jvms->caller(), n++) {
5005 ciMethod* m = caller_jvms->method();
5006 switch (n) {
5007 case 0:
5008 fatal("current JVM state does not include the Reflection.getCallerClass frame");
5009 break;
5010 case 1:
5011 // Frame 0 and 1 must be caller sensitive (see JVM_GetCallerClass).
5012 if (!m->caller_sensitive()) {
5013 #ifndef PRODUCT
5014 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5015 tty->print_cr(" Bailing out: CallerSensitive annotation expected at frame %d", n);
5016 }
5017 #endif
5018 return false; // bail-out; let JVM_GetCallerClass do the work
5019 }
5020 break;
5021 default:
5022 if (!m->is_ignored_by_security_stack_walk()) {
5023 // We have reached the desired frame; return the holder class.
5024 // Acquire method holder as java.lang.Class and push as constant.
5025 ciInstanceKlass* caller_klass = caller_jvms->method()->holder();
5026 ciInstance* caller_mirror = caller_klass->java_mirror();
5027 set_result(makecon(TypeInstPtr::make(caller_mirror)));
5028
5029 #ifndef PRODUCT
5030 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5031 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());
5032 tty->print_cr(" JVM state at this point:");
5033 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5034 ciMethod* m = jvms()->of_depth(i)->method();
5035 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5036 }
5037 }
5038 #endif
5039 return true;
5040 }
5041 break;
5042 }
5043 }
5044
5045 #ifndef PRODUCT
5046 if ((C->print_intrinsics() || C->print_inlining()) && Verbose) {
5047 tty->print_cr(" Bailing out because caller depth exceeded inlining depth = %d", jvms()->depth());
5048 tty->print_cr(" JVM state at this point:");
5049 for (int i = jvms()->depth(), n = 1; i >= 1; i--, n++) {
5050 ciMethod* m = jvms()->of_depth(i)->method();
5051 tty->print_cr(" %d) %s.%s", n, m->holder()->name()->as_utf8(), m->name()->as_utf8());
5052 }
5053 }
5054 #endif
5055
5056 return false; // bail-out; let JVM_GetCallerClass do the work
5057 }
5058
5059 bool LibraryCallKit::inline_fp_conversions(vmIntrinsics::ID id) {
5060 Node* arg = argument(0);
5061 Node* result = nullptr;
5062
5063 switch (id) {
5064 case vmIntrinsics::_floatToRawIntBits: result = new MoveF2INode(arg); break;
5065 case vmIntrinsics::_intBitsToFloat: result = new MoveI2FNode(arg); break;
5066 case vmIntrinsics::_doubleToRawLongBits: result = new MoveD2LNode(arg); break;
5067 case vmIntrinsics::_longBitsToDouble: result = new MoveL2DNode(arg); break;
5068 case vmIntrinsics::_floatToFloat16: result = new ConvF2HFNode(arg); break;
5069 case vmIntrinsics::_float16ToFloat: result = new ConvHF2FNode(arg); break;
5070
5071 case vmIntrinsics::_doubleToLongBits: {
5072 // two paths (plus control) merge in a wood
5073 RegionNode *r = new RegionNode(3);
5074 Node *phi = new PhiNode(r, TypeLong::LONG);
5075
5076 Node *cmpisnan = _gvn.transform(new CmpDNode(arg, arg));
5077 // Build the boolean node
5078 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5079
5080 // Branch either way.
5081 // NaN case is less traveled, which makes all the difference.
5082 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5083 Node *opt_isnan = _gvn.transform(ifisnan);
5084 assert( opt_isnan->is_If(), "Expect an IfNode");
5085 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5086 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5087
5088 set_control(iftrue);
5089
5090 static const jlong nan_bits = CONST64(0x7ff8000000000000);
5091 Node *slow_result = longcon(nan_bits); // return NaN
5092 phi->init_req(1, _gvn.transform( slow_result ));
5093 r->init_req(1, iftrue);
5094
5095 // Else fall through
5096 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5097 set_control(iffalse);
5098
5099 phi->init_req(2, _gvn.transform(new MoveD2LNode(arg)));
5100 r->init_req(2, iffalse);
5101
5102 // Post merge
5103 set_control(_gvn.transform(r));
5104 record_for_igvn(r);
5105
5106 C->set_has_split_ifs(true); // Has chance for split-if optimization
5107 result = phi;
5108 assert(result->bottom_type()->isa_long(), "must be");
5109 break;
5110 }
5111
5112 case vmIntrinsics::_floatToIntBits: {
5113 // two paths (plus control) merge in a wood
5114 RegionNode *r = new RegionNode(3);
5115 Node *phi = new PhiNode(r, TypeInt::INT);
5116
5117 Node *cmpisnan = _gvn.transform(new CmpFNode(arg, arg));
5118 // Build the boolean node
5119 Node *bolisnan = _gvn.transform(new BoolNode(cmpisnan, BoolTest::ne));
5120
5121 // Branch either way.
5122 // NaN case is less traveled, which makes all the difference.
5123 IfNode *ifisnan = create_and_xform_if(control(), bolisnan, PROB_STATIC_FREQUENT, COUNT_UNKNOWN);
5124 Node *opt_isnan = _gvn.transform(ifisnan);
5125 assert( opt_isnan->is_If(), "Expect an IfNode");
5126 IfNode *opt_ifisnan = (IfNode*)opt_isnan;
5127 Node *iftrue = _gvn.transform(new IfTrueNode(opt_ifisnan));
5128
5129 set_control(iftrue);
5130
5131 static const jint nan_bits = 0x7fc00000;
5132 Node *slow_result = makecon(TypeInt::make(nan_bits)); // return NaN
5133 phi->init_req(1, _gvn.transform( slow_result ));
5134 r->init_req(1, iftrue);
5135
5136 // Else fall through
5137 Node *iffalse = _gvn.transform(new IfFalseNode(opt_ifisnan));
5138 set_control(iffalse);
5139
5140 phi->init_req(2, _gvn.transform(new MoveF2INode(arg)));
5141 r->init_req(2, iffalse);
5142
5143 // Post merge
5144 set_control(_gvn.transform(r));
5145 record_for_igvn(r);
5146
5147 C->set_has_split_ifs(true); // Has chance for split-if optimization
5148 result = phi;
5149 assert(result->bottom_type()->isa_int(), "must be");
5150 break;
5151 }
5152
5153 default:
5154 fatal_unexpected_iid(id);
5155 break;
5156 }
5157 set_result(_gvn.transform(result));
5158 return true;
5159 }
5160
5161 bool LibraryCallKit::inline_fp_range_check(vmIntrinsics::ID id) {
5162 Node* arg = argument(0);
5163 Node* result = nullptr;
5164
5165 switch (id) {
5166 case vmIntrinsics::_floatIsInfinite:
5167 result = new IsInfiniteFNode(arg);
5168 break;
5169 case vmIntrinsics::_floatIsFinite:
5170 result = new IsFiniteFNode(arg);
5171 break;
5172 case vmIntrinsics::_doubleIsInfinite:
5173 result = new IsInfiniteDNode(arg);
5174 break;
5175 case vmIntrinsics::_doubleIsFinite:
5176 result = new IsFiniteDNode(arg);
5177 break;
5178 default:
5179 fatal_unexpected_iid(id);
5180 break;
5181 }
5182 set_result(_gvn.transform(result));
5183 return true;
5184 }
5185
5186 //----------------------inline_unsafe_copyMemory-------------------------
5187 // public native void Unsafe.copyMemory0(Object srcBase, long srcOffset, Object destBase, long destOffset, long bytes);
5188
5189 static bool has_wide_mem(PhaseGVN& gvn, Node* addr, Node* base) {
5190 const TypeAryPtr* addr_t = gvn.type(addr)->isa_aryptr();
5191 const Type* base_t = gvn.type(base);
5192
5193 bool in_native = (base_t == TypePtr::NULL_PTR);
5194 bool in_heap = !TypePtr::NULL_PTR->higher_equal(base_t);
5195 bool is_mixed = !in_heap && !in_native;
5196
5197 if (is_mixed) {
5198 return true; // mixed accesses can touch both on-heap and off-heap memory
5199 }
5200 if (in_heap) {
5201 bool is_prim_array = (addr_t != nullptr) && (addr_t->elem() != Type::BOTTOM);
5202 if (!is_prim_array) {
5203 // Though Unsafe.copyMemory() ensures at runtime for on-heap accesses that base is a primitive array,
5204 // there's not enough type information available to determine proper memory slice for it.
5205 return true;
5206 }
5207 }
5208 return false;
5209 }
5210
5211 bool LibraryCallKit::inline_unsafe_copyMemory() {
5212 if (callee()->is_static()) return false; // caller must have the capability!
5213 null_check_receiver(); // null-check receiver
5214 if (stopped()) return true;
5215
5216 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5217
5218 Node* src_base = argument(1); // type: oop
5219 Node* src_off = ConvL2X(argument(2)); // type: long
5220 Node* dst_base = argument(4); // type: oop
5221 Node* dst_off = ConvL2X(argument(5)); // type: long
5222 Node* size = ConvL2X(argument(7)); // type: long
5223
5224 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5225 "fieldOffset must be byte-scaled");
5226
5227 Node* src_addr = make_unsafe_address(src_base, src_off);
5228 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5229
5230 Node* thread = _gvn.transform(new ThreadLocalNode());
5231 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5232 BasicType doing_unsafe_access_bt = T_BYTE;
5233 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5234
5235 // update volatile field
5236 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5237
5238 int flags = RC_LEAF | RC_NO_FP;
5239
5240 const TypePtr* dst_type = TypePtr::BOTTOM;
5241
5242 // Adjust memory effects of the runtime call based on input values.
5243 if (!has_wide_mem(_gvn, src_addr, src_base) &&
5244 !has_wide_mem(_gvn, dst_addr, dst_base)) {
5245 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5246
5247 const TypePtr* src_type = _gvn.type(src_addr)->is_ptr();
5248 if (C->get_alias_index(src_type) == C->get_alias_index(dst_type)) {
5249 flags |= RC_NARROW_MEM; // narrow in memory
5250 }
5251 }
5252
5253 // Call it. Note that the length argument is not scaled.
5254 make_runtime_call(flags,
5255 OptoRuntime::fast_arraycopy_Type(),
5256 StubRoutines::unsafe_arraycopy(),
5257 "unsafe_arraycopy",
5258 dst_type,
5259 src_addr, dst_addr, size XTOP);
5260
5261 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5262
5263 return true;
5264 }
5265
5266 // unsafe_setmemory(void *base, ulong offset, size_t length, char fill_value);
5267 // Fill 'length' bytes starting from 'base[offset]' with 'fill_value'
5268 bool LibraryCallKit::inline_unsafe_setMemory() {
5269 if (callee()->is_static()) return false; // caller must have the capability!
5270 null_check_receiver(); // null-check receiver
5271 if (stopped()) return true;
5272
5273 C->set_has_unsafe_access(true); // Mark eventual nmethod as "unsafe".
5274
5275 Node* dst_base = argument(1); // type: oop
5276 Node* dst_off = ConvL2X(argument(2)); // type: long
5277 Node* size = ConvL2X(argument(4)); // type: long
5278 Node* byte = argument(6); // type: byte
5279
5280 assert(Unsafe_field_offset_to_byte_offset(11) == 11,
5281 "fieldOffset must be byte-scaled");
5282
5283 Node* dst_addr = make_unsafe_address(dst_base, dst_off);
5284
5285 Node* thread = _gvn.transform(new ThreadLocalNode());
5286 Node* doing_unsafe_access_addr = basic_plus_adr(top(), thread, in_bytes(JavaThread::doing_unsafe_access_offset()));
5287 BasicType doing_unsafe_access_bt = T_BYTE;
5288 assert((sizeof(bool) * CHAR_BIT) == 8, "not implemented");
5289
5290 // update volatile field
5291 store_to_memory(control(), doing_unsafe_access_addr, intcon(1), doing_unsafe_access_bt, MemNode::unordered);
5292
5293 int flags = RC_LEAF | RC_NO_FP;
5294
5295 const TypePtr* dst_type = TypePtr::BOTTOM;
5296
5297 // Adjust memory effects of the runtime call based on input values.
5298 if (!has_wide_mem(_gvn, dst_addr, dst_base)) {
5299 dst_type = _gvn.type(dst_addr)->is_ptr(); // narrow out memory
5300
5301 flags |= RC_NARROW_MEM; // narrow in memory
5302 }
5303
5304 // Call it. Note that the length argument is not scaled.
5305 make_runtime_call(flags,
5306 OptoRuntime::unsafe_setmemory_Type(),
5307 StubRoutines::unsafe_setmemory(),
5308 "unsafe_setmemory",
5309 dst_type,
5310 dst_addr, size XTOP, byte);
5311
5312 store_to_memory(control(), doing_unsafe_access_addr, intcon(0), doing_unsafe_access_bt, MemNode::unordered);
5313
5314 return true;
5315 }
5316
5317 #undef XTOP
5318
5319 //------------------------clone_coping-----------------------------------
5320 // Helper function for inline_native_clone.
5321 void LibraryCallKit::copy_to_clone(Node* obj, Node* alloc_obj, Node* obj_size, bool is_array) {
5322 assert(obj_size != nullptr, "");
5323 Node* raw_obj = alloc_obj->in(1);
5324 assert(alloc_obj->is_CheckCastPP() && raw_obj->is_Proj() && raw_obj->in(0)->is_Allocate(), "");
5325
5326 AllocateNode* alloc = nullptr;
5327 if (ReduceBulkZeroing &&
5328 // If we are implementing an array clone without knowing its source type
5329 // (can happen when compiling the array-guarded branch of a reflective
5330 // Object.clone() invocation), initialize the array within the allocation.
5331 // This is needed because some GCs (e.g. ZGC) might fall back in this case
5332 // to a runtime clone call that assumes fully initialized source arrays.
5333 (!is_array || obj->get_ptr_type()->isa_aryptr() != nullptr)) {
5334 // We will be completely responsible for initializing this object -
5335 // mark Initialize node as complete.
5336 alloc = AllocateNode::Ideal_allocation(alloc_obj);
5337 // The object was just allocated - there should be no any stores!
5338 guarantee(alloc != nullptr && alloc->maybe_set_complete(&_gvn), "");
5339 // Mark as complete_with_arraycopy so that on AllocateNode
5340 // expansion, we know this AllocateNode is initialized by an array
5341 // copy and a StoreStore barrier exists after the array copy.
5342 alloc->initialization()->set_complete_with_arraycopy();
5343 }
5344
5345 Node* size = _gvn.transform(obj_size);
5346 access_clone(obj, alloc_obj, size, is_array);
5347
5348 // Do not let reads from the cloned object float above the arraycopy.
5349 if (alloc != nullptr) {
5350 // Do not let stores that initialize this object be reordered with
5351 // a subsequent store that would make this object accessible by
5352 // other threads.
5353 // Record what AllocateNode this StoreStore protects so that
5354 // escape analysis can go from the MemBarStoreStoreNode to the
5355 // AllocateNode and eliminate the MemBarStoreStoreNode if possible
5356 // based on the escape status of the AllocateNode.
5357 insert_mem_bar(Op_MemBarStoreStore, alloc->proj_out_or_null(AllocateNode::RawAddress));
5358 } else {
5359 insert_mem_bar(Op_MemBarCPUOrder);
5360 }
5361 }
5362
5363 //------------------------inline_native_clone----------------------------
5364 // protected native Object java.lang.Object.clone();
5365 //
5366 // Here are the simple edge cases:
5367 // null receiver => normal trap
5368 // virtual and clone was overridden => slow path to out-of-line clone
5369 // not cloneable or finalizer => slow path to out-of-line Object.clone
5370 //
5371 // The general case has two steps, allocation and copying.
5372 // Allocation has two cases, and uses GraphKit::new_instance or new_array.
5373 //
5374 // Copying also has two cases, oop arrays and everything else.
5375 // Oop arrays use arrayof_oop_arraycopy (same as System.arraycopy).
5376 // Everything else uses the tight inline loop supplied by CopyArrayNode.
5377 //
5378 // These steps fold up nicely if and when the cloned object's klass
5379 // can be sharply typed as an object array, a type array, or an instance.
5380 //
5381 bool LibraryCallKit::inline_native_clone(bool is_virtual) {
5382 PhiNode* result_val;
5383
5384 // Set the reexecute bit for the interpreter to reexecute
5385 // the bytecode that invokes Object.clone if deoptimization happens.
5386 { PreserveReexecuteState preexecs(this);
5387 jvms()->set_should_reexecute(true);
5388
5389 Node* obj = null_check_receiver();
5390 if (stopped()) return true;
5391
5392 const TypeOopPtr* obj_type = _gvn.type(obj)->is_oopptr();
5393
5394 // If we are going to clone an instance, we need its exact type to
5395 // know the number and types of fields to convert the clone to
5396 // loads/stores. Maybe a speculative type can help us.
5397 if (!obj_type->klass_is_exact() &&
5398 obj_type->speculative_type() != nullptr &&
5399 obj_type->speculative_type()->is_instance_klass()) {
5400 ciInstanceKlass* spec_ik = obj_type->speculative_type()->as_instance_klass();
5401 if (spec_ik->nof_nonstatic_fields() <= ArrayCopyLoadStoreMaxElem &&
5402 !spec_ik->has_injected_fields()) {
5403 if (!obj_type->isa_instptr() ||
5404 obj_type->is_instptr()->instance_klass()->has_subklass()) {
5405 obj = maybe_cast_profiled_obj(obj, obj_type->speculative_type(), false);
5406 }
5407 }
5408 }
5409
5410 // Conservatively insert a memory barrier on all memory slices.
5411 // Do not let writes into the original float below the clone.
5412 insert_mem_bar(Op_MemBarCPUOrder);
5413
5414 // paths into result_reg:
5415 enum {
5416 _slow_path = 1, // out-of-line call to clone method (virtual or not)
5417 _objArray_path, // plain array allocation, plus arrayof_oop_arraycopy
5418 _array_path, // plain array allocation, plus arrayof_long_arraycopy
5419 _instance_path, // plain instance allocation, plus arrayof_long_arraycopy
5420 PATH_LIMIT
5421 };
5422 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
5423 result_val = new PhiNode(result_reg, TypeInstPtr::NOTNULL);
5424 PhiNode* result_i_o = new PhiNode(result_reg, Type::ABIO);
5425 PhiNode* result_mem = new PhiNode(result_reg, Type::MEMORY, TypePtr::BOTTOM);
5426 record_for_igvn(result_reg);
5427
5428 Node* obj_klass = load_object_klass(obj);
5429 Node* array_obj = obj;
5430 Node* array_ctl = generate_array_guard(obj_klass, (RegionNode*)nullptr, &array_obj);
5431 if (array_ctl != nullptr) {
5432 // It's an array.
5433 PreserveJVMState pjvms(this);
5434 set_control(array_ctl);
5435 Node* obj_length = load_array_length(array_obj);
5436 Node* array_size = nullptr; // Size of the array without object alignment padding.
5437 Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
5438
5439 BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
5440 if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
5441 // If it is an oop array, it requires very special treatment,
5442 // because gc barriers are required when accessing the array.
5443 Node* is_obja = generate_objArray_guard(obj_klass, (RegionNode*)nullptr);
5444 if (is_obja != nullptr) {
5445 PreserveJVMState pjvms2(this);
5446 set_control(is_obja);
5447 // Generate a direct call to the right arraycopy function(s).
5448 // Clones are always tightly coupled.
5449 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, array_obj, intcon(0), alloc_obj, intcon(0), obj_length, true, false);
5450 ac->set_clone_oop_array();
5451 Node* n = _gvn.transform(ac);
5452 assert(n == ac, "cannot disappear");
5453 ac->connect_outputs(this, /*deoptimize_on_exception=*/true);
5454
5455 result_reg->init_req(_objArray_path, control());
5456 result_val->init_req(_objArray_path, alloc_obj);
5457 result_i_o ->set_req(_objArray_path, i_o());
5458 result_mem ->set_req(_objArray_path, reset_memory());
5459 }
5460 }
5461 // Otherwise, there are no barriers to worry about.
5462 // (We can dispense with card marks if we know the allocation
5463 // comes out of eden (TLAB)... In fact, ReduceInitialCardMarks
5464 // causes the non-eden paths to take compensating steps to
5465 // simulate a fresh allocation, so that no further
5466 // card marks are required in compiled code to initialize
5467 // the object.)
5468
5469 if (!stopped()) {
5470 copy_to_clone(array_obj, alloc_obj, array_size, true);
5471
5472 // Present the results of the copy.
5473 result_reg->init_req(_array_path, control());
5474 result_val->init_req(_array_path, alloc_obj);
5475 result_i_o ->set_req(_array_path, i_o());
5476 result_mem ->set_req(_array_path, reset_memory());
5477 }
5478 }
5479
5480 // We only go to the instance fast case code if we pass a number of guards.
5481 // The paths which do not pass are accumulated in the slow_region.
5482 RegionNode* slow_region = new RegionNode(1);
5483 record_for_igvn(slow_region);
5484 if (!stopped()) {
5485 // It's an instance (we did array above). Make the slow-path tests.
5486 // If this is a virtual call, we generate a funny guard. We grab
5487 // the vtable entry corresponding to clone() from the target object.
5488 // If the target method which we are calling happens to be the
5489 // Object clone() method, we pass the guard. We do not need this
5490 // guard for non-virtual calls; the caller is known to be the native
5491 // Object clone().
5492 if (is_virtual) {
5493 generate_virtual_guard(obj_klass, slow_region);
5494 }
5495
5496 // The object must be easily cloneable and must not have a finalizer.
5497 // Both of these conditions may be checked in a single test.
5498 // We could optimize the test further, but we don't care.
5499 generate_misc_flags_guard(obj_klass,
5500 // Test both conditions:
5501 KlassFlags::_misc_is_cloneable_fast | KlassFlags::_misc_has_finalizer,
5502 // Must be cloneable but not finalizer:
5503 KlassFlags::_misc_is_cloneable_fast,
5504 slow_region);
5505 }
5506
5507 if (!stopped()) {
5508 // It's an instance, and it passed the slow-path tests.
5509 PreserveJVMState pjvms(this);
5510 Node* obj_size = nullptr; // Total object size, including object alignment padding.
5511 // Need to deoptimize on exception from allocation since Object.clone intrinsic
5512 // is reexecuted if deoptimization occurs and there could be problems when merging
5513 // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).
5514 Node* alloc_obj = new_instance(obj_klass, nullptr, &obj_size, /*deoptimize_on_exception=*/true);
5515
5516 copy_to_clone(obj, alloc_obj, obj_size, false);
5517
5518 // Present the results of the slow call.
5519 result_reg->init_req(_instance_path, control());
5520 result_val->init_req(_instance_path, alloc_obj);
5521 result_i_o ->set_req(_instance_path, i_o());
5522 result_mem ->set_req(_instance_path, reset_memory());
5523 }
5524
5525 // Generate code for the slow case. We make a call to clone().
5526 set_control(_gvn.transform(slow_region));
5527 if (!stopped()) {
5528 PreserveJVMState pjvms(this);
5529 CallJavaNode* slow_call = generate_method_call(vmIntrinsics::_clone, is_virtual, false, true);
5530 // We need to deoptimize on exception (see comment above)
5531 Node* slow_result = set_results_for_java_call(slow_call, false, /* deoptimize */ true);
5532 // this->control() comes from set_results_for_java_call
5533 result_reg->init_req(_slow_path, control());
5534 result_val->init_req(_slow_path, slow_result);
5535 result_i_o ->set_req(_slow_path, i_o());
5536 result_mem ->set_req(_slow_path, reset_memory());
5537 }
5538
5539 // Return the combined state.
5540 set_control( _gvn.transform(result_reg));
5541 set_i_o( _gvn.transform(result_i_o));
5542 set_all_memory( _gvn.transform(result_mem));
5543 } // original reexecute is set back here
5544
5545 set_result(_gvn.transform(result_val));
5546 return true;
5547 }
5548
5549 // If we have a tightly coupled allocation, the arraycopy may take care
5550 // of the array initialization. If one of the guards we insert between
5551 // the allocation and the arraycopy causes a deoptimization, an
5552 // uninitialized array will escape the compiled method. To prevent that
5553 // we set the JVM state for uncommon traps between the allocation and
5554 // the arraycopy to the state before the allocation so, in case of
5555 // deoptimization, we'll reexecute the allocation and the
5556 // initialization.
5557 JVMState* LibraryCallKit::arraycopy_restore_alloc_state(AllocateArrayNode* alloc, int& saved_reexecute_sp) {
5558 if (alloc != nullptr) {
5559 ciMethod* trap_method = alloc->jvms()->method();
5560 int trap_bci = alloc->jvms()->bci();
5561
5562 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
5563 !C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_null_check)) {
5564 // Make sure there's no store between the allocation and the
5565 // arraycopy otherwise visible side effects could be rexecuted
5566 // in case of deoptimization and cause incorrect execution.
5567 bool no_interfering_store = true;
5568 Node* mem = alloc->in(TypeFunc::Memory);
5569 if (mem->is_MergeMem()) {
5570 for (MergeMemStream mms(merged_memory(), mem->as_MergeMem()); mms.next_non_empty2(); ) {
5571 Node* n = mms.memory();
5572 if (n != mms.memory2() && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5573 assert(n->is_Store(), "what else?");
5574 no_interfering_store = false;
5575 break;
5576 }
5577 }
5578 } else {
5579 for (MergeMemStream mms(merged_memory()); mms.next_non_empty(); ) {
5580 Node* n = mms.memory();
5581 if (n != mem && !(n->is_Proj() && n->in(0) == alloc->initialization())) {
5582 assert(n->is_Store(), "what else?");
5583 no_interfering_store = false;
5584 break;
5585 }
5586 }
5587 }
5588
5589 if (no_interfering_store) {
5590 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5591
5592 JVMState* saved_jvms = jvms();
5593 saved_reexecute_sp = _reexecute_sp;
5594
5595 set_jvms(sfpt->jvms());
5596 _reexecute_sp = jvms()->sp();
5597
5598 return saved_jvms;
5599 }
5600 }
5601 }
5602 return nullptr;
5603 }
5604
5605 // Clone the JVMState of the array allocation and create a new safepoint with it. Re-push the array length to the stack
5606 // such that uncommon traps can be emitted to re-execute the array allocation in the interpreter.
5607 SafePointNode* LibraryCallKit::create_safepoint_with_state_before_array_allocation(const AllocateArrayNode* alloc) const {
5608 JVMState* old_jvms = alloc->jvms()->clone_shallow(C);
5609 uint size = alloc->req();
5610 SafePointNode* sfpt = new SafePointNode(size, old_jvms);
5611 old_jvms->set_map(sfpt);
5612 for (uint i = 0; i < size; i++) {
5613 sfpt->init_req(i, alloc->in(i));
5614 }
5615 // re-push array length for deoptimization
5616 sfpt->ins_req(old_jvms->stkoff() + old_jvms->sp(), alloc->in(AllocateNode::ALength));
5617 old_jvms->set_sp(old_jvms->sp()+1);
5618 old_jvms->set_monoff(old_jvms->monoff()+1);
5619 old_jvms->set_scloff(old_jvms->scloff()+1);
5620 old_jvms->set_endoff(old_jvms->endoff()+1);
5621 old_jvms->set_should_reexecute(true);
5622
5623 sfpt->set_i_o(map()->i_o());
5624 sfpt->set_memory(map()->memory());
5625 sfpt->set_control(map()->control());
5626 return sfpt;
5627 }
5628
5629 // In case of a deoptimization, we restart execution at the
5630 // allocation, allocating a new array. We would leave an uninitialized
5631 // array in the heap that GCs wouldn't expect. Move the allocation
5632 // after the traps so we don't allocate the array if we
5633 // deoptimize. This is possible because tightly_coupled_allocation()
5634 // guarantees there's no observer of the allocated array at this point
5635 // and the control flow is simple enough.
5636 void LibraryCallKit::arraycopy_move_allocation_here(AllocateArrayNode* alloc, Node* dest, JVMState* saved_jvms_before_guards,
5637 int saved_reexecute_sp, uint new_idx) {
5638 if (saved_jvms_before_guards != nullptr && !stopped()) {
5639 replace_unrelated_uncommon_traps_with_alloc_state(alloc, saved_jvms_before_guards);
5640
5641 assert(alloc != nullptr, "only with a tightly coupled allocation");
5642 // restore JVM state to the state at the arraycopy
5643 saved_jvms_before_guards->map()->set_control(map()->control());
5644 assert(saved_jvms_before_guards->map()->memory() == map()->memory(), "memory state changed?");
5645 assert(saved_jvms_before_guards->map()->i_o() == map()->i_o(), "IO state changed?");
5646 // If we've improved the types of some nodes (null check) while
5647 // emitting the guards, propagate them to the current state
5648 map()->replaced_nodes().apply(saved_jvms_before_guards->map(), new_idx);
5649 set_jvms(saved_jvms_before_guards);
5650 _reexecute_sp = saved_reexecute_sp;
5651
5652 // Remove the allocation from above the guards
5653 CallProjections callprojs;
5654 alloc->extract_projections(&callprojs, true);
5655 InitializeNode* init = alloc->initialization();
5656 Node* alloc_mem = alloc->in(TypeFunc::Memory);
5657 C->gvn_replace_by(callprojs.fallthrough_ioproj, alloc->in(TypeFunc::I_O));
5658 C->gvn_replace_by(init->proj_out(TypeFunc::Memory), alloc_mem);
5659
5660 // The CastIINode created in GraphKit::new_array (in AllocateArrayNode::make_ideal_length) must stay below
5661 // the allocation (i.e. is only valid if the allocation succeeds):
5662 // 1) replace CastIINode with AllocateArrayNode's length here
5663 // 2) Create CastIINode again once allocation has moved (see below) at the end of this method
5664 //
5665 // Multiple identical CastIINodes might exist here. Each GraphKit::load_array_length() call will generate
5666 // new separate CastIINode (arraycopy guard checks or any array length use between array allocation and ararycopy)
5667 Node* init_control = init->proj_out(TypeFunc::Control);
5668 Node* alloc_length = alloc->Ideal_length();
5669 #ifdef ASSERT
5670 Node* prev_cast = nullptr;
5671 #endif
5672 for (uint i = 0; i < init_control->outcnt(); i++) {
5673 Node* init_out = init_control->raw_out(i);
5674 if (init_out->is_CastII() && init_out->in(TypeFunc::Control) == init_control && init_out->in(1) == alloc_length) {
5675 #ifdef ASSERT
5676 if (prev_cast == nullptr) {
5677 prev_cast = init_out;
5678 } else {
5679 if (prev_cast->cmp(*init_out) == false) {
5680 prev_cast->dump();
5681 init_out->dump();
5682 assert(false, "not equal CastIINode");
5683 }
5684 }
5685 #endif
5686 C->gvn_replace_by(init_out, alloc_length);
5687 }
5688 }
5689 C->gvn_replace_by(init->proj_out(TypeFunc::Control), alloc->in(0));
5690
5691 // move the allocation here (after the guards)
5692 _gvn.hash_delete(alloc);
5693 alloc->set_req(TypeFunc::Control, control());
5694 alloc->set_req(TypeFunc::I_O, i_o());
5695 Node *mem = reset_memory();
5696 set_all_memory(mem);
5697 alloc->set_req(TypeFunc::Memory, mem);
5698 set_control(init->proj_out_or_null(TypeFunc::Control));
5699 set_i_o(callprojs.fallthrough_ioproj);
5700
5701 // Update memory as done in GraphKit::set_output_for_allocation()
5702 const TypeInt* length_type = _gvn.find_int_type(alloc->in(AllocateNode::ALength));
5703 const TypeOopPtr* ary_type = _gvn.type(alloc->in(AllocateNode::KlassNode))->is_klassptr()->as_instance_type();
5704 if (ary_type->isa_aryptr() && length_type != nullptr) {
5705 ary_type = ary_type->is_aryptr()->cast_to_size(length_type);
5706 }
5707 const TypePtr* telemref = ary_type->add_offset(Type::OffsetBot);
5708 int elemidx = C->get_alias_index(telemref);
5709 set_memory(init->proj_out_or_null(TypeFunc::Memory), Compile::AliasIdxRaw);
5710 set_memory(init->proj_out_or_null(TypeFunc::Memory), elemidx);
5711
5712 Node* allocx = _gvn.transform(alloc);
5713 assert(allocx == alloc, "where has the allocation gone?");
5714 assert(dest->is_CheckCastPP(), "not an allocation result?");
5715
5716 _gvn.hash_delete(dest);
5717 dest->set_req(0, control());
5718 Node* destx = _gvn.transform(dest);
5719 assert(destx == dest, "where has the allocation result gone?");
5720
5721 array_ideal_length(alloc, ary_type, true);
5722 }
5723 }
5724
5725 // Unrelated UCTs between the array allocation and the array copy, which are considered safe by tightly_coupled_allocation(),
5726 // need to be replaced by an UCT with a state before the array allocation (including the array length). This is necessary
5727 // because we could hit one of these UCTs (which are executed before the emitted array copy guards and the actual array
5728 // allocation which is moved down in arraycopy_move_allocation_here()). When later resuming execution in the interpreter,
5729 // we would have wrongly skipped the array allocation. To prevent this, we resume execution at the array allocation in
5730 // the interpreter similar to what we are doing for the newly emitted guards for the array copy.
5731 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(AllocateArrayNode* alloc,
5732 JVMState* saved_jvms_before_guards) {
5733 if (saved_jvms_before_guards->map()->control()->is_IfProj()) {
5734 // There is at least one unrelated uncommon trap which needs to be replaced.
5735 SafePointNode* sfpt = create_safepoint_with_state_before_array_allocation(alloc);
5736
5737 JVMState* saved_jvms = jvms();
5738 const int saved_reexecute_sp = _reexecute_sp;
5739 set_jvms(sfpt->jvms());
5740 _reexecute_sp = jvms()->sp();
5741
5742 replace_unrelated_uncommon_traps_with_alloc_state(saved_jvms_before_guards);
5743
5744 // Restore state
5745 set_jvms(saved_jvms);
5746 _reexecute_sp = saved_reexecute_sp;
5747 }
5748 }
5749
5750 // Replace the unrelated uncommon traps with new uncommon trap nodes by reusing the action and reason. The new uncommon
5751 // traps will have the state of the array allocation. Let the old uncommon trap nodes die.
5752 void LibraryCallKit::replace_unrelated_uncommon_traps_with_alloc_state(JVMState* saved_jvms_before_guards) {
5753 Node* if_proj = saved_jvms_before_guards->map()->control(); // Start the search right before the newly emitted guards
5754 while (if_proj->is_IfProj()) {
5755 CallStaticJavaNode* uncommon_trap = get_uncommon_trap_from_success_proj(if_proj);
5756 if (uncommon_trap != nullptr) {
5757 create_new_uncommon_trap(uncommon_trap);
5758 }
5759 assert(if_proj->in(0)->is_If(), "must be If");
5760 if_proj = if_proj->in(0)->in(0);
5761 }
5762 assert(if_proj->is_Proj() && if_proj->in(0)->is_Initialize(),
5763 "must have reached control projection of init node");
5764 }
5765
5766 void LibraryCallKit::create_new_uncommon_trap(CallStaticJavaNode* uncommon_trap_call) {
5767 const int trap_request = uncommon_trap_call->uncommon_trap_request();
5768 assert(trap_request != 0, "no valid UCT trap request");
5769 PreserveJVMState pjvms(this);
5770 set_control(uncommon_trap_call->in(0));
5771 uncommon_trap(Deoptimization::trap_request_reason(trap_request),
5772 Deoptimization::trap_request_action(trap_request));
5773 assert(stopped(), "Should be stopped");
5774 _gvn.hash_delete(uncommon_trap_call);
5775 uncommon_trap_call->set_req(0, top()); // not used anymore, kill it
5776 }
5777
5778 // Common checks for array sorting intrinsics arguments.
5779 // Returns `true` if checks passed.
5780 bool LibraryCallKit::check_array_sort_arguments(Node* elementType, Node* obj, BasicType& bt) {
5781 // check address of the class
5782 if (elementType == nullptr || elementType->is_top()) {
5783 return false; // dead path
5784 }
5785 const TypeInstPtr* elem_klass = gvn().type(elementType)->isa_instptr();
5786 if (elem_klass == nullptr) {
5787 return false; // dead path
5788 }
5789 // java_mirror_type() returns non-null for compile-time Class constants only
5790 ciType* elem_type = elem_klass->java_mirror_type();
5791 if (elem_type == nullptr) {
5792 return false;
5793 }
5794 bt = elem_type->basic_type();
5795 // Disable the intrinsic if the CPU does not support SIMD sort
5796 if (!Matcher::supports_simd_sort(bt)) {
5797 return false;
5798 }
5799 // check address of the array
5800 if (obj == nullptr || obj->is_top()) {
5801 return false; // dead path
5802 }
5803 const TypeAryPtr* obj_t = _gvn.type(obj)->isa_aryptr();
5804 if (obj_t == nullptr || obj_t->elem() == Type::BOTTOM) {
5805 return false; // failed input validation
5806 }
5807 return true;
5808 }
5809
5810 //------------------------------inline_array_partition-----------------------
5811 bool LibraryCallKit::inline_array_partition() {
5812 address stubAddr = StubRoutines::select_array_partition_function();
5813 if (stubAddr == nullptr) {
5814 return false; // Intrinsic's stub is not implemented on this platform
5815 }
5816 assert(callee()->signature()->size() == 9, "arrayPartition has 8 parameters (one long)");
5817
5818 // no receiver because it is a static method
5819 Node* elementType = argument(0);
5820 Node* obj = argument(1);
5821 Node* offset = argument(2); // long
5822 Node* fromIndex = argument(4);
5823 Node* toIndex = argument(5);
5824 Node* indexPivot1 = argument(6);
5825 Node* indexPivot2 = argument(7);
5826 // PartitionOperation: argument(8) is ignored
5827
5828 Node* pivotIndices = nullptr;
5829 BasicType bt = T_ILLEGAL;
5830
5831 if (!check_array_sort_arguments(elementType, obj, bt)) {
5832 return false;
5833 }
5834 null_check(obj);
5835 // If obj is dead, only null-path is taken.
5836 if (stopped()) {
5837 return true;
5838 }
5839 // Set the original stack and the reexecute bit for the interpreter to reexecute
5840 // the bytecode that invokes DualPivotQuicksort.partition() if deoptimization happens.
5841 { PreserveReexecuteState preexecs(this);
5842 jvms()->set_should_reexecute(true);
5843
5844 Node* obj_adr = make_unsafe_address(obj, offset);
5845
5846 // create the pivotIndices array of type int and size = 2
5847 Node* size = intcon(2);
5848 Node* klass_node = makecon(TypeKlassPtr::make(ciTypeArrayKlass::make(T_INT)));
5849 pivotIndices = new_array(klass_node, size, 0); // no arguments to push
5850 AllocateArrayNode* alloc = tightly_coupled_allocation(pivotIndices);
5851 guarantee(alloc != nullptr, "created above");
5852 Node* pivotIndices_adr = basic_plus_adr(pivotIndices, arrayOopDesc::base_offset_in_bytes(T_INT));
5853
5854 // pass the basic type enum to the stub
5855 Node* elemType = intcon(bt);
5856
5857 // Call the stub
5858 const char *stubName = "array_partition_stub";
5859 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_partition_Type(),
5860 stubAddr, stubName, TypePtr::BOTTOM,
5861 obj_adr, elemType, fromIndex, toIndex, pivotIndices_adr,
5862 indexPivot1, indexPivot2);
5863
5864 } // original reexecute is set back here
5865
5866 if (!stopped()) {
5867 set_result(pivotIndices);
5868 }
5869
5870 return true;
5871 }
5872
5873
5874 //------------------------------inline_array_sort-----------------------
5875 bool LibraryCallKit::inline_array_sort() {
5876 address stubAddr = StubRoutines::select_arraysort_function();
5877 if (stubAddr == nullptr) {
5878 return false; // Intrinsic's stub is not implemented on this platform
5879 }
5880 assert(callee()->signature()->size() == 7, "arraySort has 6 parameters (one long)");
5881
5882 // no receiver because it is a static method
5883 Node* elementType = argument(0);
5884 Node* obj = argument(1);
5885 Node* offset = argument(2); // long
5886 Node* fromIndex = argument(4);
5887 Node* toIndex = argument(5);
5888 // SortOperation: argument(6) is ignored
5889
5890 BasicType bt = T_ILLEGAL;
5891
5892 if (!check_array_sort_arguments(elementType, obj, bt)) {
5893 return false;
5894 }
5895 null_check(obj);
5896 // If obj is dead, only null-path is taken.
5897 if (stopped()) {
5898 return true;
5899 }
5900 Node* obj_adr = make_unsafe_address(obj, offset);
5901
5902 // pass the basic type enum to the stub
5903 Node* elemType = intcon(bt);
5904
5905 // Call the stub.
5906 const char *stubName = "arraysort_stub";
5907 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::array_sort_Type(),
5908 stubAddr, stubName, TypePtr::BOTTOM,
5909 obj_adr, elemType, fromIndex, toIndex);
5910
5911 return true;
5912 }
5913
5914
5915 //------------------------------inline_arraycopy-----------------------
5916 // public static native void java.lang.System.arraycopy(Object src, int srcPos,
5917 // Object dest, int destPos,
5918 // int length);
5919 bool LibraryCallKit::inline_arraycopy() {
5920 // Get the arguments.
5921 Node* src = argument(0); // type: oop
5922 Node* src_offset = argument(1); // type: int
5923 Node* dest = argument(2); // type: oop
5924 Node* dest_offset = argument(3); // type: int
5925 Node* length = argument(4); // type: int
5926
5927 uint new_idx = C->unique();
5928
5929 // Check for allocation before we add nodes that would confuse
5930 // tightly_coupled_allocation()
5931 AllocateArrayNode* alloc = tightly_coupled_allocation(dest);
5932
5933 int saved_reexecute_sp = -1;
5934 JVMState* saved_jvms_before_guards = arraycopy_restore_alloc_state(alloc, saved_reexecute_sp);
5935 // See arraycopy_restore_alloc_state() comment
5936 // if alloc == null we don't have to worry about a tightly coupled allocation so we can emit all needed guards
5937 // if saved_jvms_before_guards is not null (then alloc is not null) then we can handle guards and a tightly coupled allocation
5938 // if saved_jvms_before_guards is null and alloc is not null, we can't emit any guards
5939 bool can_emit_guards = (alloc == nullptr || saved_jvms_before_guards != nullptr);
5940
5941 // The following tests must be performed
5942 // (1) src and dest are arrays.
5943 // (2) src and dest arrays must have elements of the same BasicType
5944 // (3) src and dest must not be null.
5945 // (4) src_offset must not be negative.
5946 // (5) dest_offset must not be negative.
5947 // (6) length must not be negative.
5948 // (7) src_offset + length must not exceed length of src.
5949 // (8) dest_offset + length must not exceed length of dest.
5950 // (9) each element of an oop array must be assignable
5951
5952 // (3) src and dest must not be null.
5953 // always do this here because we need the JVM state for uncommon traps
5954 Node* null_ctl = top();
5955 src = saved_jvms_before_guards != nullptr ? null_check_oop(src, &null_ctl, true, true) : null_check(src, T_ARRAY);
5956 assert(null_ctl->is_top(), "no null control here");
5957 dest = null_check(dest, T_ARRAY);
5958
5959 if (!can_emit_guards) {
5960 // if saved_jvms_before_guards is null and alloc is not null, we don't emit any
5961 // guards but the arraycopy node could still take advantage of a
5962 // tightly allocated allocation. tightly_coupled_allocation() is
5963 // called again to make sure it takes the null check above into
5964 // account: the null check is mandatory and if it caused an
5965 // uncommon trap to be emitted then the allocation can't be
5966 // considered tightly coupled in this context.
5967 alloc = tightly_coupled_allocation(dest);
5968 }
5969
5970 bool validated = false;
5971
5972 const Type* src_type = _gvn.type(src);
5973 const Type* dest_type = _gvn.type(dest);
5974 const TypeAryPtr* top_src = src_type->isa_aryptr();
5975 const TypeAryPtr* top_dest = dest_type->isa_aryptr();
5976
5977 // Do we have the type of src?
5978 bool has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
5979 // Do we have the type of dest?
5980 bool has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
5981 // Is the type for src from speculation?
5982 bool src_spec = false;
5983 // Is the type for dest from speculation?
5984 bool dest_spec = false;
5985
5986 if ((!has_src || !has_dest) && can_emit_guards) {
5987 // We don't have sufficient type information, let's see if
5988 // speculative types can help. We need to have types for both src
5989 // and dest so that it pays off.
5990
5991 // Do we already have or could we have type information for src
5992 bool could_have_src = has_src;
5993 // Do we already have or could we have type information for dest
5994 bool could_have_dest = has_dest;
5995
5996 ciKlass* src_k = nullptr;
5997 if (!has_src) {
5998 src_k = src_type->speculative_type_not_null();
5999 if (src_k != nullptr && src_k->is_array_klass()) {
6000 could_have_src = true;
6001 }
6002 }
6003
6004 ciKlass* dest_k = nullptr;
6005 if (!has_dest) {
6006 dest_k = dest_type->speculative_type_not_null();
6007 if (dest_k != nullptr && dest_k->is_array_klass()) {
6008 could_have_dest = true;
6009 }
6010 }
6011
6012 if (could_have_src && could_have_dest) {
6013 // This is going to pay off so emit the required guards
6014 if (!has_src) {
6015 src = maybe_cast_profiled_obj(src, src_k, true);
6016 src_type = _gvn.type(src);
6017 top_src = src_type->isa_aryptr();
6018 has_src = (top_src != nullptr && top_src->elem() != Type::BOTTOM);
6019 src_spec = true;
6020 }
6021 if (!has_dest) {
6022 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6023 dest_type = _gvn.type(dest);
6024 top_dest = dest_type->isa_aryptr();
6025 has_dest = (top_dest != nullptr && top_dest->elem() != Type::BOTTOM);
6026 dest_spec = true;
6027 }
6028 }
6029 }
6030
6031 if (has_src && has_dest && can_emit_guards) {
6032 BasicType src_elem = top_src->isa_aryptr()->elem()->array_element_basic_type();
6033 BasicType dest_elem = top_dest->isa_aryptr()->elem()->array_element_basic_type();
6034 if (is_reference_type(src_elem, true)) src_elem = T_OBJECT;
6035 if (is_reference_type(dest_elem, true)) dest_elem = T_OBJECT;
6036
6037 if (src_elem == dest_elem && src_elem == T_OBJECT) {
6038 // If both arrays are object arrays then having the exact types
6039 // for both will remove the need for a subtype check at runtime
6040 // before the call and may make it possible to pick a faster copy
6041 // routine (without a subtype check on every element)
6042 // Do we have the exact type of src?
6043 bool could_have_src = src_spec;
6044 // Do we have the exact type of dest?
6045 bool could_have_dest = dest_spec;
6046 ciKlass* src_k = nullptr;
6047 ciKlass* dest_k = nullptr;
6048 if (!src_spec) {
6049 src_k = src_type->speculative_type_not_null();
6050 if (src_k != nullptr && src_k->is_array_klass()) {
6051 could_have_src = true;
6052 }
6053 }
6054 if (!dest_spec) {
6055 dest_k = dest_type->speculative_type_not_null();
6056 if (dest_k != nullptr && dest_k->is_array_klass()) {
6057 could_have_dest = true;
6058 }
6059 }
6060 if (could_have_src && could_have_dest) {
6061 // If we can have both exact types, emit the missing guards
6062 if (could_have_src && !src_spec) {
6063 src = maybe_cast_profiled_obj(src, src_k, true);
6064 }
6065 if (could_have_dest && !dest_spec) {
6066 dest = maybe_cast_profiled_obj(dest, dest_k, true);
6067 }
6068 }
6069 }
6070 }
6071
6072 ciMethod* trap_method = method();
6073 int trap_bci = bci();
6074 if (saved_jvms_before_guards != nullptr) {
6075 trap_method = alloc->jvms()->method();
6076 trap_bci = alloc->jvms()->bci();
6077 }
6078
6079 bool negative_length_guard_generated = false;
6080
6081 if (!C->too_many_traps(trap_method, trap_bci, Deoptimization::Reason_intrinsic) &&
6082 can_emit_guards &&
6083 !src->is_top() && !dest->is_top()) {
6084 // validate arguments: enables transformation the ArrayCopyNode
6085 validated = true;
6086
6087 RegionNode* slow_region = new RegionNode(1);
6088 record_for_igvn(slow_region);
6089
6090 // (1) src and dest are arrays.
6091 generate_non_array_guard(load_object_klass(src), slow_region, &src);
6092 generate_non_array_guard(load_object_klass(dest), slow_region, &dest);
6093
6094 // (2) src and dest arrays must have elements of the same BasicType
6095 // done at macro expansion or at Ideal transformation time
6096
6097 // (4) src_offset must not be negative.
6098 generate_negative_guard(src_offset, slow_region);
6099
6100 // (5) dest_offset must not be negative.
6101 generate_negative_guard(dest_offset, slow_region);
6102
6103 // (7) src_offset + length must not exceed length of src.
6104 generate_limit_guard(src_offset, length,
6105 load_array_length(src),
6106 slow_region);
6107
6108 // (8) dest_offset + length must not exceed length of dest.
6109 generate_limit_guard(dest_offset, length,
6110 load_array_length(dest),
6111 slow_region);
6112
6113 // (6) length must not be negative.
6114 // This is also checked in generate_arraycopy() during macro expansion, but
6115 // we also have to check it here for the case where the ArrayCopyNode will
6116 // be eliminated by Escape Analysis.
6117 if (EliminateAllocations) {
6118 generate_negative_guard(length, slow_region);
6119 negative_length_guard_generated = true;
6120 }
6121
6122 // (9) each element of an oop array must be assignable
6123 Node* dest_klass = load_object_klass(dest);
6124 if (src != dest) {
6125 Node* not_subtype_ctrl = gen_subtype_check(src, dest_klass);
6126
6127 if (not_subtype_ctrl != top()) {
6128 PreserveJVMState pjvms(this);
6129 set_control(not_subtype_ctrl);
6130 uncommon_trap(Deoptimization::Reason_intrinsic,
6131 Deoptimization::Action_make_not_entrant);
6132 assert(stopped(), "Should be stopped");
6133 }
6134 }
6135 {
6136 PreserveJVMState pjvms(this);
6137 set_control(_gvn.transform(slow_region));
6138 uncommon_trap(Deoptimization::Reason_intrinsic,
6139 Deoptimization::Action_make_not_entrant);
6140 assert(stopped(), "Should be stopped");
6141 }
6142
6143 const TypeKlassPtr* dest_klass_t = _gvn.type(dest_klass)->is_klassptr();
6144 const Type *toop = dest_klass_t->cast_to_exactness(false)->as_instance_type();
6145 src = _gvn.transform(new CheckCastPPNode(control(), src, toop));
6146 arraycopy_move_allocation_here(alloc, dest, saved_jvms_before_guards, saved_reexecute_sp, new_idx);
6147 }
6148
6149 if (stopped()) {
6150 return true;
6151 }
6152
6153 ArrayCopyNode* ac = ArrayCopyNode::make(this, true, src, src_offset, dest, dest_offset, length, alloc != nullptr, negative_length_guard_generated,
6154 // Create LoadRange and LoadKlass nodes for use during macro expansion here
6155 // so the compiler has a chance to eliminate them: during macro expansion,
6156 // we have to set their control (CastPP nodes are eliminated).
6157 load_object_klass(src), load_object_klass(dest),
6158 load_array_length(src), load_array_length(dest));
6159
6160 ac->set_arraycopy(validated);
6161
6162 Node* n = _gvn.transform(ac);
6163 if (n == ac) {
6164 ac->connect_outputs(this);
6165 } else {
6166 assert(validated, "shouldn't transform if all arguments not validated");
6167 set_all_memory(n);
6168 }
6169 clear_upper_avx();
6170
6171
6172 return true;
6173 }
6174
6175
6176 // Helper function which determines if an arraycopy immediately follows
6177 // an allocation, with no intervening tests or other escapes for the object.
6178 AllocateArrayNode*
6179 LibraryCallKit::tightly_coupled_allocation(Node* ptr) {
6180 if (stopped()) return nullptr; // no fast path
6181 if (!C->do_aliasing()) return nullptr; // no MergeMems around
6182
6183 AllocateArrayNode* alloc = AllocateArrayNode::Ideal_array_allocation(ptr);
6184 if (alloc == nullptr) return nullptr;
6185
6186 Node* rawmem = memory(Compile::AliasIdxRaw);
6187 // Is the allocation's memory state untouched?
6188 if (!(rawmem->is_Proj() && rawmem->in(0)->is_Initialize())) {
6189 // Bail out if there have been raw-memory effects since the allocation.
6190 // (Example: There might have been a call or safepoint.)
6191 return nullptr;
6192 }
6193 rawmem = rawmem->in(0)->as_Initialize()->memory(Compile::AliasIdxRaw);
6194 if (!(rawmem->is_Proj() && rawmem->in(0) == alloc)) {
6195 return nullptr;
6196 }
6197
6198 // There must be no unexpected observers of this allocation.
6199 for (DUIterator_Fast imax, i = ptr->fast_outs(imax); i < imax; i++) {
6200 Node* obs = ptr->fast_out(i);
6201 if (obs != this->map()) {
6202 return nullptr;
6203 }
6204 }
6205
6206 // This arraycopy must unconditionally follow the allocation of the ptr.
6207 Node* alloc_ctl = ptr->in(0);
6208 Node* ctl = control();
6209 while (ctl != alloc_ctl) {
6210 // There may be guards which feed into the slow_region.
6211 // Any other control flow means that we might not get a chance
6212 // to finish initializing the allocated object.
6213 // Various low-level checks bottom out in uncommon traps. These
6214 // are considered safe since we've already checked above that
6215 // there is no unexpected observer of this allocation.
6216 if (get_uncommon_trap_from_success_proj(ctl) != nullptr) {
6217 assert(ctl->in(0)->is_If(), "must be If");
6218 ctl = ctl->in(0)->in(0);
6219 } else {
6220 return nullptr;
6221 }
6222 }
6223
6224 // If we get this far, we have an allocation which immediately
6225 // precedes the arraycopy, and we can take over zeroing the new object.
6226 // The arraycopy will finish the initialization, and provide
6227 // a new control state to which we will anchor the destination pointer.
6228
6229 return alloc;
6230 }
6231
6232 CallStaticJavaNode* LibraryCallKit::get_uncommon_trap_from_success_proj(Node* node) {
6233 if (node->is_IfProj()) {
6234 Node* other_proj = node->as_IfProj()->other_if_proj();
6235 for (DUIterator_Fast jmax, j = other_proj->fast_outs(jmax); j < jmax; j++) {
6236 Node* obs = other_proj->fast_out(j);
6237 if (obs->in(0) == other_proj && obs->is_CallStaticJava() &&
6238 (obs->as_CallStaticJava()->entry_point() == OptoRuntime::uncommon_trap_blob()->entry_point())) {
6239 return obs->as_CallStaticJava();
6240 }
6241 }
6242 }
6243 return nullptr;
6244 }
6245
6246 //-------------inline_encodeISOArray-----------------------------------
6247 // encode char[] to byte[] in ISO_8859_1 or ASCII
6248 bool LibraryCallKit::inline_encodeISOArray(bool ascii) {
6249 assert(callee()->signature()->size() == 5, "encodeISOArray has 5 parameters");
6250 // no receiver since it is static method
6251 Node *src = argument(0);
6252 Node *src_offset = argument(1);
6253 Node *dst = argument(2);
6254 Node *dst_offset = argument(3);
6255 Node *length = argument(4);
6256
6257 src = must_be_not_null(src, true);
6258 dst = must_be_not_null(dst, true);
6259
6260 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6261 const TypeAryPtr* dst_type = dst->Value(&_gvn)->isa_aryptr();
6262 if (src_type == nullptr || src_type->elem() == Type::BOTTOM ||
6263 dst_type == nullptr || dst_type->elem() == Type::BOTTOM) {
6264 // failed array check
6265 return false;
6266 }
6267
6268 // Figure out the size and type of the elements we will be copying.
6269 BasicType src_elem = src_type->elem()->array_element_basic_type();
6270 BasicType dst_elem = dst_type->elem()->array_element_basic_type();
6271 if (!((src_elem == T_CHAR) || (src_elem== T_BYTE)) || dst_elem != T_BYTE) {
6272 return false;
6273 }
6274
6275 Node* src_start = array_element_address(src, src_offset, T_CHAR);
6276 Node* dst_start = array_element_address(dst, dst_offset, dst_elem);
6277 // 'src_start' points to src array + scaled offset
6278 // 'dst_start' points to dst array + scaled offset
6279
6280 const TypeAryPtr* mtype = TypeAryPtr::BYTES;
6281 Node* enc = new EncodeISOArrayNode(control(), memory(mtype), src_start, dst_start, length, ascii);
6282 enc = _gvn.transform(enc);
6283 Node* res_mem = _gvn.transform(new SCMemProjNode(enc));
6284 set_memory(res_mem, mtype);
6285 set_result(enc);
6286 clear_upper_avx();
6287
6288 return true;
6289 }
6290
6291 //-------------inline_multiplyToLen-----------------------------------
6292 bool LibraryCallKit::inline_multiplyToLen() {
6293 assert(UseMultiplyToLenIntrinsic, "not implemented on this platform");
6294
6295 address stubAddr = StubRoutines::multiplyToLen();
6296 if (stubAddr == nullptr) {
6297 return false; // Intrinsic's stub is not implemented on this platform
6298 }
6299 const char* stubName = "multiplyToLen";
6300
6301 assert(callee()->signature()->size() == 5, "multiplyToLen has 5 parameters");
6302
6303 // no receiver because it is a static method
6304 Node* x = argument(0);
6305 Node* xlen = argument(1);
6306 Node* y = argument(2);
6307 Node* ylen = argument(3);
6308 Node* z = argument(4);
6309
6310 x = must_be_not_null(x, true);
6311 y = must_be_not_null(y, true);
6312
6313 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6314 const TypeAryPtr* y_type = y->Value(&_gvn)->isa_aryptr();
6315 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6316 y_type == nullptr || y_type->elem() == Type::BOTTOM) {
6317 // failed array check
6318 return false;
6319 }
6320
6321 BasicType x_elem = x_type->elem()->array_element_basic_type();
6322 BasicType y_elem = y_type->elem()->array_element_basic_type();
6323 if (x_elem != T_INT || y_elem != T_INT) {
6324 return false;
6325 }
6326
6327 Node* x_start = array_element_address(x, intcon(0), x_elem);
6328 Node* y_start = array_element_address(y, intcon(0), y_elem);
6329 // 'x_start' points to x array + scaled xlen
6330 // 'y_start' points to y array + scaled ylen
6331
6332 Node* z_start = array_element_address(z, intcon(0), T_INT);
6333
6334 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6335 OptoRuntime::multiplyToLen_Type(),
6336 stubAddr, stubName, TypePtr::BOTTOM,
6337 x_start, xlen, y_start, ylen, z_start);
6338
6339 C->set_has_split_ifs(true); // Has chance for split-if optimization
6340 set_result(z);
6341 return true;
6342 }
6343
6344 //-------------inline_squareToLen------------------------------------
6345 bool LibraryCallKit::inline_squareToLen() {
6346 assert(UseSquareToLenIntrinsic, "not implemented on this platform");
6347
6348 address stubAddr = StubRoutines::squareToLen();
6349 if (stubAddr == nullptr) {
6350 return false; // Intrinsic's stub is not implemented on this platform
6351 }
6352 const char* stubName = "squareToLen";
6353
6354 assert(callee()->signature()->size() == 4, "implSquareToLen has 4 parameters");
6355
6356 Node* x = argument(0);
6357 Node* len = argument(1);
6358 Node* z = argument(2);
6359 Node* zlen = argument(3);
6360
6361 x = must_be_not_null(x, true);
6362 z = must_be_not_null(z, true);
6363
6364 const TypeAryPtr* x_type = x->Value(&_gvn)->isa_aryptr();
6365 const TypeAryPtr* z_type = z->Value(&_gvn)->isa_aryptr();
6366 if (x_type == nullptr || x_type->elem() == Type::BOTTOM ||
6367 z_type == nullptr || z_type->elem() == Type::BOTTOM) {
6368 // failed array check
6369 return false;
6370 }
6371
6372 BasicType x_elem = x_type->elem()->array_element_basic_type();
6373 BasicType z_elem = z_type->elem()->array_element_basic_type();
6374 if (x_elem != T_INT || z_elem != T_INT) {
6375 return false;
6376 }
6377
6378
6379 Node* x_start = array_element_address(x, intcon(0), x_elem);
6380 Node* z_start = array_element_address(z, intcon(0), z_elem);
6381
6382 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6383 OptoRuntime::squareToLen_Type(),
6384 stubAddr, stubName, TypePtr::BOTTOM,
6385 x_start, len, z_start, zlen);
6386
6387 set_result(z);
6388 return true;
6389 }
6390
6391 //-------------inline_mulAdd------------------------------------------
6392 bool LibraryCallKit::inline_mulAdd() {
6393 assert(UseMulAddIntrinsic, "not implemented on this platform");
6394
6395 address stubAddr = StubRoutines::mulAdd();
6396 if (stubAddr == nullptr) {
6397 return false; // Intrinsic's stub is not implemented on this platform
6398 }
6399 const char* stubName = "mulAdd";
6400
6401 assert(callee()->signature()->size() == 5, "mulAdd has 5 parameters");
6402
6403 Node* out = argument(0);
6404 Node* in = argument(1);
6405 Node* offset = argument(2);
6406 Node* len = argument(3);
6407 Node* k = argument(4);
6408
6409 in = must_be_not_null(in, true);
6410 out = must_be_not_null(out, true);
6411
6412 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
6413 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
6414 if (out_type == nullptr || out_type->elem() == Type::BOTTOM ||
6415 in_type == nullptr || in_type->elem() == Type::BOTTOM) {
6416 // failed array check
6417 return false;
6418 }
6419
6420 BasicType out_elem = out_type->elem()->array_element_basic_type();
6421 BasicType in_elem = in_type->elem()->array_element_basic_type();
6422 if (out_elem != T_INT || in_elem != T_INT) {
6423 return false;
6424 }
6425
6426 Node* outlen = load_array_length(out);
6427 Node* new_offset = _gvn.transform(new SubINode(outlen, offset));
6428 Node* out_start = array_element_address(out, intcon(0), out_elem);
6429 Node* in_start = array_element_address(in, intcon(0), in_elem);
6430
6431 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP,
6432 OptoRuntime::mulAdd_Type(),
6433 stubAddr, stubName, TypePtr::BOTTOM,
6434 out_start,in_start, new_offset, len, k);
6435 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6436 set_result(result);
6437 return true;
6438 }
6439
6440 //-------------inline_montgomeryMultiply-----------------------------------
6441 bool LibraryCallKit::inline_montgomeryMultiply() {
6442 address stubAddr = StubRoutines::montgomeryMultiply();
6443 if (stubAddr == nullptr) {
6444 return false; // Intrinsic's stub is not implemented on this platform
6445 }
6446
6447 assert(UseMontgomeryMultiplyIntrinsic, "not implemented on this platform");
6448 const char* stubName = "montgomery_multiply";
6449
6450 assert(callee()->signature()->size() == 7, "montgomeryMultiply has 7 parameters");
6451
6452 Node* a = argument(0);
6453 Node* b = argument(1);
6454 Node* n = argument(2);
6455 Node* len = argument(3);
6456 Node* inv = argument(4);
6457 Node* m = argument(6);
6458
6459 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6460 const TypeAryPtr* b_type = b->Value(&_gvn)->isa_aryptr();
6461 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6462 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6463 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6464 b_type == nullptr || b_type->elem() == Type::BOTTOM ||
6465 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6466 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6467 // failed array check
6468 return false;
6469 }
6470
6471 BasicType a_elem = a_type->elem()->array_element_basic_type();
6472 BasicType b_elem = b_type->elem()->array_element_basic_type();
6473 BasicType n_elem = n_type->elem()->array_element_basic_type();
6474 BasicType m_elem = m_type->elem()->array_element_basic_type();
6475 if (a_elem != T_INT || b_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6476 return false;
6477 }
6478
6479 // Make the call
6480 {
6481 Node* a_start = array_element_address(a, intcon(0), a_elem);
6482 Node* b_start = array_element_address(b, intcon(0), b_elem);
6483 Node* n_start = array_element_address(n, intcon(0), n_elem);
6484 Node* m_start = array_element_address(m, intcon(0), m_elem);
6485
6486 Node* call = make_runtime_call(RC_LEAF,
6487 OptoRuntime::montgomeryMultiply_Type(),
6488 stubAddr, stubName, TypePtr::BOTTOM,
6489 a_start, b_start, n_start, len, inv, top(),
6490 m_start);
6491 set_result(m);
6492 }
6493
6494 return true;
6495 }
6496
6497 bool LibraryCallKit::inline_montgomerySquare() {
6498 address stubAddr = StubRoutines::montgomerySquare();
6499 if (stubAddr == nullptr) {
6500 return false; // Intrinsic's stub is not implemented on this platform
6501 }
6502
6503 assert(UseMontgomerySquareIntrinsic, "not implemented on this platform");
6504 const char* stubName = "montgomery_square";
6505
6506 assert(callee()->signature()->size() == 6, "montgomerySquare has 6 parameters");
6507
6508 Node* a = argument(0);
6509 Node* n = argument(1);
6510 Node* len = argument(2);
6511 Node* inv = argument(3);
6512 Node* m = argument(5);
6513
6514 const TypeAryPtr* a_type = a->Value(&_gvn)->isa_aryptr();
6515 const TypeAryPtr* n_type = n->Value(&_gvn)->isa_aryptr();
6516 const TypeAryPtr* m_type = m->Value(&_gvn)->isa_aryptr();
6517 if (a_type == nullptr || a_type->elem() == Type::BOTTOM ||
6518 n_type == nullptr || n_type->elem() == Type::BOTTOM ||
6519 m_type == nullptr || m_type->elem() == Type::BOTTOM) {
6520 // failed array check
6521 return false;
6522 }
6523
6524 BasicType a_elem = a_type->elem()->array_element_basic_type();
6525 BasicType n_elem = n_type->elem()->array_element_basic_type();
6526 BasicType m_elem = m_type->elem()->array_element_basic_type();
6527 if (a_elem != T_INT || n_elem != T_INT || m_elem != T_INT) {
6528 return false;
6529 }
6530
6531 // Make the call
6532 {
6533 Node* a_start = array_element_address(a, intcon(0), a_elem);
6534 Node* n_start = array_element_address(n, intcon(0), n_elem);
6535 Node* m_start = array_element_address(m, intcon(0), m_elem);
6536
6537 Node* call = make_runtime_call(RC_LEAF,
6538 OptoRuntime::montgomerySquare_Type(),
6539 stubAddr, stubName, TypePtr::BOTTOM,
6540 a_start, n_start, len, inv, top(),
6541 m_start);
6542 set_result(m);
6543 }
6544
6545 return true;
6546 }
6547
6548 bool LibraryCallKit::inline_bigIntegerShift(bool isRightShift) {
6549 address stubAddr = nullptr;
6550 const char* stubName = nullptr;
6551
6552 stubAddr = isRightShift? StubRoutines::bigIntegerRightShift(): StubRoutines::bigIntegerLeftShift();
6553 if (stubAddr == nullptr) {
6554 return false; // Intrinsic's stub is not implemented on this platform
6555 }
6556
6557 stubName = isRightShift? "bigIntegerRightShiftWorker" : "bigIntegerLeftShiftWorker";
6558
6559 assert(callee()->signature()->size() == 5, "expected 5 arguments");
6560
6561 Node* newArr = argument(0);
6562 Node* oldArr = argument(1);
6563 Node* newIdx = argument(2);
6564 Node* shiftCount = argument(3);
6565 Node* numIter = argument(4);
6566
6567 const TypeAryPtr* newArr_type = newArr->Value(&_gvn)->isa_aryptr();
6568 const TypeAryPtr* oldArr_type = oldArr->Value(&_gvn)->isa_aryptr();
6569 if (newArr_type == nullptr || newArr_type->elem() == Type::BOTTOM ||
6570 oldArr_type == nullptr || oldArr_type->elem() == Type::BOTTOM) {
6571 return false;
6572 }
6573
6574 BasicType newArr_elem = newArr_type->elem()->array_element_basic_type();
6575 BasicType oldArr_elem = oldArr_type->elem()->array_element_basic_type();
6576 if (newArr_elem != T_INT || oldArr_elem != T_INT) {
6577 return false;
6578 }
6579
6580 // Make the call
6581 {
6582 Node* newArr_start = array_element_address(newArr, intcon(0), newArr_elem);
6583 Node* oldArr_start = array_element_address(oldArr, intcon(0), oldArr_elem);
6584
6585 Node* call = make_runtime_call(RC_LEAF,
6586 OptoRuntime::bigIntegerShift_Type(),
6587 stubAddr,
6588 stubName,
6589 TypePtr::BOTTOM,
6590 newArr_start,
6591 oldArr_start,
6592 newIdx,
6593 shiftCount,
6594 numIter);
6595 }
6596
6597 return true;
6598 }
6599
6600 //-------------inline_vectorizedMismatch------------------------------
6601 bool LibraryCallKit::inline_vectorizedMismatch() {
6602 assert(UseVectorizedMismatchIntrinsic, "not implemented on this platform");
6603
6604 assert(callee()->signature()->size() == 8, "vectorizedMismatch has 6 parameters");
6605 Node* obja = argument(0); // Object
6606 Node* aoffset = argument(1); // long
6607 Node* objb = argument(3); // Object
6608 Node* boffset = argument(4); // long
6609 Node* length = argument(6); // int
6610 Node* scale = argument(7); // int
6611
6612 const TypeAryPtr* obja_t = _gvn.type(obja)->isa_aryptr();
6613 const TypeAryPtr* objb_t = _gvn.type(objb)->isa_aryptr();
6614 if (obja_t == nullptr || obja_t->elem() == Type::BOTTOM ||
6615 objb_t == nullptr || objb_t->elem() == Type::BOTTOM ||
6616 scale == top()) {
6617 return false; // failed input validation
6618 }
6619
6620 Node* obja_adr = make_unsafe_address(obja, aoffset);
6621 Node* objb_adr = make_unsafe_address(objb, boffset);
6622
6623 // Partial inlining handling for inputs smaller than ArrayOperationPartialInlineSize bytes in size.
6624 //
6625 // inline_limit = ArrayOperationPartialInlineSize / element_size;
6626 // if (length <= inline_limit) {
6627 // inline_path:
6628 // vmask = VectorMaskGen length
6629 // vload1 = LoadVectorMasked obja, vmask
6630 // vload2 = LoadVectorMasked objb, vmask
6631 // result1 = VectorCmpMasked vload1, vload2, vmask
6632 // } else {
6633 // call_stub_path:
6634 // result2 = call vectorizedMismatch_stub(obja, objb, length, scale)
6635 // }
6636 // exit_block:
6637 // return Phi(result1, result2);
6638 //
6639 enum { inline_path = 1, // input is small enough to process it all at once
6640 stub_path = 2, // input is too large; call into the VM
6641 PATH_LIMIT = 3
6642 };
6643
6644 Node* exit_block = new RegionNode(PATH_LIMIT);
6645 Node* result_phi = new PhiNode(exit_block, TypeInt::INT);
6646 Node* memory_phi = new PhiNode(exit_block, Type::MEMORY, TypePtr::BOTTOM);
6647
6648 Node* call_stub_path = control();
6649
6650 BasicType elem_bt = T_ILLEGAL;
6651
6652 const TypeInt* scale_t = _gvn.type(scale)->is_int();
6653 if (scale_t->is_con()) {
6654 switch (scale_t->get_con()) {
6655 case 0: elem_bt = T_BYTE; break;
6656 case 1: elem_bt = T_SHORT; break;
6657 case 2: elem_bt = T_INT; break;
6658 case 3: elem_bt = T_LONG; break;
6659
6660 default: elem_bt = T_ILLEGAL; break; // not supported
6661 }
6662 }
6663
6664 int inline_limit = 0;
6665 bool do_partial_inline = false;
6666
6667 if (elem_bt != T_ILLEGAL && ArrayOperationPartialInlineSize > 0) {
6668 inline_limit = ArrayOperationPartialInlineSize / type2aelembytes(elem_bt);
6669 do_partial_inline = inline_limit >= 16;
6670 }
6671
6672 if (do_partial_inline) {
6673 assert(elem_bt != T_ILLEGAL, "sanity");
6674
6675 if (Matcher::match_rule_supported_vector(Op_VectorMaskGen, inline_limit, elem_bt) &&
6676 Matcher::match_rule_supported_vector(Op_LoadVectorMasked, inline_limit, elem_bt) &&
6677 Matcher::match_rule_supported_vector(Op_VectorCmpMasked, inline_limit, elem_bt)) {
6678
6679 const TypeVect* vt = TypeVect::make(elem_bt, inline_limit);
6680 Node* cmp_length = _gvn.transform(new CmpINode(length, intcon(inline_limit)));
6681 Node* bol_gt = _gvn.transform(new BoolNode(cmp_length, BoolTest::gt));
6682
6683 call_stub_path = generate_guard(bol_gt, nullptr, PROB_MIN);
6684
6685 if (!stopped()) {
6686 Node* casted_length = _gvn.transform(new CastIINode(control(), length, TypeInt::make(0, inline_limit, Type::WidenMin)));
6687
6688 const TypePtr* obja_adr_t = _gvn.type(obja_adr)->isa_ptr();
6689 const TypePtr* objb_adr_t = _gvn.type(objb_adr)->isa_ptr();
6690 Node* obja_adr_mem = memory(C->get_alias_index(obja_adr_t));
6691 Node* objb_adr_mem = memory(C->get_alias_index(objb_adr_t));
6692
6693 Node* vmask = _gvn.transform(VectorMaskGenNode::make(ConvI2X(casted_length), elem_bt));
6694 Node* vload_obja = _gvn.transform(new LoadVectorMaskedNode(control(), obja_adr_mem, obja_adr, obja_adr_t, vt, vmask));
6695 Node* vload_objb = _gvn.transform(new LoadVectorMaskedNode(control(), objb_adr_mem, objb_adr, objb_adr_t, vt, vmask));
6696 Node* result = _gvn.transform(new VectorCmpMaskedNode(vload_obja, vload_objb, vmask, TypeInt::INT));
6697
6698 exit_block->init_req(inline_path, control());
6699 memory_phi->init_req(inline_path, map()->memory());
6700 result_phi->init_req(inline_path, result);
6701
6702 C->set_max_vector_size(MAX2((uint)ArrayOperationPartialInlineSize, C->max_vector_size()));
6703 clear_upper_avx();
6704 }
6705 }
6706 }
6707
6708 if (call_stub_path != nullptr) {
6709 set_control(call_stub_path);
6710
6711 Node* call = make_runtime_call(RC_LEAF,
6712 OptoRuntime::vectorizedMismatch_Type(),
6713 StubRoutines::vectorizedMismatch(), "vectorizedMismatch", TypePtr::BOTTOM,
6714 obja_adr, objb_adr, length, scale);
6715
6716 exit_block->init_req(stub_path, control());
6717 memory_phi->init_req(stub_path, map()->memory());
6718 result_phi->init_req(stub_path, _gvn.transform(new ProjNode(call, TypeFunc::Parms)));
6719 }
6720
6721 exit_block = _gvn.transform(exit_block);
6722 memory_phi = _gvn.transform(memory_phi);
6723 result_phi = _gvn.transform(result_phi);
6724
6725 set_control(exit_block);
6726 set_all_memory(memory_phi);
6727 set_result(result_phi);
6728
6729 return true;
6730 }
6731
6732 //------------------------------inline_vectorizedHashcode----------------------------
6733 bool LibraryCallKit::inline_vectorizedHashCode() {
6734 assert(UseVectorizedHashCodeIntrinsic, "not implemented on this platform");
6735
6736 assert(callee()->signature()->size() == 5, "vectorizedHashCode has 5 parameters");
6737 Node* array = argument(0);
6738 Node* offset = argument(1);
6739 Node* length = argument(2);
6740 Node* initialValue = argument(3);
6741 Node* basic_type = argument(4);
6742
6743 if (basic_type == top()) {
6744 return false; // failed input validation
6745 }
6746
6747 const TypeInt* basic_type_t = _gvn.type(basic_type)->is_int();
6748 if (!basic_type_t->is_con()) {
6749 return false; // Only intrinsify if mode argument is constant
6750 }
6751
6752 array = must_be_not_null(array, true);
6753
6754 BasicType bt = (BasicType)basic_type_t->get_con();
6755
6756 // Resolve address of first element
6757 Node* array_start = array_element_address(array, offset, bt);
6758
6759 set_result(_gvn.transform(new VectorizedHashCodeNode(control(), memory(TypeAryPtr::get_array_body_type(bt)),
6760 array_start, length, initialValue, basic_type)));
6761 clear_upper_avx();
6762
6763 return true;
6764 }
6765
6766 /**
6767 * Calculate CRC32 for byte.
6768 * int java.util.zip.CRC32.update(int crc, int b)
6769 */
6770 bool LibraryCallKit::inline_updateCRC32() {
6771 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6772 assert(callee()->signature()->size() == 2, "update has 2 parameters");
6773 // no receiver since it is static method
6774 Node* crc = argument(0); // type: int
6775 Node* b = argument(1); // type: int
6776
6777 /*
6778 * int c = ~ crc;
6779 * b = timesXtoThe32[(b ^ c) & 0xFF];
6780 * b = b ^ (c >>> 8);
6781 * crc = ~b;
6782 */
6783
6784 Node* M1 = intcon(-1);
6785 crc = _gvn.transform(new XorINode(crc, M1));
6786 Node* result = _gvn.transform(new XorINode(crc, b));
6787 result = _gvn.transform(new AndINode(result, intcon(0xFF)));
6788
6789 Node* base = makecon(TypeRawPtr::make(StubRoutines::crc_table_addr()));
6790 Node* offset = _gvn.transform(new LShiftINode(result, intcon(0x2)));
6791 Node* adr = basic_plus_adr(top(), base, ConvI2X(offset));
6792 result = make_load(control(), adr, TypeInt::INT, T_INT, MemNode::unordered);
6793
6794 crc = _gvn.transform(new URShiftINode(crc, intcon(8)));
6795 result = _gvn.transform(new XorINode(crc, result));
6796 result = _gvn.transform(new XorINode(result, M1));
6797 set_result(result);
6798 return true;
6799 }
6800
6801 /**
6802 * Calculate CRC32 for byte[] array.
6803 * int java.util.zip.CRC32.updateBytes(int crc, byte[] buf, int off, int len)
6804 */
6805 bool LibraryCallKit::inline_updateBytesCRC32() {
6806 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6807 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6808 // no receiver since it is static method
6809 Node* crc = argument(0); // type: int
6810 Node* src = argument(1); // type: oop
6811 Node* offset = argument(2); // type: int
6812 Node* length = argument(3); // type: int
6813
6814 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6815 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6816 // failed array check
6817 return false;
6818 }
6819
6820 // Figure out the size and type of the elements we will be copying.
6821 BasicType src_elem = src_type->elem()->array_element_basic_type();
6822 if (src_elem != T_BYTE) {
6823 return false;
6824 }
6825
6826 // 'src_start' points to src array + scaled offset
6827 src = must_be_not_null(src, true);
6828 Node* src_start = array_element_address(src, offset, src_elem);
6829
6830 // We assume that range check is done by caller.
6831 // TODO: generate range check (offset+length < src.length) in debug VM.
6832
6833 // Call the stub.
6834 address stubAddr = StubRoutines::updateBytesCRC32();
6835 const char *stubName = "updateBytesCRC32";
6836
6837 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6838 stubAddr, stubName, TypePtr::BOTTOM,
6839 crc, src_start, length);
6840 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6841 set_result(result);
6842 return true;
6843 }
6844
6845 /**
6846 * Calculate CRC32 for ByteBuffer.
6847 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len)
6848 */
6849 bool LibraryCallKit::inline_updateByteBufferCRC32() {
6850 assert(UseCRC32Intrinsics, "need AVX and CLMUL instructions support");
6851 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
6852 // no receiver since it is static method
6853 Node* crc = argument(0); // type: int
6854 Node* src = argument(1); // type: long
6855 Node* offset = argument(3); // type: int
6856 Node* length = argument(4); // type: int
6857
6858 src = ConvL2X(src); // adjust Java long to machine word
6859 Node* base = _gvn.transform(new CastX2PNode(src));
6860 offset = ConvI2X(offset);
6861
6862 // 'src_start' points to src array + scaled offset
6863 Node* src_start = basic_plus_adr(top(), base, offset);
6864
6865 // Call the stub.
6866 address stubAddr = StubRoutines::updateBytesCRC32();
6867 const char *stubName = "updateBytesCRC32";
6868
6869 Node* call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::updateBytesCRC32_Type(),
6870 stubAddr, stubName, TypePtr::BOTTOM,
6871 crc, src_start, length);
6872 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6873 set_result(result);
6874 return true;
6875 }
6876
6877 //------------------------------get_table_from_crc32c_class-----------------------
6878 Node * LibraryCallKit::get_table_from_crc32c_class(ciInstanceKlass *crc32c_class) {
6879 Node* table = load_field_from_object(nullptr, "byteTable", "[I", /*decorators*/ IN_HEAP, /*is_static*/ true, crc32c_class);
6880 assert (table != nullptr, "wrong version of java.util.zip.CRC32C");
6881
6882 return table;
6883 }
6884
6885 //------------------------------inline_updateBytesCRC32C-----------------------
6886 //
6887 // Calculate CRC32C for byte[] array.
6888 // int java.util.zip.CRC32C.updateBytes(int crc, byte[] buf, int off, int end)
6889 //
6890 bool LibraryCallKit::inline_updateBytesCRC32C() {
6891 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6892 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6893 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6894 // no receiver since it is a static method
6895 Node* crc = argument(0); // type: int
6896 Node* src = argument(1); // type: oop
6897 Node* offset = argument(2); // type: int
6898 Node* end = argument(3); // type: int
6899
6900 Node* length = _gvn.transform(new SubINode(end, offset));
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 src = must_be_not_null(src, true);
6916 Node* src_start = array_element_address(src, offset, src_elem);
6917
6918 // static final int[] byteTable in class CRC32C
6919 Node* table = get_table_from_crc32c_class(callee()->holder());
6920 table = must_be_not_null(table, true);
6921 Node* table_start = array_element_address(table, intcon(0), T_INT);
6922
6923 // We assume that range check is done by caller.
6924 // TODO: generate range check (offset+length < src.length) in debug VM.
6925
6926 // Call the stub.
6927 address stubAddr = StubRoutines::updateBytesCRC32C();
6928 const char *stubName = "updateBytesCRC32C";
6929
6930 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6931 stubAddr, stubName, TypePtr::BOTTOM,
6932 crc, src_start, length, table_start);
6933 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6934 set_result(result);
6935 return true;
6936 }
6937
6938 //------------------------------inline_updateDirectByteBufferCRC32C-----------------------
6939 //
6940 // Calculate CRC32C for DirectByteBuffer.
6941 // int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long buf, int off, int end)
6942 //
6943 bool LibraryCallKit::inline_updateDirectByteBufferCRC32C() {
6944 assert(UseCRC32CIntrinsics, "need CRC32C instruction support");
6945 assert(callee()->signature()->size() == 5, "updateDirectByteBuffer has 4 parameters and one is long");
6946 assert(callee()->holder()->is_loaded(), "CRC32C class must be loaded");
6947 // no receiver since it is a static method
6948 Node* crc = argument(0); // type: int
6949 Node* src = argument(1); // type: long
6950 Node* offset = argument(3); // type: int
6951 Node* end = argument(4); // type: int
6952
6953 Node* length = _gvn.transform(new SubINode(end, offset));
6954
6955 src = ConvL2X(src); // adjust Java long to machine word
6956 Node* base = _gvn.transform(new CastX2PNode(src));
6957 offset = ConvI2X(offset);
6958
6959 // 'src_start' points to src array + scaled offset
6960 Node* src_start = basic_plus_adr(top(), base, offset);
6961
6962 // static final int[] byteTable in class CRC32C
6963 Node* table = get_table_from_crc32c_class(callee()->holder());
6964 table = must_be_not_null(table, true);
6965 Node* table_start = array_element_address(table, intcon(0), T_INT);
6966
6967 // Call the stub.
6968 address stubAddr = StubRoutines::updateBytesCRC32C();
6969 const char *stubName = "updateBytesCRC32C";
6970
6971 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesCRC32C_Type(),
6972 stubAddr, stubName, TypePtr::BOTTOM,
6973 crc, src_start, length, table_start);
6974 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
6975 set_result(result);
6976 return true;
6977 }
6978
6979 //------------------------------inline_updateBytesAdler32----------------------
6980 //
6981 // Calculate Adler32 checksum for byte[] array.
6982 // int java.util.zip.Adler32.updateBytes(int crc, byte[] buf, int off, int len)
6983 //
6984 bool LibraryCallKit::inline_updateBytesAdler32() {
6985 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
6986 assert(callee()->signature()->size() == 4, "updateBytes has 4 parameters");
6987 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
6988 // no receiver since it is static method
6989 Node* crc = argument(0); // type: int
6990 Node* src = argument(1); // type: oop
6991 Node* offset = argument(2); // type: int
6992 Node* length = argument(3); // type: int
6993
6994 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
6995 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
6996 // failed array check
6997 return false;
6998 }
6999
7000 // Figure out the size and type of the elements we will be copying.
7001 BasicType src_elem = src_type->elem()->array_element_basic_type();
7002 if (src_elem != T_BYTE) {
7003 return false;
7004 }
7005
7006 // 'src_start' points to src array + scaled offset
7007 Node* src_start = array_element_address(src, offset, src_elem);
7008
7009 // We assume that range check is done by caller.
7010 // TODO: generate range check (offset+length < src.length) in debug VM.
7011
7012 // Call the stub.
7013 address stubAddr = StubRoutines::updateBytesAdler32();
7014 const char *stubName = "updateBytesAdler32";
7015
7016 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7017 stubAddr, stubName, TypePtr::BOTTOM,
7018 crc, src_start, length);
7019 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7020 set_result(result);
7021 return true;
7022 }
7023
7024 //------------------------------inline_updateByteBufferAdler32---------------
7025 //
7026 // Calculate Adler32 checksum for DirectByteBuffer.
7027 // int java.util.zip.Adler32.updateByteBuffer(int crc, long buf, int off, int len)
7028 //
7029 bool LibraryCallKit::inline_updateByteBufferAdler32() {
7030 assert(UseAdler32Intrinsics, "Adler32 Intrinsic support need"); // check if we actually need to check this flag or check a different one
7031 assert(callee()->signature()->size() == 5, "updateByteBuffer has 4 parameters and one is long");
7032 assert(callee()->holder()->is_loaded(), "Adler32 class must be loaded");
7033 // no receiver since it is static method
7034 Node* crc = argument(0); // type: int
7035 Node* src = argument(1); // type: long
7036 Node* offset = argument(3); // type: int
7037 Node* length = argument(4); // type: int
7038
7039 src = ConvL2X(src); // adjust Java long to machine word
7040 Node* base = _gvn.transform(new CastX2PNode(src));
7041 offset = ConvI2X(offset);
7042
7043 // 'src_start' points to src array + scaled offset
7044 Node* src_start = basic_plus_adr(top(), base, offset);
7045
7046 // Call the stub.
7047 address stubAddr = StubRoutines::updateBytesAdler32();
7048 const char *stubName = "updateBytesAdler32";
7049
7050 Node* call = make_runtime_call(RC_LEAF, OptoRuntime::updateBytesAdler32_Type(),
7051 stubAddr, stubName, TypePtr::BOTTOM,
7052 crc, src_start, length);
7053
7054 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
7055 set_result(result);
7056 return true;
7057 }
7058
7059 //----------------------------inline_reference_get----------------------------
7060 // public T java.lang.ref.Reference.get();
7061 bool LibraryCallKit::inline_reference_get() {
7062 const int referent_offset = java_lang_ref_Reference::referent_offset();
7063
7064 // Get the argument:
7065 Node* reference_obj = null_check_receiver();
7066 if (stopped()) return true;
7067
7068 DecoratorSet decorators = IN_HEAP | ON_WEAK_OOP_REF;
7069 Node* result = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7070 decorators, /*is_static*/ false, nullptr);
7071 if (result == nullptr) return false;
7072
7073 // Add memory barrier to prevent commoning reads from this field
7074 // across safepoint since GC can change its value.
7075 insert_mem_bar(Op_MemBarCPUOrder);
7076
7077 set_result(result);
7078 return true;
7079 }
7080
7081 //----------------------------inline_reference_refersTo0----------------------------
7082 // bool java.lang.ref.Reference.refersTo0();
7083 // bool java.lang.ref.PhantomReference.refersTo0();
7084 bool LibraryCallKit::inline_reference_refersTo0(bool is_phantom) {
7085 // Get arguments:
7086 Node* reference_obj = null_check_receiver();
7087 Node* other_obj = argument(1);
7088 if (stopped()) return true;
7089
7090 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7091 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7092 Node* referent = load_field_from_object(reference_obj, "referent", "Ljava/lang/Object;",
7093 decorators, /*is_static*/ false, nullptr);
7094 if (referent == nullptr) return false;
7095
7096 // Add memory barrier to prevent commoning reads from this field
7097 // across safepoint since GC can change its value.
7098 insert_mem_bar(Op_MemBarCPUOrder);
7099
7100 Node* cmp = _gvn.transform(new CmpPNode(referent, other_obj));
7101 Node* bol = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
7102 IfNode* if_node = create_and_map_if(control(), bol, PROB_FAIR, COUNT_UNKNOWN);
7103
7104 RegionNode* region = new RegionNode(3);
7105 PhiNode* phi = new PhiNode(region, TypeInt::BOOL);
7106
7107 Node* if_true = _gvn.transform(new IfTrueNode(if_node));
7108 region->init_req(1, if_true);
7109 phi->init_req(1, intcon(1));
7110
7111 Node* if_false = _gvn.transform(new IfFalseNode(if_node));
7112 region->init_req(2, if_false);
7113 phi->init_req(2, intcon(0));
7114
7115 set_control(_gvn.transform(region));
7116 record_for_igvn(region);
7117 set_result(_gvn.transform(phi));
7118 return true;
7119 }
7120
7121 //----------------------------inline_reference_clear0----------------------------
7122 // void java.lang.ref.Reference.clear0();
7123 // void java.lang.ref.PhantomReference.clear0();
7124 bool LibraryCallKit::inline_reference_clear0(bool is_phantom) {
7125 // This matches the implementation in JVM_ReferenceClear, see the comments there.
7126
7127 // Get arguments
7128 Node* reference_obj = null_check_receiver();
7129 if (stopped()) return true;
7130
7131 // Common access parameters
7132 DecoratorSet decorators = IN_HEAP | AS_NO_KEEPALIVE;
7133 decorators |= (is_phantom ? ON_PHANTOM_OOP_REF : ON_WEAK_OOP_REF);
7134 Node* referent_field_addr = basic_plus_adr(reference_obj, java_lang_ref_Reference::referent_offset());
7135 const TypePtr* referent_field_addr_type = _gvn.type(referent_field_addr)->isa_ptr();
7136 const Type* val_type = TypeOopPtr::make_from_klass(env()->Object_klass());
7137
7138 Node* referent = access_load_at(reference_obj,
7139 referent_field_addr,
7140 referent_field_addr_type,
7141 val_type,
7142 T_OBJECT,
7143 decorators);
7144
7145 IdealKit ideal(this);
7146 #define __ ideal.
7147 __ if_then(referent, BoolTest::ne, null());
7148 sync_kit(ideal);
7149 access_store_at(reference_obj,
7150 referent_field_addr,
7151 referent_field_addr_type,
7152 null(),
7153 val_type,
7154 T_OBJECT,
7155 decorators);
7156 __ sync_kit(this);
7157 __ end_if();
7158 final_sync(ideal);
7159 #undef __
7160
7161 return true;
7162 }
7163
7164 Node* LibraryCallKit::load_field_from_object(Node* fromObj, const char* fieldName, const char* fieldTypeString,
7165 DecoratorSet decorators, bool is_static,
7166 ciInstanceKlass* fromKls) {
7167 if (fromKls == nullptr) {
7168 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7169 assert(tinst != nullptr, "obj is null");
7170 assert(tinst->is_loaded(), "obj is not loaded");
7171 fromKls = tinst->instance_klass();
7172 } else {
7173 assert(is_static, "only for static field access");
7174 }
7175 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7176 ciSymbol::make(fieldTypeString),
7177 is_static);
7178
7179 assert(field != nullptr, "undefined field %s %s %s", fieldTypeString, fromKls->name()->as_utf8(), fieldName);
7180 if (field == nullptr) return (Node *) nullptr;
7181
7182 if (is_static) {
7183 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7184 fromObj = makecon(tip);
7185 }
7186
7187 // Next code copied from Parse::do_get_xxx():
7188
7189 // Compute address and memory type.
7190 int offset = field->offset_in_bytes();
7191 bool is_vol = field->is_volatile();
7192 ciType* field_klass = field->type();
7193 assert(field_klass->is_loaded(), "should be loaded");
7194 const TypePtr* adr_type = C->alias_type(field)->adr_type();
7195 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7196 assert(C->get_alias_index(adr_type) == C->get_alias_index(_gvn.type(adr)->isa_ptr()),
7197 "slice of address and input slice don't match");
7198 BasicType bt = field->layout_type();
7199
7200 // Build the resultant type of the load
7201 const Type *type;
7202 if (bt == T_OBJECT) {
7203 type = TypeOopPtr::make_from_klass(field_klass->as_klass());
7204 } else {
7205 type = Type::get_const_basic_type(bt);
7206 }
7207
7208 if (is_vol) {
7209 decorators |= MO_SEQ_CST;
7210 }
7211
7212 return access_load_at(fromObj, adr, adr_type, type, bt, decorators);
7213 }
7214
7215 Node * LibraryCallKit::field_address_from_object(Node * fromObj, const char * fieldName, const char * fieldTypeString,
7216 bool is_exact /* true */, bool is_static /* false */,
7217 ciInstanceKlass * fromKls /* nullptr */) {
7218 if (fromKls == nullptr) {
7219 const TypeInstPtr* tinst = _gvn.type(fromObj)->isa_instptr();
7220 assert(tinst != nullptr, "obj is null");
7221 assert(tinst->is_loaded(), "obj is not loaded");
7222 assert(!is_exact || tinst->klass_is_exact(), "klass not exact");
7223 fromKls = tinst->instance_klass();
7224 }
7225 else {
7226 assert(is_static, "only for static field access");
7227 }
7228 ciField* field = fromKls->get_field_by_name(ciSymbol::make(fieldName),
7229 ciSymbol::make(fieldTypeString),
7230 is_static);
7231
7232 assert(field != nullptr, "undefined field");
7233 assert(!field->is_volatile(), "not defined for volatile fields");
7234
7235 if (is_static) {
7236 const TypeInstPtr* tip = TypeInstPtr::make(fromKls->java_mirror());
7237 fromObj = makecon(tip);
7238 }
7239
7240 // Next code copied from Parse::do_get_xxx():
7241
7242 // Compute address and memory type.
7243 int offset = field->offset_in_bytes();
7244 Node *adr = basic_plus_adr(fromObj, fromObj, offset);
7245
7246 return adr;
7247 }
7248
7249 //------------------------------inline_aescrypt_Block-----------------------
7250 bool LibraryCallKit::inline_aescrypt_Block(vmIntrinsics::ID id) {
7251 address stubAddr = nullptr;
7252 const char *stubName;
7253 assert(UseAES, "need AES instruction support");
7254
7255 switch(id) {
7256 case vmIntrinsics::_aescrypt_encryptBlock:
7257 stubAddr = StubRoutines::aescrypt_encryptBlock();
7258 stubName = "aescrypt_encryptBlock";
7259 break;
7260 case vmIntrinsics::_aescrypt_decryptBlock:
7261 stubAddr = StubRoutines::aescrypt_decryptBlock();
7262 stubName = "aescrypt_decryptBlock";
7263 break;
7264 default:
7265 break;
7266 }
7267 if (stubAddr == nullptr) return false;
7268
7269 Node* aescrypt_object = argument(0);
7270 Node* src = argument(1);
7271 Node* src_offset = argument(2);
7272 Node* dest = argument(3);
7273 Node* dest_offset = argument(4);
7274
7275 src = must_be_not_null(src, true);
7276 dest = must_be_not_null(dest, true);
7277
7278 // (1) src and dest are arrays.
7279 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7280 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7281 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7282 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7283
7284 // for the quick and dirty code we will skip all the checks.
7285 // we are just trying to get the call to be generated.
7286 Node* src_start = src;
7287 Node* dest_start = dest;
7288 if (src_offset != nullptr || dest_offset != nullptr) {
7289 assert(src_offset != nullptr && dest_offset != nullptr, "");
7290 src_start = array_element_address(src, src_offset, T_BYTE);
7291 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7292 }
7293
7294 // now need to get the start of its expanded key array
7295 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7296 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7297 if (k_start == nullptr) return false;
7298
7299 // Call the stub.
7300 make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::aescrypt_block_Type(),
7301 stubAddr, stubName, TypePtr::BOTTOM,
7302 src_start, dest_start, k_start);
7303
7304 return true;
7305 }
7306
7307 //------------------------------inline_cipherBlockChaining_AESCrypt-----------------------
7308 bool LibraryCallKit::inline_cipherBlockChaining_AESCrypt(vmIntrinsics::ID id) {
7309 address stubAddr = nullptr;
7310 const char *stubName = nullptr;
7311
7312 assert(UseAES, "need AES instruction support");
7313
7314 switch(id) {
7315 case vmIntrinsics::_cipherBlockChaining_encryptAESCrypt:
7316 stubAddr = StubRoutines::cipherBlockChaining_encryptAESCrypt();
7317 stubName = "cipherBlockChaining_encryptAESCrypt";
7318 break;
7319 case vmIntrinsics::_cipherBlockChaining_decryptAESCrypt:
7320 stubAddr = StubRoutines::cipherBlockChaining_decryptAESCrypt();
7321 stubName = "cipherBlockChaining_decryptAESCrypt";
7322 break;
7323 default:
7324 break;
7325 }
7326 if (stubAddr == nullptr) return false;
7327
7328 Node* cipherBlockChaining_object = argument(0);
7329 Node* src = argument(1);
7330 Node* src_offset = argument(2);
7331 Node* len = argument(3);
7332 Node* dest = argument(4);
7333 Node* dest_offset = argument(5);
7334
7335 src = must_be_not_null(src, false);
7336 dest = must_be_not_null(dest, false);
7337
7338 // (1) src and dest are arrays.
7339 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7340 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7341 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7342 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7343
7344 // checks are the responsibility of the caller
7345 Node* src_start = src;
7346 Node* dest_start = dest;
7347 if (src_offset != nullptr || dest_offset != nullptr) {
7348 assert(src_offset != nullptr && dest_offset != nullptr, "");
7349 src_start = array_element_address(src, src_offset, T_BYTE);
7350 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7351 }
7352
7353 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7354 // (because of the predicated logic executed earlier).
7355 // so we cast it here safely.
7356 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7357
7358 Node* embeddedCipherObj = load_field_from_object(cipherBlockChaining_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7359 if (embeddedCipherObj == nullptr) return false;
7360
7361 // cast it to what we know it will be at runtime
7362 const TypeInstPtr* tinst = _gvn.type(cipherBlockChaining_object)->isa_instptr();
7363 assert(tinst != nullptr, "CBC obj is null");
7364 assert(tinst->is_loaded(), "CBC obj is not loaded");
7365 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7366 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7367
7368 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7369 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7370 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7371 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7372 aescrypt_object = _gvn.transform(aescrypt_object);
7373
7374 // we need to get the start of the aescrypt_object's expanded key array
7375 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7376 if (k_start == nullptr) return false;
7377
7378 // similarly, get the start address of the r vector
7379 Node* objRvec = load_field_from_object(cipherBlockChaining_object, "r", "[B");
7380 if (objRvec == nullptr) return false;
7381 Node* r_start = array_element_address(objRvec, intcon(0), T_BYTE);
7382
7383 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7384 Node* cbcCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7385 OptoRuntime::cipherBlockChaining_aescrypt_Type(),
7386 stubAddr, stubName, TypePtr::BOTTOM,
7387 src_start, dest_start, k_start, r_start, len);
7388
7389 // return cipher length (int)
7390 Node* retvalue = _gvn.transform(new ProjNode(cbcCrypt, TypeFunc::Parms));
7391 set_result(retvalue);
7392 return true;
7393 }
7394
7395 //------------------------------inline_electronicCodeBook_AESCrypt-----------------------
7396 bool LibraryCallKit::inline_electronicCodeBook_AESCrypt(vmIntrinsics::ID id) {
7397 address stubAddr = nullptr;
7398 const char *stubName = nullptr;
7399
7400 assert(UseAES, "need AES instruction support");
7401
7402 switch (id) {
7403 case vmIntrinsics::_electronicCodeBook_encryptAESCrypt:
7404 stubAddr = StubRoutines::electronicCodeBook_encryptAESCrypt();
7405 stubName = "electronicCodeBook_encryptAESCrypt";
7406 break;
7407 case vmIntrinsics::_electronicCodeBook_decryptAESCrypt:
7408 stubAddr = StubRoutines::electronicCodeBook_decryptAESCrypt();
7409 stubName = "electronicCodeBook_decryptAESCrypt";
7410 break;
7411 default:
7412 break;
7413 }
7414
7415 if (stubAddr == nullptr) return false;
7416
7417 Node* electronicCodeBook_object = argument(0);
7418 Node* src = argument(1);
7419 Node* src_offset = argument(2);
7420 Node* len = argument(3);
7421 Node* dest = argument(4);
7422 Node* dest_offset = argument(5);
7423
7424 // (1) src and dest are arrays.
7425 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7426 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7427 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7428 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7429
7430 // checks are the responsibility of the caller
7431 Node* src_start = src;
7432 Node* dest_start = dest;
7433 if (src_offset != nullptr || dest_offset != nullptr) {
7434 assert(src_offset != nullptr && dest_offset != nullptr, "");
7435 src_start = array_element_address(src, src_offset, T_BYTE);
7436 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7437 }
7438
7439 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7440 // (because of the predicated logic executed earlier).
7441 // so we cast it here safely.
7442 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7443
7444 Node* embeddedCipherObj = load_field_from_object(electronicCodeBook_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7445 if (embeddedCipherObj == nullptr) return false;
7446
7447 // cast it to what we know it will be at runtime
7448 const TypeInstPtr* tinst = _gvn.type(electronicCodeBook_object)->isa_instptr();
7449 assert(tinst != nullptr, "ECB obj is null");
7450 assert(tinst->is_loaded(), "ECB obj is not loaded");
7451 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7452 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7453
7454 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7455 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7456 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7457 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7458 aescrypt_object = _gvn.transform(aescrypt_object);
7459
7460 // we need to get the start of the aescrypt_object's expanded key array
7461 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7462 if (k_start == nullptr) return false;
7463
7464 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7465 Node* ecbCrypt = make_runtime_call(RC_LEAF | RC_NO_FP,
7466 OptoRuntime::electronicCodeBook_aescrypt_Type(),
7467 stubAddr, stubName, TypePtr::BOTTOM,
7468 src_start, dest_start, k_start, len);
7469
7470 // return cipher length (int)
7471 Node* retvalue = _gvn.transform(new ProjNode(ecbCrypt, TypeFunc::Parms));
7472 set_result(retvalue);
7473 return true;
7474 }
7475
7476 //------------------------------inline_counterMode_AESCrypt-----------------------
7477 bool LibraryCallKit::inline_counterMode_AESCrypt(vmIntrinsics::ID id) {
7478 assert(UseAES, "need AES instruction support");
7479 if (!UseAESCTRIntrinsics) return false;
7480
7481 address stubAddr = nullptr;
7482 const char *stubName = nullptr;
7483 if (id == vmIntrinsics::_counterMode_AESCrypt) {
7484 stubAddr = StubRoutines::counterMode_AESCrypt();
7485 stubName = "counterMode_AESCrypt";
7486 }
7487 if (stubAddr == nullptr) return false;
7488
7489 Node* counterMode_object = argument(0);
7490 Node* src = argument(1);
7491 Node* src_offset = argument(2);
7492 Node* len = argument(3);
7493 Node* dest = argument(4);
7494 Node* dest_offset = argument(5);
7495
7496 // (1) src and dest are arrays.
7497 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
7498 const TypeAryPtr* dest_type = dest->Value(&_gvn)->isa_aryptr();
7499 assert( src_type != nullptr && src_type->elem() != Type::BOTTOM &&
7500 dest_type != nullptr && dest_type->elem() != Type::BOTTOM, "args are strange");
7501
7502 // checks are the responsibility of the caller
7503 Node* src_start = src;
7504 Node* dest_start = dest;
7505 if (src_offset != nullptr || dest_offset != nullptr) {
7506 assert(src_offset != nullptr && dest_offset != nullptr, "");
7507 src_start = array_element_address(src, src_offset, T_BYTE);
7508 dest_start = array_element_address(dest, dest_offset, T_BYTE);
7509 }
7510
7511 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
7512 // (because of the predicated logic executed earlier).
7513 // so we cast it here safely.
7514 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
7515 Node* embeddedCipherObj = load_field_from_object(counterMode_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7516 if (embeddedCipherObj == nullptr) return false;
7517 // cast it to what we know it will be at runtime
7518 const TypeInstPtr* tinst = _gvn.type(counterMode_object)->isa_instptr();
7519 assert(tinst != nullptr, "CTR obj is null");
7520 assert(tinst->is_loaded(), "CTR obj is not loaded");
7521 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7522 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
7523 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7524 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
7525 const TypeOopPtr* xtype = aklass->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
7526 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
7527 aescrypt_object = _gvn.transform(aescrypt_object);
7528 // we need to get the start of the aescrypt_object's expanded key array
7529 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
7530 if (k_start == nullptr) return false;
7531 // similarly, get the start address of the r vector
7532 Node* obj_counter = load_field_from_object(counterMode_object, "counter", "[B");
7533 if (obj_counter == nullptr) return false;
7534 Node* cnt_start = array_element_address(obj_counter, intcon(0), T_BYTE);
7535
7536 Node* saved_encCounter = load_field_from_object(counterMode_object, "encryptedCounter", "[B");
7537 if (saved_encCounter == nullptr) return false;
7538 Node* saved_encCounter_start = array_element_address(saved_encCounter, intcon(0), T_BYTE);
7539 Node* used = field_address_from_object(counterMode_object, "used", "I", /*is_exact*/ false);
7540
7541 // Call the stub, passing src_start, dest_start, k_start, r_start and src_len
7542 Node* ctrCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
7543 OptoRuntime::counterMode_aescrypt_Type(),
7544 stubAddr, stubName, TypePtr::BOTTOM,
7545 src_start, dest_start, k_start, cnt_start, len, saved_encCounter_start, used);
7546
7547 // return cipher length (int)
7548 Node* retvalue = _gvn.transform(new ProjNode(ctrCrypt, TypeFunc::Parms));
7549 set_result(retvalue);
7550 return true;
7551 }
7552
7553 //------------------------------get_key_start_from_aescrypt_object-----------------------
7554 Node * LibraryCallKit::get_key_start_from_aescrypt_object(Node *aescrypt_object) {
7555 #if defined(PPC64) || defined(S390) || defined(RISCV64)
7556 // MixColumns for decryption can be reduced by preprocessing MixColumns with round keys.
7557 // Intel's extension is based on this optimization and AESCrypt generates round keys by preprocessing MixColumns.
7558 // However, ppc64 vncipher processes MixColumns and requires the same round keys with encryption.
7559 // The ppc64 and riscv64 stubs of encryption and decryption use the same round keys (sessionK[0]).
7560 Node* objSessionK = load_field_from_object(aescrypt_object, "sessionK", "[[I");
7561 assert (objSessionK != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7562 if (objSessionK == nullptr) {
7563 return (Node *) nullptr;
7564 }
7565 Node* objAESCryptKey = load_array_element(objSessionK, intcon(0), TypeAryPtr::OOPS, /* set_ctrl */ true);
7566 #else
7567 Node* objAESCryptKey = load_field_from_object(aescrypt_object, "K", "[I");
7568 #endif // PPC64
7569 assert (objAESCryptKey != nullptr, "wrong version of com.sun.crypto.provider.AESCrypt");
7570 if (objAESCryptKey == nullptr) return (Node *) nullptr;
7571
7572 // now have the array, need to get the start address of the K array
7573 Node* k_start = array_element_address(objAESCryptKey, intcon(0), T_INT);
7574 return k_start;
7575 }
7576
7577 //----------------------------inline_cipherBlockChaining_AESCrypt_predicate----------------------------
7578 // Return node representing slow path of predicate check.
7579 // the pseudo code we want to emulate with this predicate is:
7580 // for encryption:
7581 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7582 // for decryption:
7583 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7584 // note cipher==plain is more conservative than the original java code but that's OK
7585 //
7586 Node* LibraryCallKit::inline_cipherBlockChaining_AESCrypt_predicate(bool decrypting) {
7587 // The receiver was checked for null already.
7588 Node* objCBC = argument(0);
7589
7590 Node* src = argument(1);
7591 Node* dest = argument(4);
7592
7593 // Load embeddedCipher field of CipherBlockChaining object.
7594 Node* embeddedCipherObj = load_field_from_object(objCBC, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7595
7596 // get AESCrypt klass for instanceOf check
7597 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7598 // will have same classloader as CipherBlockChaining object
7599 const TypeInstPtr* tinst = _gvn.type(objCBC)->isa_instptr();
7600 assert(tinst != nullptr, "CBCobj is null");
7601 assert(tinst->is_loaded(), "CBCobj is not loaded");
7602
7603 // we want to do an instanceof comparison against the AESCrypt class
7604 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7605 if (!klass_AESCrypt->is_loaded()) {
7606 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7607 Node* ctrl = control();
7608 set_control(top()); // no regular fast path
7609 return ctrl;
7610 }
7611
7612 src = must_be_not_null(src, true);
7613 dest = must_be_not_null(dest, true);
7614
7615 // Resolve oops to stable for CmpP below.
7616 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7617
7618 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7619 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7620 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7621
7622 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7623
7624 // for encryption, we are done
7625 if (!decrypting)
7626 return instof_false; // even if it is null
7627
7628 // for decryption, we need to add a further check to avoid
7629 // taking the intrinsic path when cipher and plain are the same
7630 // see the original java code for why.
7631 RegionNode* region = new RegionNode(3);
7632 region->init_req(1, instof_false);
7633
7634 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7635 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7636 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7637 region->init_req(2, src_dest_conjoint);
7638
7639 record_for_igvn(region);
7640 return _gvn.transform(region);
7641 }
7642
7643 //----------------------------inline_electronicCodeBook_AESCrypt_predicate----------------------------
7644 // Return node representing slow path of predicate check.
7645 // the pseudo code we want to emulate with this predicate is:
7646 // for encryption:
7647 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7648 // for decryption:
7649 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7650 // note cipher==plain is more conservative than the original java code but that's OK
7651 //
7652 Node* LibraryCallKit::inline_electronicCodeBook_AESCrypt_predicate(bool decrypting) {
7653 // The receiver was checked for null already.
7654 Node* objECB = argument(0);
7655
7656 // Load embeddedCipher field of ElectronicCodeBook object.
7657 Node* embeddedCipherObj = load_field_from_object(objECB, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7658
7659 // get AESCrypt klass for instanceOf check
7660 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7661 // will have same classloader as ElectronicCodeBook object
7662 const TypeInstPtr* tinst = _gvn.type(objECB)->isa_instptr();
7663 assert(tinst != nullptr, "ECBobj is null");
7664 assert(tinst->is_loaded(), "ECBobj is not loaded");
7665
7666 // we want to do an instanceof comparison against the AESCrypt class
7667 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7668 if (!klass_AESCrypt->is_loaded()) {
7669 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7670 Node* ctrl = control();
7671 set_control(top()); // no regular fast path
7672 return ctrl;
7673 }
7674 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7675
7676 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7677 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7678 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7679
7680 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7681
7682 // for encryption, we are done
7683 if (!decrypting)
7684 return instof_false; // even if it is null
7685
7686 // for decryption, we need to add a further check to avoid
7687 // taking the intrinsic path when cipher and plain are the same
7688 // see the original java code for why.
7689 RegionNode* region = new RegionNode(3);
7690 region->init_req(1, instof_false);
7691 Node* src = argument(1);
7692 Node* dest = argument(4);
7693 Node* cmp_src_dest = _gvn.transform(new CmpPNode(src, dest));
7694 Node* bool_src_dest = _gvn.transform(new BoolNode(cmp_src_dest, BoolTest::eq));
7695 Node* src_dest_conjoint = generate_guard(bool_src_dest, nullptr, PROB_MIN);
7696 region->init_req(2, src_dest_conjoint);
7697
7698 record_for_igvn(region);
7699 return _gvn.transform(region);
7700 }
7701
7702 //----------------------------inline_counterMode_AESCrypt_predicate----------------------------
7703 // Return node representing slow path of predicate check.
7704 // the pseudo code we want to emulate with this predicate is:
7705 // for encryption:
7706 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
7707 // for decryption:
7708 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
7709 // note cipher==plain is more conservative than the original java code but that's OK
7710 //
7711
7712 Node* LibraryCallKit::inline_counterMode_AESCrypt_predicate() {
7713 // The receiver was checked for null already.
7714 Node* objCTR = argument(0);
7715
7716 // Load embeddedCipher field of CipherBlockChaining object.
7717 Node* embeddedCipherObj = load_field_from_object(objCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
7718
7719 // get AESCrypt klass for instanceOf check
7720 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
7721 // will have same classloader as CipherBlockChaining object
7722 const TypeInstPtr* tinst = _gvn.type(objCTR)->isa_instptr();
7723 assert(tinst != nullptr, "CTRobj is null");
7724 assert(tinst->is_loaded(), "CTRobj is not loaded");
7725
7726 // we want to do an instanceof comparison against the AESCrypt class
7727 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
7728 if (!klass_AESCrypt->is_loaded()) {
7729 // if AESCrypt is not even loaded, we never take the intrinsic fast path
7730 Node* ctrl = control();
7731 set_control(top()); // no regular fast path
7732 return ctrl;
7733 }
7734
7735 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
7736 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
7737 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
7738 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
7739 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
7740
7741 return instof_false; // even if it is null
7742 }
7743
7744 //------------------------------inline_ghash_processBlocks
7745 bool LibraryCallKit::inline_ghash_processBlocks() {
7746 address stubAddr;
7747 const char *stubName;
7748 assert(UseGHASHIntrinsics, "need GHASH intrinsics support");
7749
7750 stubAddr = StubRoutines::ghash_processBlocks();
7751 stubName = "ghash_processBlocks";
7752
7753 Node* data = argument(0);
7754 Node* offset = argument(1);
7755 Node* len = argument(2);
7756 Node* state = argument(3);
7757 Node* subkeyH = argument(4);
7758
7759 state = must_be_not_null(state, true);
7760 subkeyH = must_be_not_null(subkeyH, true);
7761 data = must_be_not_null(data, true);
7762
7763 Node* state_start = array_element_address(state, intcon(0), T_LONG);
7764 assert(state_start, "state is null");
7765 Node* subkeyH_start = array_element_address(subkeyH, intcon(0), T_LONG);
7766 assert(subkeyH_start, "subkeyH is null");
7767 Node* data_start = array_element_address(data, offset, T_BYTE);
7768 assert(data_start, "data is null");
7769
7770 Node* ghash = make_runtime_call(RC_LEAF|RC_NO_FP,
7771 OptoRuntime::ghash_processBlocks_Type(),
7772 stubAddr, stubName, TypePtr::BOTTOM,
7773 state_start, subkeyH_start, data_start, len);
7774 return true;
7775 }
7776
7777 //------------------------------inline_chacha20Block
7778 bool LibraryCallKit::inline_chacha20Block() {
7779 address stubAddr;
7780 const char *stubName;
7781 assert(UseChaCha20Intrinsics, "need ChaCha20 intrinsics support");
7782
7783 stubAddr = StubRoutines::chacha20Block();
7784 stubName = "chacha20Block";
7785
7786 Node* state = argument(0);
7787 Node* result = argument(1);
7788
7789 state = must_be_not_null(state, true);
7790 result = must_be_not_null(result, true);
7791
7792 Node* state_start = array_element_address(state, intcon(0), T_INT);
7793 assert(state_start, "state is null");
7794 Node* result_start = array_element_address(result, intcon(0), T_BYTE);
7795 assert(result_start, "result is null");
7796
7797 Node* cc20Blk = make_runtime_call(RC_LEAF|RC_NO_FP,
7798 OptoRuntime::chacha20Block_Type(),
7799 stubAddr, stubName, TypePtr::BOTTOM,
7800 state_start, result_start);
7801 // return key stream length (int)
7802 Node* retvalue = _gvn.transform(new ProjNode(cc20Blk, TypeFunc::Parms));
7803 set_result(retvalue);
7804 return true;
7805 }
7806
7807 //------------------------------inline_kyberNtt
7808 bool LibraryCallKit::inline_kyberNtt() {
7809 address stubAddr;
7810 const char *stubName;
7811 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7812 assert(callee()->signature()->size() == 2, "kyberNtt has 2 parameters");
7813
7814 stubAddr = StubRoutines::kyberNtt();
7815 stubName = "kyberNtt";
7816 if (!stubAddr) return false;
7817
7818 Node* coeffs = argument(0);
7819 Node* ntt_zetas = argument(1);
7820
7821 coeffs = must_be_not_null(coeffs, true);
7822 ntt_zetas = must_be_not_null(ntt_zetas, true);
7823
7824 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7825 assert(coeffs_start, "coeffs is null");
7826 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_SHORT);
7827 assert(ntt_zetas_start, "ntt_zetas is null");
7828 Node* kyberNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7829 OptoRuntime::kyberNtt_Type(),
7830 stubAddr, stubName, TypePtr::BOTTOM,
7831 coeffs_start, ntt_zetas_start);
7832 // return an int
7833 Node* retvalue = _gvn.transform(new ProjNode(kyberNtt, TypeFunc::Parms));
7834 set_result(retvalue);
7835 return true;
7836 }
7837
7838 //------------------------------inline_kyberInverseNtt
7839 bool LibraryCallKit::inline_kyberInverseNtt() {
7840 address stubAddr;
7841 const char *stubName;
7842 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7843 assert(callee()->signature()->size() == 2, "kyberInverseNtt has 2 parameters");
7844
7845 stubAddr = StubRoutines::kyberInverseNtt();
7846 stubName = "kyberInverseNtt";
7847 if (!stubAddr) return false;
7848
7849 Node* coeffs = argument(0);
7850 Node* zetas = argument(1);
7851
7852 coeffs = must_be_not_null(coeffs, true);
7853 zetas = must_be_not_null(zetas, true);
7854
7855 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
7856 assert(coeffs_start, "coeffs is null");
7857 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7858 assert(zetas_start, "inverseNtt_zetas is null");
7859 Node* kyberInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
7860 OptoRuntime::kyberInverseNtt_Type(),
7861 stubAddr, stubName, TypePtr::BOTTOM,
7862 coeffs_start, zetas_start);
7863
7864 // return an int
7865 Node* retvalue = _gvn.transform(new ProjNode(kyberInverseNtt, TypeFunc::Parms));
7866 set_result(retvalue);
7867 return true;
7868 }
7869
7870 //------------------------------inline_kyberNttMult
7871 bool LibraryCallKit::inline_kyberNttMult() {
7872 address stubAddr;
7873 const char *stubName;
7874 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7875 assert(callee()->signature()->size() == 4, "kyberNttMult has 4 parameters");
7876
7877 stubAddr = StubRoutines::kyberNttMult();
7878 stubName = "kyberNttMult";
7879 if (!stubAddr) return false;
7880
7881 Node* result = argument(0);
7882 Node* ntta = argument(1);
7883 Node* nttb = argument(2);
7884 Node* zetas = argument(3);
7885
7886 result = must_be_not_null(result, true);
7887 ntta = must_be_not_null(ntta, true);
7888 nttb = must_be_not_null(nttb, true);
7889 zetas = must_be_not_null(zetas, true);
7890 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7891 assert(result_start, "result is null");
7892 Node* ntta_start = array_element_address(ntta, intcon(0), T_SHORT);
7893 assert(ntta_start, "ntta is null");
7894 Node* nttb_start = array_element_address(nttb, intcon(0), T_SHORT);
7895 assert(nttb_start, "nttb is null");
7896 Node* zetas_start = array_element_address(zetas, intcon(0), T_SHORT);
7897 assert(zetas_start, "nttMult_zetas is null");
7898 Node* kyberNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
7899 OptoRuntime::kyberNttMult_Type(),
7900 stubAddr, stubName, TypePtr::BOTTOM,
7901 result_start, ntta_start, nttb_start,
7902 zetas_start);
7903
7904 // return an int
7905 Node* retvalue = _gvn.transform(new ProjNode(kyberNttMult, TypeFunc::Parms));
7906 set_result(retvalue);
7907
7908 return true;
7909 }
7910
7911 //------------------------------inline_kyberAddPoly_2
7912 bool LibraryCallKit::inline_kyberAddPoly_2() {
7913 address stubAddr;
7914 const char *stubName;
7915 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7916 assert(callee()->signature()->size() == 3, "kyberAddPoly_2 has 3 parameters");
7917
7918 stubAddr = StubRoutines::kyberAddPoly_2();
7919 stubName = "kyberAddPoly_2";
7920 if (!stubAddr) return false;
7921
7922 Node* result = argument(0);
7923 Node* a = argument(1);
7924 Node* b = argument(2);
7925
7926 result = must_be_not_null(result, true);
7927 a = must_be_not_null(a, true);
7928 b = must_be_not_null(b, true);
7929
7930 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7931 assert(result_start, "result is null");
7932 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7933 assert(a_start, "a is null");
7934 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7935 assert(b_start, "b is null");
7936 Node* kyberAddPoly_2 = make_runtime_call(RC_LEAF|RC_NO_FP,
7937 OptoRuntime::kyberAddPoly_2_Type(),
7938 stubAddr, stubName, TypePtr::BOTTOM,
7939 result_start, a_start, b_start);
7940 // return an int
7941 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_2, TypeFunc::Parms));
7942 set_result(retvalue);
7943 return true;
7944 }
7945
7946 //------------------------------inline_kyberAddPoly_3
7947 bool LibraryCallKit::inline_kyberAddPoly_3() {
7948 address stubAddr;
7949 const char *stubName;
7950 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7951 assert(callee()->signature()->size() == 4, "kyberAddPoly_3 has 4 parameters");
7952
7953 stubAddr = StubRoutines::kyberAddPoly_3();
7954 stubName = "kyberAddPoly_3";
7955 if (!stubAddr) return false;
7956
7957 Node* result = argument(0);
7958 Node* a = argument(1);
7959 Node* b = argument(2);
7960 Node* c = argument(3);
7961
7962 result = must_be_not_null(result, true);
7963 a = must_be_not_null(a, true);
7964 b = must_be_not_null(b, true);
7965 c = must_be_not_null(c, true);
7966
7967 Node* result_start = array_element_address(result, intcon(0), T_SHORT);
7968 assert(result_start, "result is null");
7969 Node* a_start = array_element_address(a, intcon(0), T_SHORT);
7970 assert(a_start, "a is null");
7971 Node* b_start = array_element_address(b, intcon(0), T_SHORT);
7972 assert(b_start, "b is null");
7973 Node* c_start = array_element_address(c, intcon(0), T_SHORT);
7974 assert(c_start, "c is null");
7975 Node* kyberAddPoly_3 = make_runtime_call(RC_LEAF|RC_NO_FP,
7976 OptoRuntime::kyberAddPoly_3_Type(),
7977 stubAddr, stubName, TypePtr::BOTTOM,
7978 result_start, a_start, b_start, c_start);
7979 // return an int
7980 Node* retvalue = _gvn.transform(new ProjNode(kyberAddPoly_3, TypeFunc::Parms));
7981 set_result(retvalue);
7982 return true;
7983 }
7984
7985 //------------------------------inline_kyber12To16
7986 bool LibraryCallKit::inline_kyber12To16() {
7987 address stubAddr;
7988 const char *stubName;
7989 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
7990 assert(callee()->signature()->size() == 4, "kyber12To16 has 4 parameters");
7991
7992 stubAddr = StubRoutines::kyber12To16();
7993 stubName = "kyber12To16";
7994 if (!stubAddr) return false;
7995
7996 Node* condensed = argument(0);
7997 Node* condensedOffs = argument(1);
7998 Node* parsed = argument(2);
7999 Node* parsedLength = argument(3);
8000
8001 condensed = must_be_not_null(condensed, true);
8002 parsed = must_be_not_null(parsed, true);
8003
8004 Node* condensed_start = array_element_address(condensed, intcon(0), T_BYTE);
8005 assert(condensed_start, "condensed is null");
8006 Node* parsed_start = array_element_address(parsed, intcon(0), T_SHORT);
8007 assert(parsed_start, "parsed is null");
8008 Node* kyber12To16 = make_runtime_call(RC_LEAF|RC_NO_FP,
8009 OptoRuntime::kyber12To16_Type(),
8010 stubAddr, stubName, TypePtr::BOTTOM,
8011 condensed_start, condensedOffs, parsed_start, parsedLength);
8012 // return an int
8013 Node* retvalue = _gvn.transform(new ProjNode(kyber12To16, TypeFunc::Parms));
8014 set_result(retvalue);
8015 return true;
8016
8017 }
8018
8019 //------------------------------inline_kyberBarrettReduce
8020 bool LibraryCallKit::inline_kyberBarrettReduce() {
8021 address stubAddr;
8022 const char *stubName;
8023 assert(UseKyberIntrinsics, "need Kyber intrinsics support");
8024 assert(callee()->signature()->size() == 1, "kyberBarrettReduce has 1 parameters");
8025
8026 stubAddr = StubRoutines::kyberBarrettReduce();
8027 stubName = "kyberBarrettReduce";
8028 if (!stubAddr) return false;
8029
8030 Node* coeffs = argument(0);
8031
8032 coeffs = must_be_not_null(coeffs, true);
8033
8034 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_SHORT);
8035 assert(coeffs_start, "coeffs is null");
8036 Node* kyberBarrettReduce = make_runtime_call(RC_LEAF|RC_NO_FP,
8037 OptoRuntime::kyberBarrettReduce_Type(),
8038 stubAddr, stubName, TypePtr::BOTTOM,
8039 coeffs_start);
8040 // return an int
8041 Node* retvalue = _gvn.transform(new ProjNode(kyberBarrettReduce, TypeFunc::Parms));
8042 set_result(retvalue);
8043 return true;
8044 }
8045
8046 //------------------------------inline_dilithiumAlmostNtt
8047 bool LibraryCallKit::inline_dilithiumAlmostNtt() {
8048 address stubAddr;
8049 const char *stubName;
8050 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8051 assert(callee()->signature()->size() == 2, "dilithiumAlmostNtt has 2 parameters");
8052
8053 stubAddr = StubRoutines::dilithiumAlmostNtt();
8054 stubName = "dilithiumAlmostNtt";
8055 if (!stubAddr) return false;
8056
8057 Node* coeffs = argument(0);
8058 Node* ntt_zetas = argument(1);
8059
8060 coeffs = must_be_not_null(coeffs, true);
8061 ntt_zetas = must_be_not_null(ntt_zetas, true);
8062
8063 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8064 assert(coeffs_start, "coeffs is null");
8065 Node* ntt_zetas_start = array_element_address(ntt_zetas, intcon(0), T_INT);
8066 assert(ntt_zetas_start, "ntt_zetas is null");
8067 Node* dilithiumAlmostNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8068 OptoRuntime::dilithiumAlmostNtt_Type(),
8069 stubAddr, stubName, TypePtr::BOTTOM,
8070 coeffs_start, ntt_zetas_start);
8071 // return an int
8072 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostNtt, TypeFunc::Parms));
8073 set_result(retvalue);
8074 return true;
8075 }
8076
8077 //------------------------------inline_dilithiumAlmostInverseNtt
8078 bool LibraryCallKit::inline_dilithiumAlmostInverseNtt() {
8079 address stubAddr;
8080 const char *stubName;
8081 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8082 assert(callee()->signature()->size() == 2, "dilithiumAlmostInverseNtt has 2 parameters");
8083
8084 stubAddr = StubRoutines::dilithiumAlmostInverseNtt();
8085 stubName = "dilithiumAlmostInverseNtt";
8086 if (!stubAddr) return false;
8087
8088 Node* coeffs = argument(0);
8089 Node* zetas = argument(1);
8090
8091 coeffs = must_be_not_null(coeffs, true);
8092 zetas = must_be_not_null(zetas, true);
8093
8094 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8095 assert(coeffs_start, "coeffs is null");
8096 Node* zetas_start = array_element_address(zetas, intcon(0), T_INT);
8097 assert(zetas_start, "inverseNtt_zetas is null");
8098 Node* dilithiumAlmostInverseNtt = make_runtime_call(RC_LEAF|RC_NO_FP,
8099 OptoRuntime::dilithiumAlmostInverseNtt_Type(),
8100 stubAddr, stubName, TypePtr::BOTTOM,
8101 coeffs_start, zetas_start);
8102 // return an int
8103 Node* retvalue = _gvn.transform(new ProjNode(dilithiumAlmostInverseNtt, TypeFunc::Parms));
8104 set_result(retvalue);
8105 return true;
8106 }
8107
8108 //------------------------------inline_dilithiumNttMult
8109 bool LibraryCallKit::inline_dilithiumNttMult() {
8110 address stubAddr;
8111 const char *stubName;
8112 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8113 assert(callee()->signature()->size() == 3, "dilithiumNttMult has 3 parameters");
8114
8115 stubAddr = StubRoutines::dilithiumNttMult();
8116 stubName = "dilithiumNttMult";
8117 if (!stubAddr) return false;
8118
8119 Node* result = argument(0);
8120 Node* ntta = argument(1);
8121 Node* nttb = argument(2);
8122 Node* zetas = argument(3);
8123
8124 result = must_be_not_null(result, true);
8125 ntta = must_be_not_null(ntta, true);
8126 nttb = must_be_not_null(nttb, true);
8127 zetas = must_be_not_null(zetas, true);
8128
8129 Node* result_start = array_element_address(result, intcon(0), T_INT);
8130 assert(result_start, "result is null");
8131 Node* ntta_start = array_element_address(ntta, intcon(0), T_INT);
8132 assert(ntta_start, "ntta is null");
8133 Node* nttb_start = array_element_address(nttb, intcon(0), T_INT);
8134 assert(nttb_start, "nttb is null");
8135 Node* dilithiumNttMult = make_runtime_call(RC_LEAF|RC_NO_FP,
8136 OptoRuntime::dilithiumNttMult_Type(),
8137 stubAddr, stubName, TypePtr::BOTTOM,
8138 result_start, ntta_start, nttb_start);
8139
8140 // return an int
8141 Node* retvalue = _gvn.transform(new ProjNode(dilithiumNttMult, TypeFunc::Parms));
8142 set_result(retvalue);
8143
8144 return true;
8145 }
8146
8147 //------------------------------inline_dilithiumMontMulByConstant
8148 bool LibraryCallKit::inline_dilithiumMontMulByConstant() {
8149 address stubAddr;
8150 const char *stubName;
8151 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8152 assert(callee()->signature()->size() == 2, "dilithiumMontMulByConstant has 2 parameters");
8153
8154 stubAddr = StubRoutines::dilithiumMontMulByConstant();
8155 stubName = "dilithiumMontMulByConstant";
8156 if (!stubAddr) return false;
8157
8158 Node* coeffs = argument(0);
8159 Node* constant = argument(1);
8160
8161 coeffs = must_be_not_null(coeffs, true);
8162
8163 Node* coeffs_start = array_element_address(coeffs, intcon(0), T_INT);
8164 assert(coeffs_start, "coeffs is null");
8165 Node* dilithiumMontMulByConstant = make_runtime_call(RC_LEAF|RC_NO_FP,
8166 OptoRuntime::dilithiumMontMulByConstant_Type(),
8167 stubAddr, stubName, TypePtr::BOTTOM,
8168 coeffs_start, constant);
8169
8170 // return an int
8171 Node* retvalue = _gvn.transform(new ProjNode(dilithiumMontMulByConstant, TypeFunc::Parms));
8172 set_result(retvalue);
8173 return true;
8174 }
8175
8176
8177 //------------------------------inline_dilithiumDecomposePoly
8178 bool LibraryCallKit::inline_dilithiumDecomposePoly() {
8179 address stubAddr;
8180 const char *stubName;
8181 assert(UseDilithiumIntrinsics, "need Dilithium intrinsics support");
8182 assert(callee()->signature()->size() == 5, "dilithiumDecomposePoly has 5 parameters");
8183
8184 stubAddr = StubRoutines::dilithiumDecomposePoly();
8185 stubName = "dilithiumDecomposePoly";
8186 if (!stubAddr) return false;
8187
8188 Node* input = argument(0);
8189 Node* lowPart = argument(1);
8190 Node* highPart = argument(2);
8191 Node* twoGamma2 = argument(3);
8192 Node* multiplier = argument(4);
8193
8194 input = must_be_not_null(input, true);
8195 lowPart = must_be_not_null(lowPart, true);
8196 highPart = must_be_not_null(highPart, true);
8197
8198 Node* input_start = array_element_address(input, intcon(0), T_INT);
8199 assert(input_start, "input is null");
8200 Node* lowPart_start = array_element_address(lowPart, intcon(0), T_INT);
8201 assert(lowPart_start, "lowPart is null");
8202 Node* highPart_start = array_element_address(highPart, intcon(0), T_INT);
8203 assert(highPart_start, "highPart is null");
8204
8205 Node* dilithiumDecomposePoly = make_runtime_call(RC_LEAF|RC_NO_FP,
8206 OptoRuntime::dilithiumDecomposePoly_Type(),
8207 stubAddr, stubName, TypePtr::BOTTOM,
8208 input_start, lowPart_start, highPart_start,
8209 twoGamma2, multiplier);
8210
8211 // return an int
8212 Node* retvalue = _gvn.transform(new ProjNode(dilithiumDecomposePoly, TypeFunc::Parms));
8213 set_result(retvalue);
8214 return true;
8215 }
8216
8217 bool LibraryCallKit::inline_base64_encodeBlock() {
8218 address stubAddr;
8219 const char *stubName;
8220 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8221 assert(callee()->signature()->size() == 6, "base64_encodeBlock has 6 parameters");
8222 stubAddr = StubRoutines::base64_encodeBlock();
8223 stubName = "encodeBlock";
8224
8225 if (!stubAddr) return false;
8226 Node* base64obj = argument(0);
8227 Node* src = argument(1);
8228 Node* offset = argument(2);
8229 Node* len = argument(3);
8230 Node* dest = argument(4);
8231 Node* dp = argument(5);
8232 Node* isURL = argument(6);
8233
8234 src = must_be_not_null(src, true);
8235 dest = must_be_not_null(dest, true);
8236
8237 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8238 assert(src_start, "source array is null");
8239 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8240 assert(dest_start, "destination array is null");
8241
8242 Node* base64 = make_runtime_call(RC_LEAF,
8243 OptoRuntime::base64_encodeBlock_Type(),
8244 stubAddr, stubName, TypePtr::BOTTOM,
8245 src_start, offset, len, dest_start, dp, isURL);
8246 return true;
8247 }
8248
8249 bool LibraryCallKit::inline_base64_decodeBlock() {
8250 address stubAddr;
8251 const char *stubName;
8252 assert(UseBASE64Intrinsics, "need Base64 intrinsics support");
8253 assert(callee()->signature()->size() == 7, "base64_decodeBlock has 7 parameters");
8254 stubAddr = StubRoutines::base64_decodeBlock();
8255 stubName = "decodeBlock";
8256
8257 if (!stubAddr) return false;
8258 Node* base64obj = argument(0);
8259 Node* src = argument(1);
8260 Node* src_offset = argument(2);
8261 Node* len = argument(3);
8262 Node* dest = argument(4);
8263 Node* dest_offset = argument(5);
8264 Node* isURL = argument(6);
8265 Node* isMIME = argument(7);
8266
8267 src = must_be_not_null(src, true);
8268 dest = must_be_not_null(dest, true);
8269
8270 Node* src_start = array_element_address(src, intcon(0), T_BYTE);
8271 assert(src_start, "source array is null");
8272 Node* dest_start = array_element_address(dest, intcon(0), T_BYTE);
8273 assert(dest_start, "destination array is null");
8274
8275 Node* call = make_runtime_call(RC_LEAF,
8276 OptoRuntime::base64_decodeBlock_Type(),
8277 stubAddr, stubName, TypePtr::BOTTOM,
8278 src_start, src_offset, len, dest_start, dest_offset, isURL, isMIME);
8279 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8280 set_result(result);
8281 return true;
8282 }
8283
8284 bool LibraryCallKit::inline_poly1305_processBlocks() {
8285 address stubAddr;
8286 const char *stubName;
8287 assert(UsePoly1305Intrinsics, "need Poly intrinsics support");
8288 assert(callee()->signature()->size() == 5, "poly1305_processBlocks has %d parameters", callee()->signature()->size());
8289 stubAddr = StubRoutines::poly1305_processBlocks();
8290 stubName = "poly1305_processBlocks";
8291
8292 if (!stubAddr) return false;
8293 null_check_receiver(); // null-check receiver
8294 if (stopped()) return true;
8295
8296 Node* input = argument(1);
8297 Node* input_offset = argument(2);
8298 Node* len = argument(3);
8299 Node* alimbs = argument(4);
8300 Node* rlimbs = argument(5);
8301
8302 input = must_be_not_null(input, true);
8303 alimbs = must_be_not_null(alimbs, true);
8304 rlimbs = must_be_not_null(rlimbs, true);
8305
8306 Node* input_start = array_element_address(input, input_offset, T_BYTE);
8307 assert(input_start, "input array is null");
8308 Node* acc_start = array_element_address(alimbs, intcon(0), T_LONG);
8309 assert(acc_start, "acc array is null");
8310 Node* r_start = array_element_address(rlimbs, intcon(0), T_LONG);
8311 assert(r_start, "r array is null");
8312
8313 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8314 OptoRuntime::poly1305_processBlocks_Type(),
8315 stubAddr, stubName, TypePtr::BOTTOM,
8316 input_start, len, acc_start, r_start);
8317 return true;
8318 }
8319
8320 bool LibraryCallKit::inline_intpoly_montgomeryMult_P256() {
8321 address stubAddr;
8322 const char *stubName;
8323 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8324 assert(callee()->signature()->size() == 3, "intpoly_montgomeryMult_P256 has %d parameters", callee()->signature()->size());
8325 stubAddr = StubRoutines::intpoly_montgomeryMult_P256();
8326 stubName = "intpoly_montgomeryMult_P256";
8327
8328 if (!stubAddr) return false;
8329 null_check_receiver(); // null-check receiver
8330 if (stopped()) return true;
8331
8332 Node* a = argument(1);
8333 Node* b = argument(2);
8334 Node* r = argument(3);
8335
8336 a = must_be_not_null(a, true);
8337 b = must_be_not_null(b, true);
8338 r = must_be_not_null(r, true);
8339
8340 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8341 assert(a_start, "a array is null");
8342 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8343 assert(b_start, "b array is null");
8344 Node* r_start = array_element_address(r, intcon(0), T_LONG);
8345 assert(r_start, "r array is null");
8346
8347 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8348 OptoRuntime::intpoly_montgomeryMult_P256_Type(),
8349 stubAddr, stubName, TypePtr::BOTTOM,
8350 a_start, b_start, r_start);
8351 return true;
8352 }
8353
8354 bool LibraryCallKit::inline_intpoly_assign() {
8355 assert(UseIntPolyIntrinsics, "need intpoly intrinsics support");
8356 assert(callee()->signature()->size() == 3, "intpoly_assign has %d parameters", callee()->signature()->size());
8357 const char *stubName = "intpoly_assign";
8358 address stubAddr = StubRoutines::intpoly_assign();
8359 if (!stubAddr) return false;
8360
8361 Node* set = argument(0);
8362 Node* a = argument(1);
8363 Node* b = argument(2);
8364 Node* arr_length = load_array_length(a);
8365
8366 a = must_be_not_null(a, true);
8367 b = must_be_not_null(b, true);
8368
8369 Node* a_start = array_element_address(a, intcon(0), T_LONG);
8370 assert(a_start, "a array is null");
8371 Node* b_start = array_element_address(b, intcon(0), T_LONG);
8372 assert(b_start, "b array is null");
8373
8374 Node* call = make_runtime_call(RC_LEAF | RC_NO_FP,
8375 OptoRuntime::intpoly_assign_Type(),
8376 stubAddr, stubName, TypePtr::BOTTOM,
8377 set, a_start, b_start, arr_length);
8378 return true;
8379 }
8380
8381 //------------------------------inline_digestBase_implCompress-----------------------
8382 //
8383 // Calculate MD5 for single-block byte[] array.
8384 // void com.sun.security.provider.MD5.implCompress(byte[] buf, int ofs)
8385 //
8386 // Calculate SHA (i.e., SHA-1) for single-block byte[] array.
8387 // void com.sun.security.provider.SHA.implCompress(byte[] buf, int ofs)
8388 //
8389 // Calculate SHA2 (i.e., SHA-244 or SHA-256) for single-block byte[] array.
8390 // void com.sun.security.provider.SHA2.implCompress(byte[] buf, int ofs)
8391 //
8392 // Calculate SHA5 (i.e., SHA-384 or SHA-512) for single-block byte[] array.
8393 // void com.sun.security.provider.SHA5.implCompress(byte[] buf, int ofs)
8394 //
8395 // Calculate SHA3 (i.e., SHA3-224 or SHA3-256 or SHA3-384 or SHA3-512) for single-block byte[] array.
8396 // void com.sun.security.provider.SHA3.implCompress(byte[] buf, int ofs)
8397 //
8398 bool LibraryCallKit::inline_digestBase_implCompress(vmIntrinsics::ID id) {
8399 assert(callee()->signature()->size() == 2, "sha_implCompress has 2 parameters");
8400
8401 Node* digestBase_obj = argument(0);
8402 Node* src = argument(1); // type oop
8403 Node* ofs = argument(2); // type int
8404
8405 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8406 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8407 // failed array check
8408 return false;
8409 }
8410 // Figure out the size and type of the elements we will be copying.
8411 BasicType src_elem = src_type->elem()->array_element_basic_type();
8412 if (src_elem != T_BYTE) {
8413 return false;
8414 }
8415 // 'src_start' points to src array + offset
8416 src = must_be_not_null(src, true);
8417 Node* src_start = array_element_address(src, ofs, src_elem);
8418 Node* state = nullptr;
8419 Node* block_size = nullptr;
8420 address stubAddr;
8421 const char *stubName;
8422
8423 switch(id) {
8424 case vmIntrinsics::_md5_implCompress:
8425 assert(UseMD5Intrinsics, "need MD5 instruction support");
8426 state = get_state_from_digest_object(digestBase_obj, T_INT);
8427 stubAddr = StubRoutines::md5_implCompress();
8428 stubName = "md5_implCompress";
8429 break;
8430 case vmIntrinsics::_sha_implCompress:
8431 assert(UseSHA1Intrinsics, "need SHA1 instruction support");
8432 state = get_state_from_digest_object(digestBase_obj, T_INT);
8433 stubAddr = StubRoutines::sha1_implCompress();
8434 stubName = "sha1_implCompress";
8435 break;
8436 case vmIntrinsics::_sha2_implCompress:
8437 assert(UseSHA256Intrinsics, "need SHA256 instruction support");
8438 state = get_state_from_digest_object(digestBase_obj, T_INT);
8439 stubAddr = StubRoutines::sha256_implCompress();
8440 stubName = "sha256_implCompress";
8441 break;
8442 case vmIntrinsics::_sha5_implCompress:
8443 assert(UseSHA512Intrinsics, "need SHA512 instruction support");
8444 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8445 stubAddr = StubRoutines::sha512_implCompress();
8446 stubName = "sha512_implCompress";
8447 break;
8448 case vmIntrinsics::_sha3_implCompress:
8449 assert(UseSHA3Intrinsics, "need SHA3 instruction support");
8450 state = get_state_from_digest_object(digestBase_obj, T_LONG);
8451 stubAddr = StubRoutines::sha3_implCompress();
8452 stubName = "sha3_implCompress";
8453 block_size = get_block_size_from_digest_object(digestBase_obj);
8454 if (block_size == nullptr) return false;
8455 break;
8456 default:
8457 fatal_unexpected_iid(id);
8458 return false;
8459 }
8460 if (state == nullptr) return false;
8461
8462 assert(stubAddr != nullptr, "Stub %s is not generated", stubName);
8463 if (stubAddr == nullptr) return false;
8464
8465 // Call the stub.
8466 Node* call;
8467 if (block_size == nullptr) {
8468 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(false),
8469 stubAddr, stubName, TypePtr::BOTTOM,
8470 src_start, state);
8471 } else {
8472 call = make_runtime_call(RC_LEAF|RC_NO_FP, OptoRuntime::digestBase_implCompress_Type(true),
8473 stubAddr, stubName, TypePtr::BOTTOM,
8474 src_start, state, block_size);
8475 }
8476
8477 return true;
8478 }
8479
8480 //------------------------------inline_double_keccak
8481 bool LibraryCallKit::inline_double_keccak() {
8482 address stubAddr;
8483 const char *stubName;
8484 assert(UseSHA3Intrinsics, "need SHA3 intrinsics support");
8485 assert(callee()->signature()->size() == 2, "double_keccak has 2 parameters");
8486
8487 stubAddr = StubRoutines::double_keccak();
8488 stubName = "double_keccak";
8489 if (!stubAddr) return false;
8490
8491 Node* status0 = argument(0);
8492 Node* status1 = argument(1);
8493
8494 status0 = must_be_not_null(status0, true);
8495 status1 = must_be_not_null(status1, true);
8496
8497 Node* status0_start = array_element_address(status0, intcon(0), T_LONG);
8498 assert(status0_start, "status0 is null");
8499 Node* status1_start = array_element_address(status1, intcon(0), T_LONG);
8500 assert(status1_start, "status1 is null");
8501 Node* double_keccak = make_runtime_call(RC_LEAF|RC_NO_FP,
8502 OptoRuntime::double_keccak_Type(),
8503 stubAddr, stubName, TypePtr::BOTTOM,
8504 status0_start, status1_start);
8505 // return an int
8506 Node* retvalue = _gvn.transform(new ProjNode(double_keccak, TypeFunc::Parms));
8507 set_result(retvalue);
8508 return true;
8509 }
8510
8511
8512 //------------------------------inline_digestBase_implCompressMB-----------------------
8513 //
8514 // Calculate MD5/SHA/SHA2/SHA5/SHA3 for multi-block byte[] array.
8515 // int com.sun.security.provider.DigestBase.implCompressMultiBlock(byte[] b, int ofs, int limit)
8516 //
8517 bool LibraryCallKit::inline_digestBase_implCompressMB(int predicate) {
8518 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8519 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8520 assert((uint)predicate < 5, "sanity");
8521 assert(callee()->signature()->size() == 3, "digestBase_implCompressMB has 3 parameters");
8522
8523 Node* digestBase_obj = argument(0); // The receiver was checked for null already.
8524 Node* src = argument(1); // byte[] array
8525 Node* ofs = argument(2); // type int
8526 Node* limit = argument(3); // type int
8527
8528 const TypeAryPtr* src_type = src->Value(&_gvn)->isa_aryptr();
8529 if (src_type == nullptr || src_type->elem() == Type::BOTTOM) {
8530 // failed array check
8531 return false;
8532 }
8533 // Figure out the size and type of the elements we will be copying.
8534 BasicType src_elem = src_type->elem()->array_element_basic_type();
8535 if (src_elem != T_BYTE) {
8536 return false;
8537 }
8538 // 'src_start' points to src array + offset
8539 src = must_be_not_null(src, false);
8540 Node* src_start = array_element_address(src, ofs, src_elem);
8541
8542 const char* klass_digestBase_name = nullptr;
8543 const char* stub_name = nullptr;
8544 address stub_addr = nullptr;
8545 BasicType elem_type = T_INT;
8546
8547 switch (predicate) {
8548 case 0:
8549 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_md5_implCompress)) {
8550 klass_digestBase_name = "sun/security/provider/MD5";
8551 stub_name = "md5_implCompressMB";
8552 stub_addr = StubRoutines::md5_implCompressMB();
8553 }
8554 break;
8555 case 1:
8556 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha_implCompress)) {
8557 klass_digestBase_name = "sun/security/provider/SHA";
8558 stub_name = "sha1_implCompressMB";
8559 stub_addr = StubRoutines::sha1_implCompressMB();
8560 }
8561 break;
8562 case 2:
8563 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha2_implCompress)) {
8564 klass_digestBase_name = "sun/security/provider/SHA2";
8565 stub_name = "sha256_implCompressMB";
8566 stub_addr = StubRoutines::sha256_implCompressMB();
8567 }
8568 break;
8569 case 3:
8570 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha5_implCompress)) {
8571 klass_digestBase_name = "sun/security/provider/SHA5";
8572 stub_name = "sha512_implCompressMB";
8573 stub_addr = StubRoutines::sha512_implCompressMB();
8574 elem_type = T_LONG;
8575 }
8576 break;
8577 case 4:
8578 if (vmIntrinsics::is_intrinsic_available(vmIntrinsics::_sha3_implCompress)) {
8579 klass_digestBase_name = "sun/security/provider/SHA3";
8580 stub_name = "sha3_implCompressMB";
8581 stub_addr = StubRoutines::sha3_implCompressMB();
8582 elem_type = T_LONG;
8583 }
8584 break;
8585 default:
8586 fatal("unknown DigestBase intrinsic predicate: %d", predicate);
8587 }
8588 if (klass_digestBase_name != nullptr) {
8589 assert(stub_addr != nullptr, "Stub is generated");
8590 if (stub_addr == nullptr) return false;
8591
8592 // get DigestBase klass to lookup for SHA klass
8593 const TypeInstPtr* tinst = _gvn.type(digestBase_obj)->isa_instptr();
8594 assert(tinst != nullptr, "digestBase_obj is not instance???");
8595 assert(tinst->is_loaded(), "DigestBase is not loaded");
8596
8597 ciKlass* klass_digestBase = tinst->instance_klass()->find_klass(ciSymbol::make(klass_digestBase_name));
8598 assert(klass_digestBase->is_loaded(), "predicate checks that this class is loaded");
8599 ciInstanceKlass* instklass_digestBase = klass_digestBase->as_instance_klass();
8600 return inline_digestBase_implCompressMB(digestBase_obj, instklass_digestBase, elem_type, stub_addr, stub_name, src_start, ofs, limit);
8601 }
8602 return false;
8603 }
8604
8605 //------------------------------inline_digestBase_implCompressMB-----------------------
8606 bool LibraryCallKit::inline_digestBase_implCompressMB(Node* digestBase_obj, ciInstanceKlass* instklass_digestBase,
8607 BasicType elem_type, address stubAddr, const char *stubName,
8608 Node* src_start, Node* ofs, Node* limit) {
8609 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_digestBase);
8610 const TypeOopPtr* xtype = aklass->cast_to_exactness(false)->as_instance_type()->cast_to_ptr_type(TypePtr::NotNull);
8611 Node* digest_obj = new CheckCastPPNode(control(), digestBase_obj, xtype);
8612 digest_obj = _gvn.transform(digest_obj);
8613
8614 Node* state = get_state_from_digest_object(digest_obj, elem_type);
8615 if (state == nullptr) return false;
8616
8617 Node* block_size = nullptr;
8618 if (strcmp("sha3_implCompressMB", stubName) == 0) {
8619 block_size = get_block_size_from_digest_object(digest_obj);
8620 if (block_size == nullptr) return false;
8621 }
8622
8623 // Call the stub.
8624 Node* call;
8625 if (block_size == nullptr) {
8626 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8627 OptoRuntime::digestBase_implCompressMB_Type(false),
8628 stubAddr, stubName, TypePtr::BOTTOM,
8629 src_start, state, ofs, limit);
8630 } else {
8631 call = make_runtime_call(RC_LEAF|RC_NO_FP,
8632 OptoRuntime::digestBase_implCompressMB_Type(true),
8633 stubAddr, stubName, TypePtr::BOTTOM,
8634 src_start, state, block_size, ofs, limit);
8635 }
8636
8637 // return ofs (int)
8638 Node* result = _gvn.transform(new ProjNode(call, TypeFunc::Parms));
8639 set_result(result);
8640
8641 return true;
8642 }
8643
8644 //------------------------------inline_galoisCounterMode_AESCrypt-----------------------
8645 bool LibraryCallKit::inline_galoisCounterMode_AESCrypt() {
8646 assert(UseAES, "need AES instruction support");
8647 address stubAddr = nullptr;
8648 const char *stubName = nullptr;
8649 stubAddr = StubRoutines::galoisCounterMode_AESCrypt();
8650 stubName = "galoisCounterMode_AESCrypt";
8651
8652 if (stubAddr == nullptr) return false;
8653
8654 Node* in = argument(0);
8655 Node* inOfs = argument(1);
8656 Node* len = argument(2);
8657 Node* ct = argument(3);
8658 Node* ctOfs = argument(4);
8659 Node* out = argument(5);
8660 Node* outOfs = argument(6);
8661 Node* gctr_object = argument(7);
8662 Node* ghash_object = argument(8);
8663
8664 // (1) in, ct and out are arrays.
8665 const TypeAryPtr* in_type = in->Value(&_gvn)->isa_aryptr();
8666 const TypeAryPtr* ct_type = ct->Value(&_gvn)->isa_aryptr();
8667 const TypeAryPtr* out_type = out->Value(&_gvn)->isa_aryptr();
8668 assert( in_type != nullptr && in_type->elem() != Type::BOTTOM &&
8669 ct_type != nullptr && ct_type->elem() != Type::BOTTOM &&
8670 out_type != nullptr && out_type->elem() != Type::BOTTOM, "args are strange");
8671
8672 // checks are the responsibility of the caller
8673 Node* in_start = in;
8674 Node* ct_start = ct;
8675 Node* out_start = out;
8676 if (inOfs != nullptr || ctOfs != nullptr || outOfs != nullptr) {
8677 assert(inOfs != nullptr && ctOfs != nullptr && outOfs != nullptr, "");
8678 in_start = array_element_address(in, inOfs, T_BYTE);
8679 ct_start = array_element_address(ct, ctOfs, T_BYTE);
8680 out_start = array_element_address(out, outOfs, T_BYTE);
8681 }
8682
8683 // if we are in this set of code, we "know" the embeddedCipher is an AESCrypt object
8684 // (because of the predicated logic executed earlier).
8685 // so we cast it here safely.
8686 // this requires a newer class file that has this array as littleEndian ints, otherwise we revert to java
8687 Node* embeddedCipherObj = load_field_from_object(gctr_object, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8688 Node* counter = load_field_from_object(gctr_object, "counter", "[B");
8689 Node* subkeyHtbl = load_field_from_object(ghash_object, "subkeyHtbl", "[J");
8690 Node* state = load_field_from_object(ghash_object, "state", "[J");
8691
8692 if (embeddedCipherObj == nullptr || counter == nullptr || subkeyHtbl == nullptr || state == nullptr) {
8693 return false;
8694 }
8695 // cast it to what we know it will be at runtime
8696 const TypeInstPtr* tinst = _gvn.type(gctr_object)->isa_instptr();
8697 assert(tinst != nullptr, "GCTR obj is null");
8698 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8699 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8700 assert(klass_AESCrypt->is_loaded(), "predicate checks that this class is loaded");
8701 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8702 const TypeKlassPtr* aklass = TypeKlassPtr::make(instklass_AESCrypt);
8703 const TypeOopPtr* xtype = aklass->as_instance_type();
8704 Node* aescrypt_object = new CheckCastPPNode(control(), embeddedCipherObj, xtype);
8705 aescrypt_object = _gvn.transform(aescrypt_object);
8706 // we need to get the start of the aescrypt_object's expanded key array
8707 Node* k_start = get_key_start_from_aescrypt_object(aescrypt_object);
8708 if (k_start == nullptr) return false;
8709 // similarly, get the start address of the r vector
8710 Node* cnt_start = array_element_address(counter, intcon(0), T_BYTE);
8711 Node* state_start = array_element_address(state, intcon(0), T_LONG);
8712 Node* subkeyHtbl_start = array_element_address(subkeyHtbl, intcon(0), T_LONG);
8713
8714
8715 // Call the stub, passing params
8716 Node* gcmCrypt = make_runtime_call(RC_LEAF|RC_NO_FP,
8717 OptoRuntime::galoisCounterMode_aescrypt_Type(),
8718 stubAddr, stubName, TypePtr::BOTTOM,
8719 in_start, len, ct_start, out_start, k_start, state_start, subkeyHtbl_start, cnt_start);
8720
8721 // return cipher length (int)
8722 Node* retvalue = _gvn.transform(new ProjNode(gcmCrypt, TypeFunc::Parms));
8723 set_result(retvalue);
8724
8725 return true;
8726 }
8727
8728 //----------------------------inline_galoisCounterMode_AESCrypt_predicate----------------------------
8729 // Return node representing slow path of predicate check.
8730 // the pseudo code we want to emulate with this predicate is:
8731 // for encryption:
8732 // if (embeddedCipherObj instanceof AESCrypt) do_intrinsic, else do_javapath
8733 // for decryption:
8734 // if ((embeddedCipherObj instanceof AESCrypt) && (cipher!=plain)) do_intrinsic, else do_javapath
8735 // note cipher==plain is more conservative than the original java code but that's OK
8736 //
8737
8738 Node* LibraryCallKit::inline_galoisCounterMode_AESCrypt_predicate() {
8739 // The receiver was checked for null already.
8740 Node* objGCTR = argument(7);
8741 // Load embeddedCipher field of GCTR object.
8742 Node* embeddedCipherObj = load_field_from_object(objGCTR, "embeddedCipher", "Lcom/sun/crypto/provider/SymmetricCipher;");
8743 assert(embeddedCipherObj != nullptr, "embeddedCipherObj is null");
8744
8745 // get AESCrypt klass for instanceOf check
8746 // AESCrypt might not be loaded yet if some other SymmetricCipher got us to this compile point
8747 // will have same classloader as CipherBlockChaining object
8748 const TypeInstPtr* tinst = _gvn.type(objGCTR)->isa_instptr();
8749 assert(tinst != nullptr, "GCTR obj is null");
8750 assert(tinst->is_loaded(), "GCTR obj is not loaded");
8751
8752 // we want to do an instanceof comparison against the AESCrypt class
8753 ciKlass* klass_AESCrypt = tinst->instance_klass()->find_klass(ciSymbol::make("com/sun/crypto/provider/AESCrypt"));
8754 if (!klass_AESCrypt->is_loaded()) {
8755 // if AESCrypt is not even loaded, we never take the intrinsic fast path
8756 Node* ctrl = control();
8757 set_control(top()); // no regular fast path
8758 return ctrl;
8759 }
8760
8761 ciInstanceKlass* instklass_AESCrypt = klass_AESCrypt->as_instance_klass();
8762 Node* instof = gen_instanceof(embeddedCipherObj, makecon(TypeKlassPtr::make(instklass_AESCrypt)));
8763 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8764 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8765 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8766
8767 return instof_false; // even if it is null
8768 }
8769
8770 //------------------------------get_state_from_digest_object-----------------------
8771 Node * LibraryCallKit::get_state_from_digest_object(Node *digest_object, BasicType elem_type) {
8772 const char* state_type;
8773 switch (elem_type) {
8774 case T_BYTE: state_type = "[B"; break;
8775 case T_INT: state_type = "[I"; break;
8776 case T_LONG: state_type = "[J"; break;
8777 default: ShouldNotReachHere();
8778 }
8779 Node* digest_state = load_field_from_object(digest_object, "state", state_type);
8780 assert (digest_state != nullptr, "wrong version of sun.security.provider.MD5/SHA/SHA2/SHA5/SHA3");
8781 if (digest_state == nullptr) return (Node *) nullptr;
8782
8783 // now have the array, need to get the start address of the state array
8784 Node* state = array_element_address(digest_state, intcon(0), elem_type);
8785 return state;
8786 }
8787
8788 //------------------------------get_block_size_from_sha3_object----------------------------------
8789 Node * LibraryCallKit::get_block_size_from_digest_object(Node *digest_object) {
8790 Node* block_size = load_field_from_object(digest_object, "blockSize", "I");
8791 assert (block_size != nullptr, "sanity");
8792 return block_size;
8793 }
8794
8795 //----------------------------inline_digestBase_implCompressMB_predicate----------------------------
8796 // Return node representing slow path of predicate check.
8797 // the pseudo code we want to emulate with this predicate is:
8798 // if (digestBaseObj instanceof MD5/SHA/SHA2/SHA5/SHA3) do_intrinsic, else do_javapath
8799 //
8800 Node* LibraryCallKit::inline_digestBase_implCompressMB_predicate(int predicate) {
8801 assert(UseMD5Intrinsics || UseSHA1Intrinsics || UseSHA256Intrinsics || UseSHA512Intrinsics || UseSHA3Intrinsics,
8802 "need MD5/SHA1/SHA256/SHA512/SHA3 instruction support");
8803 assert((uint)predicate < 5, "sanity");
8804
8805 // The receiver was checked for null already.
8806 Node* digestBaseObj = argument(0);
8807
8808 // get DigestBase klass for instanceOf check
8809 const TypeInstPtr* tinst = _gvn.type(digestBaseObj)->isa_instptr();
8810 assert(tinst != nullptr, "digestBaseObj is null");
8811 assert(tinst->is_loaded(), "DigestBase is not loaded");
8812
8813 const char* klass_name = nullptr;
8814 switch (predicate) {
8815 case 0:
8816 if (UseMD5Intrinsics) {
8817 // we want to do an instanceof comparison against the MD5 class
8818 klass_name = "sun/security/provider/MD5";
8819 }
8820 break;
8821 case 1:
8822 if (UseSHA1Intrinsics) {
8823 // we want to do an instanceof comparison against the SHA class
8824 klass_name = "sun/security/provider/SHA";
8825 }
8826 break;
8827 case 2:
8828 if (UseSHA256Intrinsics) {
8829 // we want to do an instanceof comparison against the SHA2 class
8830 klass_name = "sun/security/provider/SHA2";
8831 }
8832 break;
8833 case 3:
8834 if (UseSHA512Intrinsics) {
8835 // we want to do an instanceof comparison against the SHA5 class
8836 klass_name = "sun/security/provider/SHA5";
8837 }
8838 break;
8839 case 4:
8840 if (UseSHA3Intrinsics) {
8841 // we want to do an instanceof comparison against the SHA3 class
8842 klass_name = "sun/security/provider/SHA3";
8843 }
8844 break;
8845 default:
8846 fatal("unknown SHA intrinsic predicate: %d", predicate);
8847 }
8848
8849 ciKlass* klass = nullptr;
8850 if (klass_name != nullptr) {
8851 klass = tinst->instance_klass()->find_klass(ciSymbol::make(klass_name));
8852 }
8853 if ((klass == nullptr) || !klass->is_loaded()) {
8854 // if none of MD5/SHA/SHA2/SHA5 is loaded, we never take the intrinsic fast path
8855 Node* ctrl = control();
8856 set_control(top()); // no intrinsic path
8857 return ctrl;
8858 }
8859 ciInstanceKlass* instklass = klass->as_instance_klass();
8860
8861 Node* instof = gen_instanceof(digestBaseObj, makecon(TypeKlassPtr::make(instklass)));
8862 Node* cmp_instof = _gvn.transform(new CmpINode(instof, intcon(1)));
8863 Node* bool_instof = _gvn.transform(new BoolNode(cmp_instof, BoolTest::ne));
8864 Node* instof_false = generate_guard(bool_instof, nullptr, PROB_MIN);
8865
8866 return instof_false; // even if it is null
8867 }
8868
8869 //-------------inline_fma-----------------------------------
8870 bool LibraryCallKit::inline_fma(vmIntrinsics::ID id) {
8871 Node *a = nullptr;
8872 Node *b = nullptr;
8873 Node *c = nullptr;
8874 Node* result = nullptr;
8875 switch (id) {
8876 case vmIntrinsics::_fmaD:
8877 assert(callee()->signature()->size() == 6, "fma has 3 parameters of size 2 each.");
8878 // no receiver since it is static method
8879 a = argument(0);
8880 b = argument(2);
8881 c = argument(4);
8882 result = _gvn.transform(new FmaDNode(a, b, c));
8883 break;
8884 case vmIntrinsics::_fmaF:
8885 assert(callee()->signature()->size() == 3, "fma has 3 parameters of size 1 each.");
8886 a = argument(0);
8887 b = argument(1);
8888 c = argument(2);
8889 result = _gvn.transform(new FmaFNode(a, b, c));
8890 break;
8891 default:
8892 fatal_unexpected_iid(id); break;
8893 }
8894 set_result(result);
8895 return true;
8896 }
8897
8898 bool LibraryCallKit::inline_character_compare(vmIntrinsics::ID id) {
8899 // argument(0) is receiver
8900 Node* codePoint = argument(1);
8901 Node* n = nullptr;
8902
8903 switch (id) {
8904 case vmIntrinsics::_isDigit :
8905 n = new DigitNode(control(), codePoint);
8906 break;
8907 case vmIntrinsics::_isLowerCase :
8908 n = new LowerCaseNode(control(), codePoint);
8909 break;
8910 case vmIntrinsics::_isUpperCase :
8911 n = new UpperCaseNode(control(), codePoint);
8912 break;
8913 case vmIntrinsics::_isWhitespace :
8914 n = new WhitespaceNode(control(), codePoint);
8915 break;
8916 default:
8917 fatal_unexpected_iid(id);
8918 }
8919
8920 set_result(_gvn.transform(n));
8921 return true;
8922 }
8923
8924 bool LibraryCallKit::inline_profileBoolean() {
8925 Node* counts = argument(1);
8926 const TypeAryPtr* ary = nullptr;
8927 ciArray* aobj = nullptr;
8928 if (counts->is_Con()
8929 && (ary = counts->bottom_type()->isa_aryptr()) != nullptr
8930 && (aobj = ary->const_oop()->as_array()) != nullptr
8931 && (aobj->length() == 2)) {
8932 // Profile is int[2] where [0] and [1] correspond to false and true value occurrences respectively.
8933 jint false_cnt = aobj->element_value(0).as_int();
8934 jint true_cnt = aobj->element_value(1).as_int();
8935
8936 if (C->log() != nullptr) {
8937 C->log()->elem("observe source='profileBoolean' false='%d' true='%d'",
8938 false_cnt, true_cnt);
8939 }
8940
8941 if (false_cnt + true_cnt == 0) {
8942 // According to profile, never executed.
8943 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8944 Deoptimization::Action_reinterpret);
8945 return true;
8946 }
8947
8948 // result is a boolean (0 or 1) and its profile (false_cnt & true_cnt)
8949 // is a number of each value occurrences.
8950 Node* result = argument(0);
8951 if (false_cnt == 0 || true_cnt == 0) {
8952 // According to profile, one value has been never seen.
8953 int expected_val = (false_cnt == 0) ? 1 : 0;
8954
8955 Node* cmp = _gvn.transform(new CmpINode(result, intcon(expected_val)));
8956 Node* test = _gvn.transform(new BoolNode(cmp, BoolTest::eq));
8957
8958 IfNode* check = create_and_map_if(control(), test, PROB_ALWAYS, COUNT_UNKNOWN);
8959 Node* fast_path = _gvn.transform(new IfTrueNode(check));
8960 Node* slow_path = _gvn.transform(new IfFalseNode(check));
8961
8962 { // Slow path: uncommon trap for never seen value and then reexecute
8963 // MethodHandleImpl::profileBoolean() to bump the count, so JIT knows
8964 // the value has been seen at least once.
8965 PreserveJVMState pjvms(this);
8966 PreserveReexecuteState preexecs(this);
8967 jvms()->set_should_reexecute(true);
8968
8969 set_control(slow_path);
8970 set_i_o(i_o());
8971
8972 uncommon_trap_exact(Deoptimization::Reason_intrinsic,
8973 Deoptimization::Action_reinterpret);
8974 }
8975 // The guard for never seen value enables sharpening of the result and
8976 // returning a constant. It allows to eliminate branches on the same value
8977 // later on.
8978 set_control(fast_path);
8979 result = intcon(expected_val);
8980 }
8981 // Stop profiling.
8982 // MethodHandleImpl::profileBoolean() has profiling logic in its bytecode.
8983 // By replacing method body with profile data (represented as ProfileBooleanNode
8984 // on IR level) we effectively disable profiling.
8985 // It enables full speed execution once optimized code is generated.
8986 Node* profile = _gvn.transform(new ProfileBooleanNode(result, false_cnt, true_cnt));
8987 C->record_for_igvn(profile);
8988 set_result(profile);
8989 return true;
8990 } else {
8991 // Continue profiling.
8992 // Profile data isn't available at the moment. So, execute method's bytecode version.
8993 // Usually, when GWT LambdaForms are profiled it means that a stand-alone nmethod
8994 // is compiled and counters aren't available since corresponding MethodHandle
8995 // isn't a compile-time constant.
8996 return false;
8997 }
8998 }
8999
9000 bool LibraryCallKit::inline_isCompileConstant() {
9001 Node* n = argument(0);
9002 set_result(n->is_Con() ? intcon(1) : intcon(0));
9003 return true;
9004 }
9005
9006 //------------------------------- inline_getObjectSize --------------------------------------
9007 //
9008 // Calculate the runtime size of the object/array.
9009 // native long sun.instrument.InstrumentationImpl.getObjectSize0(long nativeAgent, Object objectToSize);
9010 //
9011 bool LibraryCallKit::inline_getObjectSize() {
9012 Node* obj = argument(3);
9013 Node* klass_node = load_object_klass(obj);
9014
9015 jint layout_con = Klass::_lh_neutral_value;
9016 Node* layout_val = get_layout_helper(klass_node, layout_con);
9017 int layout_is_con = (layout_val == nullptr);
9018
9019 if (layout_is_con) {
9020 // Layout helper is constant, can figure out things at compile time.
9021
9022 if (Klass::layout_helper_is_instance(layout_con)) {
9023 // Instance case: layout_con contains the size itself.
9024 Node *size = longcon(Klass::layout_helper_size_in_bytes(layout_con));
9025 set_result(size);
9026 } else {
9027 // Array case: size is round(header + element_size*arraylength).
9028 // Since arraylength is different for every array instance, we have to
9029 // compute the whole thing at runtime.
9030
9031 Node* arr_length = load_array_length(obj);
9032
9033 int round_mask = MinObjAlignmentInBytes - 1;
9034 int hsize = Klass::layout_helper_header_size(layout_con);
9035 int eshift = Klass::layout_helper_log2_element_size(layout_con);
9036
9037 if ((round_mask & ~right_n_bits(eshift)) == 0) {
9038 round_mask = 0; // strength-reduce it if it goes away completely
9039 }
9040 assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
9041 Node* header_size = intcon(hsize + round_mask);
9042
9043 Node* lengthx = ConvI2X(arr_length);
9044 Node* headerx = ConvI2X(header_size);
9045
9046 Node* abody = lengthx;
9047 if (eshift != 0) {
9048 abody = _gvn.transform(new LShiftXNode(lengthx, intcon(eshift)));
9049 }
9050 Node* size = _gvn.transform( new AddXNode(headerx, abody) );
9051 if (round_mask != 0) {
9052 size = _gvn.transform( new AndXNode(size, MakeConX(~round_mask)) );
9053 }
9054 size = ConvX2L(size);
9055 set_result(size);
9056 }
9057 } else {
9058 // Layout helper is not constant, need to test for array-ness at runtime.
9059
9060 enum { _instance_path = 1, _array_path, PATH_LIMIT };
9061 RegionNode* result_reg = new RegionNode(PATH_LIMIT);
9062 PhiNode* result_val = new PhiNode(result_reg, TypeLong::LONG);
9063 record_for_igvn(result_reg);
9064
9065 Node* array_ctl = generate_array_guard(klass_node, nullptr, &obj);
9066 if (array_ctl != nullptr) {
9067 // Array case: size is round(header + element_size*arraylength).
9068 // Since arraylength is different for every array instance, we have to
9069 // compute the whole thing at runtime.
9070
9071 PreserveJVMState pjvms(this);
9072 set_control(array_ctl);
9073 Node* arr_length = load_array_length(obj);
9074
9075 int round_mask = MinObjAlignmentInBytes - 1;
9076 Node* mask = intcon(round_mask);
9077
9078 Node* hss = intcon(Klass::_lh_header_size_shift);
9079 Node* hsm = intcon(Klass::_lh_header_size_mask);
9080 Node* header_size = _gvn.transform(new URShiftINode(layout_val, hss));
9081 header_size = _gvn.transform(new AndINode(header_size, hsm));
9082 header_size = _gvn.transform(new AddINode(header_size, mask));
9083
9084 // There is no need to mask or shift this value.
9085 // The semantics of LShiftINode include an implicit mask to 0x1F.
9086 assert(Klass::_lh_log2_element_size_shift == 0, "use shift in place");
9087 Node* elem_shift = layout_val;
9088
9089 Node* lengthx = ConvI2X(arr_length);
9090 Node* headerx = ConvI2X(header_size);
9091
9092 Node* abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
9093 Node* size = _gvn.transform(new AddXNode(headerx, abody));
9094 if (round_mask != 0) {
9095 size = _gvn.transform(new AndXNode(size, MakeConX(~round_mask)));
9096 }
9097 size = ConvX2L(size);
9098
9099 result_reg->init_req(_array_path, control());
9100 result_val->init_req(_array_path, size);
9101 }
9102
9103 if (!stopped()) {
9104 // Instance case: the layout helper gives us instance size almost directly,
9105 // but we need to mask out the _lh_instance_slow_path_bit.
9106 Node* size = ConvI2X(layout_val);
9107 assert((int) Klass::_lh_instance_slow_path_bit < BytesPerLong, "clear bit");
9108 Node* mask = MakeConX(~(intptr_t) right_n_bits(LogBytesPerLong));
9109 size = _gvn.transform(new AndXNode(size, mask));
9110 size = ConvX2L(size);
9111
9112 result_reg->init_req(_instance_path, control());
9113 result_val->init_req(_instance_path, size);
9114 }
9115
9116 set_result(result_reg, result_val);
9117 }
9118
9119 return true;
9120 }
9121
9122 //------------------------------- inline_blackhole --------------------------------------
9123 //
9124 // Make sure all arguments to this node are alive.
9125 // This matches methods that were requested to be blackholed through compile commands.
9126 //
9127 bool LibraryCallKit::inline_blackhole() {
9128 assert(callee()->is_static(), "Should have been checked before: only static methods here");
9129 assert(callee()->is_empty(), "Should have been checked before: only empty methods here");
9130 assert(callee()->holder()->is_loaded(), "Should have been checked before: only methods for loaded classes here");
9131
9132 // Blackhole node pinches only the control, not memory. This allows
9133 // the blackhole to be pinned in the loop that computes blackholed
9134 // values, but have no other side effects, like breaking the optimizations
9135 // across the blackhole.
9136
9137 Node* bh = _gvn.transform(new BlackholeNode(control()));
9138 set_control(_gvn.transform(new ProjNode(bh, TypeFunc::Control)));
9139
9140 // Bind call arguments as blackhole arguments to keep them alive
9141 uint nargs = callee()->arg_size();
9142 for (uint i = 0; i < nargs; i++) {
9143 bh->add_req(argument(i));
9144 }
9145
9146 return true;
9147 }
9148
9149 Node* LibraryCallKit::unbox_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* box) {
9150 const TypeInstPtr* box_type = _gvn.type(box)->isa_instptr();
9151 if (box_type == nullptr || box_type->instance_klass() != float16_box_type->instance_klass()) {
9152 return nullptr; // box klass is not Float16
9153 }
9154
9155 // Null check; get notnull casted pointer
9156 Node* null_ctl = top();
9157 Node* not_null_box = null_check_oop(box, &null_ctl, true);
9158 // If not_null_box is dead, only null-path is taken
9159 if (stopped()) {
9160 set_control(null_ctl);
9161 return nullptr;
9162 }
9163 assert(not_null_box->bottom_type()->is_instptr()->maybe_null() == false, "");
9164 const TypePtr* adr_type = C->alias_type(field)->adr_type();
9165 Node* adr = basic_plus_adr(not_null_box, field->offset_in_bytes());
9166 return access_load_at(not_null_box, adr, adr_type, TypeInt::SHORT, T_SHORT, IN_HEAP);
9167 }
9168
9169 Node* LibraryCallKit::box_fp16_value(const TypeInstPtr* float16_box_type, ciField* field, Node* value) {
9170 PreserveReexecuteState preexecs(this);
9171 jvms()->set_should_reexecute(true);
9172
9173 const TypeKlassPtr* klass_type = float16_box_type->as_klass_type();
9174 Node* klass_node = makecon(klass_type);
9175 Node* box = new_instance(klass_node);
9176
9177 Node* value_field = basic_plus_adr(box, field->offset_in_bytes());
9178 const TypePtr* value_adr_type = value_field->bottom_type()->is_ptr();
9179
9180 Node* field_store = _gvn.transform(access_store_at(box,
9181 value_field,
9182 value_adr_type,
9183 value,
9184 TypeInt::SHORT,
9185 T_SHORT,
9186 IN_HEAP));
9187 set_memory(field_store, value_adr_type);
9188 return box;
9189 }
9190
9191 bool LibraryCallKit::inline_fp16_operations(vmIntrinsics::ID id, int num_args) {
9192 if (!Matcher::match_rule_supported(Op_ReinterpretS2HF) ||
9193 !Matcher::match_rule_supported(Op_ReinterpretHF2S)) {
9194 return false;
9195 }
9196
9197 const TypeInstPtr* box_type = _gvn.type(argument(0))->isa_instptr();
9198 if (box_type == nullptr || box_type->const_oop() == nullptr) {
9199 return false;
9200 }
9201
9202 ciInstanceKlass* float16_klass = box_type->const_oop()->as_instance()->java_lang_Class_klass()->as_instance_klass();
9203 const TypeInstPtr* float16_box_type = TypeInstPtr::make_exact(TypePtr::NotNull, float16_klass);
9204 ciField* field = float16_klass->get_field_by_name(ciSymbols::value_name(),
9205 ciSymbols::short_signature(),
9206 false);
9207 assert(field != nullptr, "");
9208
9209 // Transformed nodes
9210 Node* fld1 = nullptr;
9211 Node* fld2 = nullptr;
9212 Node* fld3 = nullptr;
9213 switch(num_args) {
9214 case 3:
9215 fld3 = unbox_fp16_value(float16_box_type, field, argument(3));
9216 if (fld3 == nullptr) {
9217 return false;
9218 }
9219 fld3 = _gvn.transform(new ReinterpretS2HFNode(fld3));
9220 // fall-through
9221 case 2:
9222 fld2 = unbox_fp16_value(float16_box_type, field, argument(2));
9223 if (fld2 == nullptr) {
9224 return false;
9225 }
9226 fld2 = _gvn.transform(new ReinterpretS2HFNode(fld2));
9227 // fall-through
9228 case 1:
9229 fld1 = unbox_fp16_value(float16_box_type, field, argument(1));
9230 if (fld1 == nullptr) {
9231 return false;
9232 }
9233 fld1 = _gvn.transform(new ReinterpretS2HFNode(fld1));
9234 break;
9235 default: fatal("Unsupported number of arguments %d", num_args);
9236 }
9237
9238 Node* result = nullptr;
9239 switch (id) {
9240 // Unary operations
9241 case vmIntrinsics::_sqrt_float16:
9242 result = _gvn.transform(new SqrtHFNode(C, control(), fld1));
9243 break;
9244 // Ternary operations
9245 case vmIntrinsics::_fma_float16:
9246 result = _gvn.transform(new FmaHFNode(fld1, fld2, fld3));
9247 break;
9248 default:
9249 fatal_unexpected_iid(id);
9250 break;
9251 }
9252 result = _gvn.transform(new ReinterpretHF2SNode(result));
9253 set_result(box_fp16_value(float16_box_type, field, result));
9254 return true;
9255 }
9256